JP4351684B2 - Digital signal decoding method, apparatus, program, and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a coding method of compressing information of a digital signal within controlled distortion, without having not so large a deterioration in the quality of decoded signal, even if bits are missing on a transmission line and increasing the compression efficiency, while guaranteeing precision at bit rates over a wide range, a decoding method therefor, and apparatuses for them. <P>SOLUTION: Input codes are decoded to generate a plurality of sets of reversible data of a bit sequence, where samples of a tentative digital signal are arranged across their bit positions. A signal, generated by inserting values included in data of the plurality of sets where values are obtained from the data of the plurality of sets and inserting an arbitrary value where values are not obtained from the data of the plurality of sets, is reproduced as a tentative digital signal (431). Input auxiliary information is decoded to find a decoded spectral envelope (450). A decoded digital signal is obtained (434) by correcting the tentative digital signal so that a spectral envelope of the tentative digital signal approximates the decoded spectrum envelope. <P>COPYRIGHT: (C)2006,JPO&amp;NCIPI

Description

  The present invention relates to a decoding method, apparatus, program, and recording medium for converting into a code with a reduced number of bits representing various digital signals such as audio signals, music signals, and image signals.

  There are irreversible encoding that allows distortion and reversible encoding that does not allow distortion as methods of compressing information such as sound and image. As irreversible compression, various methods are known as standards of ITU-T (International Telecommunications Union-Telecom Standardization) and ISO / IEC MPEG (International Organization for Standardization / International Electrotechnical Commission Moving Picture Experts Group). By using these lossy compression methods, it is possible to compress the original digital signal to 1/10 or less while suppressing a slight distortion. However, the distortion depends on the encoding condition and the input signal, and deterioration of the reproduction signal may be a problem depending on the application.

On the other hand, as a reversible compression method capable of completely reproducing the original signal, universal compression coding often used for compression of computer files and texts is known. It can compress any signal while learning the statistics of the input sequence, and can compress up to about 1/2 for text etc., but even if it is applied directly to audio or image data, the compression effect Only around 20%.
Performs irreversible encoding with a high compression ratio and reversibly compresses the error between the reproduced signal and the original signal to flexibly use both lossy encoding with high compression ratio and lossless compression. Is possible.

The present inventor proposed this combination compression method in Patent Document 1. This method is described in detail in the above publication, but will be briefly described below with reference to FIG.
In the encoder, a digital input signal (hereinafter also referred to as an input signal sample sequence) is input from the input terminal 100, and the frame division unit 110 converts the input signal sample sequence into frame units including, for example, 1024 input signal samples. Are sequentially divided.

  The lossy quantization unit 120 performs lossy compression coding on the output of the frame division unit 110. This encoding may be any system suitable for the input signal as long as the original digital input signal can be reproduced to some extent at the time of decoding. For example, if the input signal is an audio signal, ITU-T audio coding can be used, if it is music, MPEG or TwinVQ (Transform-Domain Weighted Interleaved Vector Quantization) can be used, and if it is video, MPEG etc. Is available. In addition, various irreversible quantization methods described in the above publication can also be used. The output from the irreversible quantization unit 120 is referred to as “irreversible compression code I (n)”.

A local reproduction signal is generated from the irreversible compression code I (n) by the inverse quantization unit 130 having the same configuration as the decoding unit (that is, the inverse quantization unit 230) corresponding to the irreversible quantization unit 120. An error signal between the local reproduction signal and the original digital input signal is obtained by the subtraction unit 140. Usually, the amplitude of this error signal is much smaller than the amplitude of the original digital input signal. Therefore, the amount of information can be reduced by reversibly compressing and encoding the error signal rather than performing lossless compression encoding of the digital input signal as it is.
In order to increase the efficiency of the lossless compression encoding, the rearrangement unit 160 rearranges the bits of the error signal (that is, the bit string). Details of the processing by the rearrangement unit 160 will be described below with reference to FIG. In the digital input signal (FIG. 2A), positive and negative integers of each sample value (amplitude) are expressed by 2's complement expression. An error signal sample between the digital input signal and the local reproduction signal corresponding to the digital input signal is shown in FIG. 2B. The rearrangement unit 160 converts the error signal (that is, the bit string) from a bit string expressed in 2's complement to a bit string expressed in code absolute value (binary number of polarity and absolute value) (FIG. 2C). In the converted error signal, the MSB (Most Significant Bit) to the second LSB (Least Significant Bit) indicate the absolute value of the amplitude, and the LSB indicates the polarity code of the amplitude.

  Next, the rearrangement unit 160 concatenates the error signal samples converted into the code absolute value representation for each corresponding bit position (MSB, second MSB,..., LSB), that is, in the time axis direction of FIG. (FIG. 2D). Each bit string (that is, a bit string composed of 1024 bits at the same bit position) is hereinafter referred to as a “coordinate bit string”. With the above rearrangement, the value of each error signal does not change at all. However, since the error signal has a small amplitude, in the output of the rearrangement unit 160, the upper bits are often all “0”. As a result, the continuous loss of “0” increases the lossless encoding efficiency of the error signal.

Next, the lossless encoding unit 150 performs lossless compression encoding on the output of the rearrangement unit 160. The lossless encoding unit 150 performs lossless compression encoding of the equivalent bit string by entropy encoding such as Huffman encoding or arithmetic encoding using, for example, a case where there are continuous sequences or a sequence that frequently appears. To the decoder 20. The compression efficiency is also improved by applying universal coding for reversibly compressing text or the like to the output of the rearrangement unit 160.
Through the above processing, the encoder outputs the lossy compression code I (n) from the lossy quantization unit 120 and the lossless compression code I (e) from the lossless encoding unit 150.

In the decoder 20, the lossless decoding unit 210 decodes the lossless compression code I (e). Then, the rearrangement unit 220 performs reverse processing to the rearrangement unit 160 to sequentially output error signals in units of frames. In addition, the inverse quantization unit 230 decodes the lossy compression code I (n). Adder 240 adds the output of inverse quantization unit 230 and the output of rearrangement unit 160. Finally, the frame synthesizing unit 250 reproduces the original input signal sample sequence by sequentially connecting the outputs of the adding unit 240. Through the above processing, the input signal sample sequence is output from the output terminal 260.
JP 2001-44847 A

This conventional coding method has a problem that if a bit loss occurs on the transmission path, bits of other samples are mixed in each sample reconstructed by the rearrangement unit of the decoder 20, and the quality of the decoded signal is greatly deteriorated. . Also, the compression ratio that can be realized is about 1/2, and compression with 1/3 or 1/4 or guaranteed accuracy can not be realized with almost no loss in quality.
In addition, waveform reproduction that guarantees the same accuracy as when the number of bits of the digital value representing the amplitude of the original signal is reduced by 1 bit is possible. There was a problem that I could hear.

  A first object of the present invention is to provide an encoding method, a decoding method, and an apparatus thereof that do not cause a great quality degradation in a decoded signal even if a bit loss occurs in a transmission line. A second object of the present invention is to compress a digital signal within a controlled distortion and to encode the signal to ensure high accuracy with a wide range of bit rates, a decoding method thereof, It is to provide these devices.

According to the present invention, an encoding method for encoding a digital signal for each frame including a plurality of samples includes the following steps.
(a) Generate a plurality of pairs of reversible data of bit strings straddling digital signal samples for each frame at each bit position, or reversible data sets of irreversible data and error signals derived from the irreversible data. ,
(b) The above data set is output as a code.

According to the present invention, there is provided a decoding method for reproducing a sample of a digital signal for each frame,
(a) Multiple sets of reversible data of bit strings obtained by decoding input codes and straddling digital signal samples at their respective bit positions, or reversible data sets of error signals derived from irreversible data and the irreversible data A step of generating
(b) regenerating a digital signal based on the plurality of sets of data;
Including.

According to the present invention, an encoder for encoding a digital signal for each frame composed of a plurality of samples,
Means for generating a plurality of sets of reversible data of bit strings straddling the digital signal samples for each frame at each bit position, or a set of reversible data of error signals derived from the irreversible data and the irreversible data for each frame When,
An output means for outputting the code set of the data;
Including.

According to the present invention, there is provided a decoder for reproducing a sample of a digital signal for each frame composed of a plurality of samples,
Decodes the input code and generates a reversible set of reversible data of the bit string straddling digital signal samples at each bit position, or a reversible set of error signals derived from the irreversible data and the irreversible data. Data generation means for
Means for reproducing a digital signal from the data set;
Including.

  According to the above encoding method, encoder and decoding method, and decoder according to the present invention, packetization is performed for each transmission record unit data. Therefore, in the encoding process, packets are intentionally dropped according to transmission capacity, storage capacity, etc. Even if the amount of information is reduced, according to the decoding method of the present invention, transmission record unit data can be corrected for packet loss.

According to the lossless encoding method of the present invention, packetization is performed for each transmission record unit data, so that the amount of information is reduced by intentionally dropping packets according to transmission capacity, storage capacity, etc., that is, 1/3 or 1 / 4, according to the lossless decoding method of the present invention, transmission record unit data can be corrected for packet loss.
In this specification, packets in one frame are intentionally removed in order to adjust the amount of information, so that all the packets in one frame are not input to the decoder, or an exchange station due to traffic congestion in the communication network. In case of packet loss based on not sending out some packets or due to transmission path failure, recording / reproducing device abnormality, etc., there is an error in the input packet, and the transmission record unit data cannot be decoded, A case where it cannot be used is generically referred to as packet loss.

Embodiments of the present invention will be described below with reference to the drawings. In each figure, the same reference numerals are given to the parts corresponding to those already explained.
First Embodiment FIG. 3 shows embodiments of the encoder 10 and the decoder 20 according to the present invention. This embodiment is different from the prior art of FIG. 1 in that a header adding unit 320 is provided, and a signal is transmitted in a packet, thereby reducing the influence of quality degradation on a decoded signal due to missing bits in the transmission path. It is that.

  In the encoder 10, the input signal sample sequence from the input terminal is sequentially divided by the frame dividing unit 110 into frame units including, for example, 1024 input signal samples (that is, n = 1024 point samples). For each frame, the input signal sample sequence is reordered by the rearrangement unit 160 so that bits at corresponding bit positions in the amplitude bit string of each sample are connected in the time direction over the frame. At this time, the input signal sample is expressed in two's complement with a positive or negative integer amplitude, and each two-complement expressed sample is converted into a bit string by this two's complement expression as in the description of the prior art. The polarity absolute value conversion unit 161 converts it into an m-bit binary representation composed of a polarity code and an absolute value, and then, for each m-bit sample of 1 frame, the bit array conversion unit 162 converts each m-bit sample. It is preferable to rearrange the bits at the corresponding bit positions into a bit array (that is, a coordinate bit string) concatenated in the time direction over the frame. However, as indicated by the broken line in the figure, the bit array conversion unit 162 converts the bit positions at the corresponding bit positions in the time direction without changing the polarity absolute value conversion unit 161, that is, with the 2's complement representation. It may be rearranged into a coordinate bit string.

  In the present invention, the equidistant bit string from the rearrangement unit 160 is divided into transmission unit or recording unit data by the transmission / recording unit dividing unit 310. This unit is hereinafter referred to as a block, but is also referred to as a packet generally used as a transmission unit. These divided transmission record unit data are losslessly encoded by the lossless encoding unit 150 as necessary, and the header addition unit 320 re-transmits the transmission record unit data separated from each other as one frame at the time of decoding. For example, a header including a serial number is attached so as to be configurable, and is output to the output terminal 170 as a packet. The lossless encoding unit 150 is the same as the lossless encoding unit 150 of the prior art.

  A normal input signal sample expresses a positive or negative integer in two's complement. In this example, the bit sequence is rearranged in a numerical sequence for each frame after the digital input signal is converted into a binary number having a polarity code and an absolute value of amplitude. That is, FIG. 4A shows an example of conversion to a coordinate bit string and division into transmission recording unit data. Each m-bit sample (amplitude bit string) expressed as a polarity code and an absolute value is arranged in the time direction as shown in the left side of FIG. 4A. In order to easily understand the state of the amplitude bit string of the nth sample, it is shown as an amplitude bit string DV (n) surrounded by a thick line. n represents a sample number in the time axis direction in the frame, for example, n = 1, 2,. As will be described later, for example, if the input data is an audio signal, if bit loss occurs due to packet loss in the transmission path, higher-order bits of each amplitude bit string DV (n) are given distortion to the reproduced audio quality, Since the polarity code gives the largest distortion, in this example, the polarity code of each amplitude bit string DV (n) is adjacent to the MSB of the absolute value, and in the figure, the polarity code is positioned immediately above the MSB.

  In this example, as shown on the right side of FIG. 4A, such a sample string is first connected in the time axis direction with only the polarity bit (sign) of the amplitude value of each sample to create a coordinate bit string within the frame. Next, only MSBs with m-bit amplitude values are connected in a frame to create a coordinate bit string, and in the same manner, bits are sequentially connected in the time direction over the frame for each corresponding bit position to create a coordinate bit string. Finally, only the LSB is concatenated in the frame to create a coordinate bit string. One example of these coordinate bit strings is shown as DH (i) with a thick frame in the right bit arrangement in FIG. 4A. i indicates each bit position in the amplitude bit string DV (n), for example, i = 1, 2,..., 17 from the MSB side. With the above rearrangement, the content of the data contained in the frame does not change at all.

The transmission record unit dividing unit 310 divides the data into transmission record unit data for each coordinate bit string DH (i) or for each of a plurality of adjacent coordinate bit strings DH (i). In this case, in one frame, data composed of one coordinate bit string and data composed of a plurality of coordinate bit strings may be mixed as transmission recording unit data.
Each transmission record unit data divided in this way is added by a header adding unit 320. For example, a header 31 is attached to the transmission record unit data (payload) 32 as shown in FIG. 4B. The header 31 is provided with, for example, a packet number 33 including a frame number and a transmission record unit data number in the frame, and a priority 34 and a data length 35 of the packet as necessary. The digital signal train can be reconstructed for each frame.

The data length 35 is not necessary if the data length of the transmission record unit data 32 is fixed, but the data length may differ depending on the packet when the lossless encoding unit 150 performs compression encoding.
More generally, an error detection code 36 such as a CRC code for detecting whether or not an error has occurred in the entire packet is added at the end to form one packet 30, and this packet 30 is output from the output terminal 170. The
When priority is given to the packet 30, a code indicating a higher priority is given to a packet including transmission recording unit data corresponding to a higher-order bit position of the amplitude bit string. That is, when dividing each transmission bit unit DH (i) into transmission recording unit data, a packet including data having a smaller i of DH (i) is given higher priority. Therefore, in the example of FIG. 4A, the highest priority is given to DH (1) including the polarity equivalence bit string corresponding to i = 1. The priorities do not necessarily have to be given different monotonically increasing priorities for the LSB to MSB of the amplitude bit string. For example, different priorities are given to the upper bits, and the same priority is given to the lower bits. May be. That is, the predetermined number of priorities may be equal to or smaller than the number of bits of the amplitude bit string. As will be described later, when high-reciprocal lossy quantization is used together as described in the related art, the highest priority may be given to a bit string representing the high-compression encoded code.

  In the decoder 20, the packet 30 from the input terminal 200 is losslessly decoded by a lossless decoding unit 210 (same as the lossless decoding unit 210 of the prior art) when the transmission record unit data is losslessly compressed. The recording unit data is integrated by one frame from a plurality of packets by the transmission recording unit integration unit 410 based on the packet number, and is integrated, for example, as shown in the right-order bit string DH (i) in FIG. 4A. The integrated equidistant bit string is converted into an amplitude bit string DV (n), that is, a sample string (waveform sample string) by the rearrangement unit 220. At this time, if the transmission recording unit data is created from what each sample represents with a polarity code and an absolute value, the bit array inverse conversion unit 221 converts the equidistant bit string in reverse to the rearrangement described in FIG. 4A. Are rearranged into amplitude bit strings, and each two-bit complement conversion unit 222 converts each amplitude bit string into a two's complement representation. When the transmission record unit data is based on the one directly replaced from the two-complement expressed amplitude bit string to the equidistant bit string, the bit array inverse conversion unit 221 starts from the bit array inverse conversion unit 221 as indicated by a broken line in FIG. Are directly supplied to the frame synthesis unit 250 as decoded samples.

In the present invention, when the loss detection unit 420 detects that a series of packet numbers of an input packet is missing, it determines that the packet of that packet number is missing, and the amplitude bit string from the rearrangement unit 220 Is not supplied directly to the frame synthesizing unit 250, but is supplied to the missing correction unit 430. Correction of the missing information is performed on the amplitude bit string (sample) and supplied to the frame synthesizing unit 250.
The correction in the missing correction unit 430 is performed by estimating missing information from known information. If some packets, packets containing bits on the LSB side with low priority, are lost, the value of the bit corresponding to the missing portion cannot be determined, so a small numerical value such as 0 or the missing portion can be taken. The waveform must be replayed using the intermediate value. In this case, the accuracy of the determined number of bits on the MSB side can be maintained, but auditory distortion due to missing bits is large. The reason for this is that, as shown in the schematic diagram of FIG. 5, the spectrum of the original sound often has a biased energy in the low range as shown by the solid line, whereas the distortion component due to missing bits is almost flat as shown by the broken line. Since it has a spectral shape, its high-frequency component is larger than the original sound and can be heard as noise. Therefore, the value of the uncertain waveform is corrected so that the spectrum of the uncertain component is close to the average spectrum or the spectrum determined every frame. As a result, the corrected distortion component spectrum has a high-frequency component that is smaller as shown by the one-dot chain line in FIG. 5, and the distortion is masked by the original sound, thereby improving the quality.

  In other words, when the average spectrum of the past several frames or the determined spectrum obtained separately in that frame as described later is close to the solid line in FIG. 5, for example, it is obtained from information other than the missing information of the frame. The missing information is corrected so that the obtained spectrum approaches the solid line in FIG. A preferred technique for this correction will be described in a later embodiment. As a simple method, the missing correction unit 430 may smooth the input reproduction sample sequence through a low-pass filter to remove a high-frequency noise component. If the spectral shape (envelope) of the original signal is known in advance, the cutoff characteristic of the low-pass filter is selected so as to attenuate the high-frequency component according to the characteristic. Alternatively, as described above, an average spectrum may be obtained, or the cutoff characteristic may be adaptively changed in accordance with the shape of the spectrum determined for each frame.

Since the decoder 20 can correct information loss due to packet loss in this way, the LSB side packet is not sent intentionally as necessary, and decoding can be performed by the decoder 20 even if the encoding compression efficiency is increased. Alternatively, it is possible to perform reproduction that does not cause a problem with hearing.
Second Embodiment FIG. 6 shows a second embodiment of the present invention. In the second embodiment, the input signal sample sequence divided for each frame by the frame dividing unit 110 in the encoder 10 is supplied to the auxiliary information generating unit 350. The auxiliary information generation unit 350 includes a spectrum envelope calculation unit 351, a power calculation unit 354, and an auxiliary information encoding unit 352. In the input signal sample sequence divided for each frame by the frame dividing unit 110, a parameter expressing the spectral envelope is obtained by the spectral envelope calculating unit 351, for example, as a linear prediction coefficient LPC by linear prediction analysis, and the frame is calculated by the power calculating unit 354. The average power Pa of the input signal sample sequence for each is obtained.

  Alternatively, the input signal sample sequence is input to the inverse filter 355 configured based on the obtained linear prediction coefficient LPC, and the spectrum envelope is flattened, and the average power of the flattened signal is calculated by the flattening power calculation unit 356. Desired. The linear prediction coefficient LPC and the average power Pa are encoded by the auxiliary information encoding unit 352, for example, with low bits of about 30 to 50 bits and used as auxiliary information. This auxiliary information is supplied to the header adding unit 320 and added to a representative packet of each frame, for example, a packet including transmission recording unit data including a polarity code, or output as an independent packet.

Note that the spectrum envelope coefficient obtained by the spectrum envelope calculation unit 351 does not change so much even if the frame length is increased, and the information amount of the spectrum envelope coefficient encoded by the auxiliary information encoding unit 340 does not change so much. Therefore, as shown by the broken line in the figure, the frame length is longer than the frame length of the digital input signal given to the rearrangement unit 160 by the frame division unit 110 ′ and given to the auxiliary information generation unit 350 to increase the efficiency of lossless compression. You may do it.
In the decoder 20, the packet input to the input terminal 200 is separated into transmission recording unit data and auxiliary information in the separation unit 440, the transmission recording unit data is supplied to the lossless decoding unit 210, and the auxiliary information is auxiliary information. The auxiliary information decoding unit 450 is supplied to the decoding unit 450, and the auxiliary information decoding unit 450 decodes the parameter representing the spectral envelope of the frame, here, the linear prediction coefficient LPC and the average power Pa, and the spectral envelope coefficient LPC and the average power Pa are deleted. To supply. The missing correction unit 430 uses the spectrum envelope coefficient LPC and the average power Pa for correcting missing information, and corrects missing information as described in the first embodiment, for example.

Alternatively, each combination of possible values of missing information (bits) is added to each sample value to create correction sample string (waveform) candidates, and the spectrum envelopes of these candidates are obtained, and the spectrum envelope is the decoded spectrum of the auxiliary information. A correction sample sequence (waveform) candidate corresponding to the closest envelope is output to the frame synthesis unit 250 as a correction sample sequence. In FIG. 6, the lossless encoding unit 150 and the lossless decoding unit 210 may be omitted.
When candidates for correction sample sequences are created using combinations of possible values of correction missing information by auxiliary information, if the missing information (bits) increases, the number of correction sample sequence (waveform) candidates increases significantly, and the processing amount is reduced. May become realistic. The processing of the missing correction unit 430 to prevent such a problem and the functional configuration thereof will be described below.

FIG. 7 shows an example of the processing procedure, and FIG. 8 shows an example of the functional configuration of the missing correction unit 430. First, the provisional waveform generator 431 reproduces the provisional waveform (provisional sample string) in the frame using only the confirmed bits input from the rearrangement unit 220 (S1).
In the reproduction of the provisional waveform, missing bits are fixed to 0, for example, or set to an intermediate value that can be taken by missing bits. For example, if the lower 4 bits are missing, the value of the missing 4 bits is any value from 0 to 15, but is set to 8 or 7.

  Next, the spectrum envelope calculation unit 432 calculates the spectrum envelope coefficient of the provisional waveform (S2). For example, the spectrum envelope coefficient can be estimated by performing all-pole linear prediction analysis used in speech analysis on the provisional waveform. On the other hand, in the error calculation unit 433, the received auxiliary information Ax is decoded by the auxiliary information decoding unit 450 to obtain the spectrum envelope coefficient of the original sound, and the original sound spectrum envelope is compared with the spectrum envelope coefficient of the provisional waveform to determine the error. Is within the allowable range, the provisional waveform is output as a corrected output waveform signal to the frame synthesis unit 250 via the switch SW1 (S3).

If the error between the estimated spectral envelope coefficient and the decoded spectral envelope coefficient is not within the allowable range in step S3, first, the inverse characteristic of the estimated spectral envelope coefficient of the provisional waveform is given to the temporary waveform (S4). Specifically, using the linear prediction coefficient that represents the spectrum envelope coefficient of the provisional waveform obtained in step S2, for example, the provisional signal given to the all-pole linear prediction coefficient inverse filter (all zero type) 434 via the switch SW2. The spectrum of the provisional waveform is flattened by passing the waveform.
Next, the average power of the waveform signal whose spectrum is flattened is calculated by the power calculation unit 438, and the correction amount is calculated from the average power and the average power decoded from the auxiliary information decoding unit 450. For example, the ratio or difference between the two is calculated, and the power correction unit 501 performs amplitude correction on the flattened signal based on the correction amount. That is, the output of the inverse filter 434 is multiplied or added by a correction amount to match the decoding power (S5). Next, a provisional waveform in which the spectrum envelope coefficient is corrected is generated by multiplying the amplitude-corrected average signal by the spectrum envelope coefficient of the auxiliary information (S6). That is, a spectrum correction waveform is created through the output of the power correction unit 501 in the all-pole synthesis filter 435 using the parameter LPC representing the spectrum envelope of the auxiliary information. The resulting spectral envelope of the waveform is close to the original sound.

  However, since the amplitude value of the spectrum correction waveform may include a bit inconsistent with the already determined bit in the amplitude value of the missing waveform, the correction unit 436 corrects it to a correct value (S7). For example, when the lower 4 bits of the amplitude value of 16-bit precision are unknown, the uncertainty range of the value that each sample can take is 16, so the value closest to the spectrum correction waveform among those possible values To correct. That is, when the sample value corrected for each sample is out of the possible range (XXXXXXXXXXXX0000 to XXXXXXXXXXXX1111), it is corrected to the limit value closest to the sample of the spectrum correction waveform in the possible range. With this modification, all the bits whose amplitude values have been determined match and the spectrum envelope approaches that of the original sound at the same time.

This corrected waveform can be returned to the provisional waveform generation unit 431 via the switch SW3 as the provisional waveform of step S1, and the processing after step S2 can be repeated. Note that the waveform (sample value) is premised on an integer value, but it is treated as a real number in the filter calculation, and the output value of the filter must be converted to an integer. In the case of a synthesis filter, the result differs depending on whether it is converted into an integer for each sample or later in a batch for each frame, but both are possible.
As shown by broken lines in FIG. 7 and FIG. 8, after the provisional waveform is flattened in step S4, the flattened provisional waveform (flattened signal) is first passed through the synthesis filter 435 and the spectrum envelope corrected. A constituent sample series (waveform) may be obtained (S5 ′), the amplitude of the spectrum envelope corrected waveform may be corrected by the power correction unit 501 ′ (S6 ′), and the process may proceed to step S7. In this case, the average power of the spectrum envelope-corrected waveform from the synthesis filter 435 is calculated by the power calculation unit 438 ′, and this average power and the decoding power of the auxiliary information (the output of the power calculation unit 356 in the encoder 10) are calculated. ), The correction amount calculation unit 439 ′ obtains the correction amount, and the power correction unit 501 ′ performs amplitude correction on the output of the synthesis filter 435 based on the correction amount.

Next, another example of missing information correction processing using auxiliary information will be described with reference to FIG.
Steps S1 to S3 and S6 are the same processes as steps S1 to S3 and S7 in the process shown in FIG. In this example, in step S4 of FIG. 9, one synthesis is performed by combining the inverse filter 434 having the inverse characteristic with respect to the spectrum envelope estimated in step S2 and the synthesis filter 435 using the linear prediction coefficient LPC representing the spectrum envelope in the auxiliary information. The filter coefficient of the filter unit 435 ′ is calculated (S4), and the correction waveform is synthesized through the provisional waveform through this synthesis filter in step S5. Amplitude correction is performed on the corrected spectrum envelope waveform (S6). This amplitude correction is performed by a power calculation unit 438 ′, a correction amount calculation unit 439 ′, and a power correction unit 501 ′ indicated by broken lines in FIG.

  The functional configuration of the process shown in FIG. 9 is based on the estimated spectral envelope coefficient from the spectral envelope calculation unit 432 and the decoded spectral envelope coefficient from the auxiliary information decoding unit 450, as indicated by a broken line in FIG. The filter coefficient of one synthesis filter unit 435 ′ combining the filter 435 and the synthesis filter 435 is calculated by the synthesis spectrum envelope calculation unit 437, and the filter coefficient is set in the synthesis filter unit 435 ′. Is passed. The power correction unit 501 ′ performs amplitude correction on the output of the synthesis filter 438.

  As shown in FIG. 10, for example, the synthesis filter unit 435 ′ calculates the filter coefficient by converting the linear prediction coefficient obtained as the provisional waveform spectrum envelope coefficient into the linear prediction cepstrum coefficient Ca by the coefficient conversion unit 437a. A linear prediction coefficient obtained as a spectrum envelope coefficient is converted into a linear prediction cepstrum coefficient Cb by a coefficient conversion unit 437b, and Cb-Ca is obtained by subtracting these coefficients Ca and Cb by a subtraction unit 437c, and this Cb-Ca is inversely converted. The unit 437d performs inverse conversion to a linear prediction coefficient, and this linear prediction coefficient may be used as the filter coefficient of the synthesis filter 438. For the conversion to the prediction coefficient, for example, a method described in Japanese Patent Laid-Open No. 8-248996 “Filter coefficient determination method of digital filter” may be used.

In the processing shown in FIG. 9, calculation by the combined spectrum envelope calculation unit 437 is necessary, but the filter processing operation for the provisional waveform is only required once. The correction processing shown in FIG. 7 and FIG. 9 may be performed by converting into the frequency domain.
Spectral correction based on LPC decoded from auxiliary information in steps S6 (or S5 ′) and S5 in the loops of the flowcharts shown in FIGS. 7 and 9 is to reduce the spectral distortion of the original sound and the reproduced signal. The correction of S7 is to reduce the waveform distortion of the original sound and the reproduction signal, and the correction amount can be reduced to prevent the correction waveform from being divergenced. 7 and 9, the linear prediction coefficient α _k of the inverse filter 434 and the linear prediction coefficients β _k and β _k ′ of the synthesis filters 435 and 438 are both raised to the kth power of a constant γ that is less than 1. There is a method of multiplying (γ ^k ) (k is the order of the parameter). In the example shown in FIG. 10, a constant less than 1 may be applied to all the linear prediction cepstrum coefficients Ca and Cb. 7 and 9, it is also possible to set γ to a value close to 1 at the beginning of the iteration, and to reduce the estimation error by decreasing the value of γ as it converges.

Multiplying α _k , β _k , β _k ′ by γ ^k, and multiplying the linear prediction cepstrum coefficient by a constant less than 1 will multiply the prediction coefficient by expanding the band of the spectral envelope characteristic. The spectrum envelope of the provisional waveform (sample sequence) and the decoded spectrum envelope of the auxiliary information are each rounded.
7 and 9, step S3 is omitted, and steps S1, S2, S4, S5 (S5 ′), S6 (S6 ′), and S7 are repeated only once or a predetermined number of times, and a corrected waveform ( Sample sequence) may be output. Further, as shown in the figure, when the number of repetitions exceeds the specified number in step S3, the corrected waveform obtained at the end may be output.
Third Embodiment The present invention is applied to a coding method that combines the lossy coding with a high compression ratio and the lossless compression of the reproduced signal and the error signal of the original signal, as described with reference to FIG. 1 of the prior art. The third embodiment is shown in FIG. As in FIG. 1, in the encoder 10, the digital input signal sequence from the input terminal 100 is irreversibly quantized by the irreversible quantization unit 120 having a high compression rate, and the quantized information is dequantized by the inverse quantization unit 130. And the difference between the inversely quantized signal (reproduced signal) and the original digital input signal sequence from the input terminal 100 is taken by the subtractor 140, and an error signal which is this difference signal Is supplied to the frame dividing unit 110. In other words, the error signal for each frame is rearranged in bit arrangement, divided into transmission recording unit data, and lossless compression of the transmission recording unit data is performed as described above, and the header 31 is attached to the packet by the header adding unit 320. It becomes.

  The quantization code I (n) from the irreversible quantization unit 120 is also supplied to the header adding unit 320, and the header 31 is attached to form a packet. At this time, when giving priority, a code indicating the highest priority is given to the bit string of the quantization code. In addition, as indicated by the lines in FIG. 4, the error signal or sample sequence is generated by the auxiliary information generating unit 350 as auxiliary information in which the parameter expressing the spectral envelope of the original input signal sample sequence and the average power are generated as auxiliary information. It is stored and transmitted as a packet or in a high priority packet.

  The decoder 20 is an example of performing correction after the original sound is synthesized, that is, after adding the outline and error waveform of the original sound signal waveform reproduced by inverse quantization by the adder 240 and reproducing the original sound signal waveform. The irreversible quantization code I (n), the auxiliary information, and the transmission recording unit data are separated from the packet input to 200 by the separation unit 440, and the irreversible quantization code I (n) is separated from the inverse quantization unit 230. Then, the auxiliary information is decoded by the auxiliary information decoding unit 450, and the parameter and average power representing the decoded spectrum envelope are supplied to the missing correction unit 430. On the other hand, as described above, transmission recording unit data is reversibly expanded, transmission recording unit integration, and rearrangement are sequentially performed, and an amplitude bit string, that is, an error signal sample string is reproduced. The addition unit 240 adds the inverse quantized signal. This addition signal is supplied to the loss correction unit 430 when a packet loss is detected by the loss detection unit 420.

  Any of the various methods described above may be used as the correction method in the missing correction unit 430. When the parameter representing the auxiliary information decoding spectrum envelope is used, if the inverse quantization unit 230 can obtain the parameter representing the corresponding spectrum envelope, this may be used. Alternatively, as indicated by a broken line in the figure, in the encoder 10, the auxiliary information generating unit 350 uses the output error signal of the frame dividing unit 110 or the input signal from the input terminal 100 divided into frames by the frame dividing unit 110 ′. And this auxiliary information is added in any packet or output by an independent packet. In the decoder 20, the auxiliary information is separated by the separation unit 440, and the auxiliary information is decoded by the auxiliary information decoding unit 450 and supplied to the missing correction unit 430.

  When the lossy quantization code and the error signal are used in this way, the decoder 20 may perform missing information correction on the error signal before the original sound synthesis. That is, for example, as shown in FIG. 12, when the packet is missing, the rearrangement output from the rearrangement unit 220 is supplied to the missing correction unit 430 to correct the missing information and is supplied to the frame synthesis unit 250. The correction by the missing correction unit 430 can use the various methods described above. When the auxiliary information decoding spectrum envelope is used, the decoding output of the auxiliary information decoding unit 450 is used, or the inverse quantization unit 230 is used. If a parameter representing the spectral envelope to be obtained is obtained, this may be used. The adder 240 adds the reproduction error signal from the frame synthesizer 250 and the inverse quantized signal from the inverse quantizer 230 to reproduce the original sound signal.

Also in the embodiment shown in FIGS. 11 and 12, the average power may not be used as auxiliary information.
As shown in FIG. 11, in the encoder 10, when the error signal is frame-divided by the frame dividing unit 110, the frame division is performed on the quantization processing frame (for example, 1024 samples) in the irreversible quantization unit 120. By making the divided frame of the unit 110 longer, for example, about 16 times, the compression efficiency in the lossless encoding unit 150 can be increased. However, if the divided frame length in the frame dividing unit 110 is made too long, the delay becomes longer in decoding.

In each embodiment shown in FIGS. 6 and 11, the rearrangement unit 160 does not necessarily include the polarity absolute value conversion unit 161. That is, the bit arrangement may be rearranged with the 2's complement representation. In this case, the efficiency of lossless encoding is reduced, but it is effective for applications in which only selection of amplitude resolution (scalability of amplitude accuracy) is a problem rather than compression rate.
In the example shown in FIGS. 11 and 12, not only the signal from the terminal 260 but also the output from the inverse quantization unit 230 may be used as the reproduction signal in response to a request for fidelity of the reproduction signal.
In each of the above-described embodiments, the encoder 10 and the decoder 20 can function by causing a computer to execute a lossless encoding program and a lossless decoding program. In these cases, a lossless encoding program and a lossless decoding program are downloaded to a program memory of the computer from a CD-ROM, a flexible magnetic disk, or the like or through a communication line.

The effect of the present invention was calculated by computer simulation for the third embodiment (FIG. 11 for encoding and FIG. 12 for decoding). The number of digits of the sample value of the error signal (including the polarity signal) is 16, the 6th to 10th bits are missing, and the missing signal correction unit 430 corrects the error signal by smoothing it with a low-pass filter, and outputs the decoded signal (output) The SNR (Weighted SNR) of the terminal 260) corrected for auditory sensation and the cepstrum distance (spectrum envelope distortion) of the decoded signal with respect to the original sound signal were obtained. The results are shown in FIGS. 13A and 13B. For reference, the SNR and cepstrum distance when the missing information is not corrected are also shown. From these, it is understood that when the sixth bit is missing, the missing information correction (missing correction) according to the present invention is considerably improved as compared with the case where the missing correction is not performed.
Fourth Embodiment FIG. 14 shows another embodiment of the encoder 10 and the decoder 20 according to the present invention. This embodiment is a modification of the auxiliary information generator 350 in the embodiment of FIG. As described above, in the encoder 10, the input signal sample sequence from the input terminal 100 is sequentially divided by the frame dividing unit 110 into frame units including, for example, 1024 samples. In this embodiment, the effective digit number detecting unit 353 in the auxiliary information generating unit 350 detects the number of digits representing the maximum absolute value of the input digital signal for each frame as the effective digit number Fe.

  For each frame divided by the frame dividing unit 110, only the portion within the effective number of digits of the input signal sample series is arranged in the time direction by the reordering unit 160 in which the corresponding bits in each sample (amplitude bit string) are arranged. At this time, each input signal sample is expressed in the complement of a positive or negative integer having an amplitude of 2, and each sample expressed in the 2's complement is converted into a polarity absolute value converter 161 as in the description of the prior art. The bit sequence (equal bit sequence) in which the corresponding bits (digits) of each sample (amplitude bit sequence) are concatenated in the time direction by the bit sequence conversion unit 162. It is preferable to reappear. However, as indicated by a broken line in the figure, the bit array conversion unit 162 re-arranges the corresponding bits into bit sequences in the time direction without using the polarity absolute value conversion unit 161, that is, in the 2's complement representation. May be.

  That is, a normal input signal sample represents a positive or negative integer with two's complement, but in this example, the input signal sample is converted into a binary number with a polarity code and an absolute value of the amplitude, and then, etc. Conversion to a unit bit string is performed, and further, the transmission recording unit division unit 310 divides the data into transmission recording unit data. An example is shown in FIG. 15A. As shown in the amplitude bit string on the left side of FIG. 15A, one frame of each sample expressed as a polarity code and an absolute value is sequentially arranged in the time direction. In order to make it easy to understand the state of one amplitude bit string, n represents an amplitude bit string DV (n) surrounded by a thick line, and n represents a time in the frame, for example, n = 1, 2,. In this example, the polarity code of each amplitude bit string DV (n) is adjacent to the absolute value MSB, and in the figure, the polarity code is located immediately above the MSB.

A bit position having “1” in the digit closest to the MSB in the amplitude bit string of each sample for one frame is detected, and the number of digits from the LSB to that digit is obtained as the effective digit number Fe. Only the range 361 within the effective number of digits and the polarity code in the digital signal of one frame are converted into a bit string. That is, the range 362 from the digit higher than the effective number of digits to the MSB is not converted into a coordinate bit string.
In this example, as shown on the right side of FIG. 15A, such amplitude bit string data is first connected in the time direction with only the polarity bit (sign) of the numerical value of the amplitude of each sample, and is a series (coordinate bit string) within the frame. make. Next, a digit (coordinate bit string) is created by concatenating only the digits corresponding to the largest numerical value within the effective number of digits Fe within the frame, and in the same manner, sequentially connected in the time direction for each digit (corresponding bit). Thus, a coordinate bit string is created, and finally, a bit string (equal bit string) in which only LSBs are connected in a frame is created. One example of these coordinate bit strings is shown as DH (i) in the bit arrangement thick frame on the right side in FIG. 15A. i indicates the order in which coordinate bit strings are created. The above sort does not change the content of the data in the frame.

  When the digital signal expresses a positive or negative integer in two's complement, for example, as shown in FIG. 15B, an amplitude bit string for one frame, a portion having an absolute value larger than the digit expressing the maximum absolute value is As shown in a range 363 in FIG. 15B, if the amplitude bit string is a positive value, all are “0”, and if it is a negative value, all are “1”. The number of digits in the range 364 other than such a portion is detected as the number of significant digits. Only the range 364 and the bits (digits) adjacent thereto, that is, the polarity code, may be converted into a coordinate bit string as shown on the right side of FIG. 15B.

The output from the rearrangement unit 160 is divided into transmission recording unit data by the transmission recording unit dividing unit 310 for each equal bit string DH (i) or for each of a plurality of adjacent coordinate bit strings DH (i). In this case, in one frame, data composed of one coordinate bit string and data composed of a plurality of coordinate bit strings may be mixed as transmission recording unit data.
Each transmission record unit data divided in this way is attached to the transmission record unit data (payload) 32 by the header attaching unit 320, for example, as shown in FIG. 4C.

  In the fourth embodiment, the auxiliary information generating unit 350 includes a spectrum envelope calculating unit 351, an auxiliary information encoding unit 352, an effective digit number detecting unit 353, and a power calculating unit 354, and an input signal from the frame dividing unit 110 The effective digit number Fe detected by the effective digit number detection unit 353 from the sample string is encoded by the auxiliary information encoding unit 352 and transmitted. Alternatively, if each sample has an m-bit configuration, it is obvious that m-Fe may be encoded and sent instead of the effective digit number Fe. In this example, the input signal sample sequence divided for each frame by the frame dividing unit 110 is also obtained by the spectral envelope calculation unit 351 as a parameter for expressing the spectral envelope by the linear prediction coefficient LPC by linear prediction analysis, for example. The unit 354 calculates the average power Pa of the divided sample sequence for each frame. Alternatively, the input signal sample sequence is input to the inverse filter 355 configured based on the linear prediction coefficient LPC obtained by the spectrum envelope calculation unit 351 to flatten the spectrum envelope, and the average power Pa of the flattened signal is the power. It is obtained by the calculation unit 356.

  The linear prediction coefficient LPC and the average power Pa are also encoded by the auxiliary information encoding unit 340 with, for example, a low bit of about 30 to 50 bits to be auxiliary information. The auxiliary information obtained by encoding the effective digit number Fe, the spectral envelope parameter LPC, and the average power Pa is supplied to the header adding unit 320, and a representative packet of each frame, for example, a packet in which transmission recording unit data including a polarity code is stored. It is added inside or output as an independent packet. As indicated by a broken line in the figure, the auxiliary information such as the spectral envelope parameter LPC and the average power Pa is converted by the frame dividing unit 110 ′ into the frame length longer than the division length by the frame dividing unit 110 as in the embodiment of FIG. The efficiency of lossless compression may be increased by acquiring the input signal frame divided by.

  In the decoder 20, the packet 30 input to the input terminal 200 is separated into transmission recording unit data and auxiliary information in the separation unit 440, and the transmission recording unit data is converted into the lossless decoding unit 210 (the decoding unit 210 in FIG. 1). The auxiliary information is supplied to the auxiliary information decoding unit 450, and the auxiliary information decoding unit 450 decodes the effective digit number Fe of the frame, the parameter LPC indicating the spectrum envelope and the average power Pa, and the effective number of digits. Fe is supplied to the digit aligning unit 460, and the spectral envelope parameter LPC and the average power Pa are supplied to the missing correction unit 430. The missing correction unit 430, the auxiliary information decoding unit 450, and the digit alignment unit 460 constitute an information correction unit 480.

  If the transmission record unit data is losslessly compressed, the lossless decoding unit 210 performs lossless decoding. Each transmission record unit data is transmitted from the plurality of packets by one frame based on the packet number in the transmission record unit integration unit 410. Are integrated, for example, as shown in the coordinate arrangement on the right side of FIG. 15A. The integrated data is converted by the rearrangement unit 220 from an equivalence bit string to an amplitude bit string, that is, an original sample string (waveform). At this time, when each sample is expressed by the polarity code and the absolute value of the amplitude, the bit array inverse transform unit 221 reverses the rearrangement described with reference to FIG. After rearrangement into the amplitude bit string, the 2's complement conversion unit 222 converts each amplitude bit string from the bit array inverse conversion unit 221 into a 2's complement expression, that is, an amplitude bit string having a negative polarity sign is “1”. And "0".

  If the transmission record unit data is based on the amplitude bit string that is expressed in two's complement form and directly rearranged into the equidistant bit string, the bit array inverse conversion unit 221 starts from the bit array inverse conversion unit 221 as shown by a broken line in FIG. The amplitude bit string is supplied to the digit aligning unit 460 as it is. Digit aligning section 460 aligns each amplitude bit string according to the number of significant decoding digits Fe. In other words, 0 is added to the higher order of the amplitude bit string so that the number of bits (digits) m of the original amplitude bit string is positive, and 1 is added if the polarity code is negative, that is, for example, the range 363 in FIG. Add a bit of. The digitized amplitude bit string is supplied to the frame synthesis unit 250 as a decoded sample.

When a packet loss occurs, the packet number of the packet is detected by the loss detection unit 420, and the amplitude bit string from the rearrangement unit 220 is supplied to the loss correction unit 430 without being directly supplied to the digit alignment unit 460. Then, correction for the missing information is performed on the amplitude bit string (sample) and supplied to the digit aligning unit 460.
The correction in the missing correction unit 430 is performed by estimating missing information from known information. As in each of the embodiments described above, the missing correction unit 430 adds the missing frame to the average spectrum of the past several frames, or the determined spectrum of the frame obtained as a result of decoding the auxiliary information as described later. Correction for missing information is performed so that a spectrum obtained from information other than information approaches. As a simple correction method, the missing correction unit 430 may smooth the input reproduction sample sequence through a low-pass filter to remove high-frequency noise components. If the spectral shape (envelope) of the original signal is known in advance, the cutoff characteristic of the low-pass filter is selected so as to attenuate the high-frequency component according to the characteristic. Alternatively, as described above, an average spectrum may be obtained, or the cutoff characteristic may be adaptively changed according to the shape of the spectrum determined for each frame.

In this way, by utilizing the fact that the decoder 20 can correct the information loss due to the packet loss, the LSB side packet is not intentionally sent as necessary to reduce the amount of information, and even if the bit rate is lowered, the decoder 20 Makes it possible to perform lossless decoding. Alternatively, it is possible to perform reproduction that does not cause a problem in hearing.
Alternatively, all combinations of possible values of missing information (bits) are added to each sample value to create correction sample string (waveform) candidates, and the spectral envelopes of these candidates are obtained. A correction sample string (waveform) candidate corresponding to the one closest to the decoded spectrum envelope is output to the digit aligning unit 460 as a correction sample string. In FIG. 14, the lossless encoding unit 150 and the lossless decoding unit 210 may be omitted.
When correction sample sequence candidates are generated using all combinations of possible values of missing information, correction sample sequence (waveform) candidates increase remarkably as missing information (bits) increases. The amount of processing may increase unrealistically. The processing of the missing correction unit 430 that avoids such a problem and its functional configuration will be described below.

  FIG. 16 shows an example of the missing information correction processing procedure, and FIG. 17 shows an example of the functional configuration of the missing correction unit 430. The processes in steps S1 to S6 are the same as those in steps S1 to S4, S6, and S7 in FIG. 7. First, only the confirmed bits input from the rearrangement unit 220 to the temporary waveform generation unit 431 are used. A provisional waveform (provisional sample string) is reproduced (S1). In the reproduction of the provisional waveform, missing bits are fixed to 0, for example, or set to an intermediate value that can be taken by missing bits. For example, if the lower 4 bits are missing, the value of the missing 4 bits is any value from 0 to 15, but it is set to 8 or 7.

  Next, the spectrum envelope of the provisional waveform is calculated by the spectrum envelope calculation unit 432 (S2). For example, spectral envelope coefficients can be analyzed by performing all-pole linear prediction analysis used in speech analysis on provisional waveforms. On the other hand, the error calculation unit 433 decodes the received auxiliary information Ax by the auxiliary information decoding unit 450 to obtain the spectrum envelope coefficient of the original sound, and compares the original sound spectrum envelope coefficient with the spectrum envelope coefficient of the provisional waveform to determine the error. Is smaller than the predetermined value Δd, the provisional waveform is output to the frame synthesis unit 250 as a corrected output waveform (S3).

  In step S3, if the error between the estimated spectral envelope coefficient and the decoded spectral envelope coefficient is not smaller than the predetermined value Δd, first, the inverse characteristic of the estimated spectral envelope coefficient of the temporary waveform is given to the temporary waveform (S4). Specifically, the parameter representing the provisional waveform spectrum envelope obtained in step S2 is set in, for example, the all-pole type (all zero type) linear prediction inverse filter 434, and the provisional waveform is passed through the inverse filter 434 to pass the provisional waveform. The spectrum is flattened to obtain a flattened signal. The average power of the flattened signal is calculated by the power calculation unit 438 and corrected from the average power and the decoded average power from the auxiliary information decoding unit 450 (output of the flattening power calculation unit 331 of the encoder 10). The amount is calculated by the correction amount calculation unit 439, for example, as a ratio or difference between the two, and the power correction unit 501 performs amplitude correction on the output of the inverse filter 434 based on the correction amount. That is, the output of the inverse filter 434 is multiplied or added by the correction amount to match the power of the provisional waveform with the decoding power (S5).

Next, the spectral envelope characteristic of the auxiliary information is given to the flattened signal whose amplitude has been corrected to correct the spectral envelope (S6). That is, a spectrum correction waveform is created through the output of the power correction unit 501 in the all-pole type synthesis filter 435 using the parameter LPC representing the decoded spectrum envelope of the auxiliary information. The resulting spectral envelope of the waveform is close to the original sound.
However, since this spectrum correction waveform may be inconsistent with the already determined bit in the amplitude bit string, it is corrected by the correcting unit 436 to the correct value as described above (S7).

  By using this corrected waveform as a provisional waveform in step S1, the processing after step S2 can be repeated. In addition, when the decoding effective digit number Fe is different for each frame, it is a target of each process of the linear prediction analysis (step S2), the inverse filter 434 (step S4), and the synthesis filter 435 (step S6) of the spectrum envelope calculation unit 432. Samples may span current and past frames. In this case, even if the target of processing is the current frame, it is necessary to perform analysis and filtering processing after matching the effective digit number Fe of the previous frame with the effective digit number Fe of the current frame. . If the number of significant digits Fe in the past frame is N digits greater than the number of significant digits in the current frame, the amplitude value is reduced by shifting the sample of the past frame, for example, by N digits lower. Match the number of digits to the number of significant digits in the current frame. Conversely, if the number of significant digits of one previous frame is M digits greater than the number of significant digits of the current frame, the amplitude of the sample of the past frame is temporarily shifted to the upper M digits by, for example, floating point display. Increase the value to match the number of significant digits to that of the current frame. In this case, the accuracy of the amplitude value of the sample of the past frame is lowered when the information overflows from the register due to the upper shift and the information loss is large, so the sample of the past frame is not used or the sample of the current frame is not used. The correction process may be omitted.

  As shown by the broken line in FIG. 16, such correction of the effective number of digits is performed before the above-described effective number of digits correction (S2 ') when necessary for the analysis processing in step S2 (step S4'). If necessary for the inverse filter processing, the effective digit number correction is performed before (S4 ′), and if necessary for the synthesis filter processing in step S6, the effective digit number correction is performed before that (S6 ′). In FIG. 17, the significant digits decoded from the auxiliary information decoding unit 450 as shown by a broken line for those requiring samples of past frames in the spectrum envelope calculation unit 432, the inverse filter 434, and the synthesis filter 435. The number Fe is also supplied, and after the spectrum envelope calculation unit 432, the inverse filter 434, and the synthesis filter 435 perform the process of matching the number of significant digits for the sample of the past frame with the number of significant digits of the current frame, the original process To be done.

  As shown by the broken lines in FIGS. 16 and 17, after the provisional waveform is flattened in step S4, the flattened provisional waveform (flattened signal) is first passed through the synthesis filter 435 and the spectrum envelope corrected. A constituent sample series (waveform) may be obtained (S5 ′), the amplitude of the spectrum envelope corrected waveform may be corrected by the power correction unit 501 ′ (S6 ″), and the process may proceed to step S7. The average power of the spectrum envelope-corrected waveform from the synthesis filter 435 is calculated by the power calculation unit 438 ′, and the average power and the decoding power of the auxiliary information (output of the power calculation unit 331 in the encoder 10) are calculated. The correction amount is calculated by the correction amount calculation unit 439 ′, and the amplitude correction is performed by the power correction unit 501 ′ on the output of the synthesis filter 435 based on the correction amount, and the processing by the synthesis filter 435 in step S5 ′ is 1. When it is performed across the previous frame, the number of effective digits is corrected in advance as shown in step S5 ″.

Still another example of the missing information correction process using the correction information will be described with reference to FIG. This missing information correction process uses the same principle as the correction process using the combined spectrum envelope calculation unit 437 and the combined filter unit 435 ′ in FIG.
Steps S1 to S3 and S6 are the same processes as steps S1 to S3 and S7 in the process shown in FIG. In this example, after step S3, the filter coefficient of one synthesis filter unit 435 ′, which is a combination of the inverse filter 434 of the spectrum envelope estimated in step S2 and the synthesis filter 435 of the decoded spectrum envelope of the auxiliary information, is used as a synthesis spectrum envelope calculation unit 437. (S4), and in step S5, a waveform obtained by correcting the spectrum envelope through the provisional waveform is synthesized in the synthesis filter unit 435 ′, and the amplitude correction is performed by the power correction unit 501 ′ on the waveform whose spectrum envelope has been corrected. (S6). The amplitude correction is performed by a power calculation unit 438 ′, a correction amount calculation unit 439 ′, and a power correction unit 501 ′ indicated by broken lines in FIG. The calculation of the filter coefficient of the synthesis filter 501 ′ is performed by the calculation described with reference to FIG.

In the processing shown in FIG. 18, it is necessary to calculate the filter coefficient of the synthesis filter unit 435 ′ by the synthesis spectrum envelope calculation unit 437, but the filter operation for the provisional waveform only needs to be performed once. The correction processing shown in FIGS. 16 and 18 may be performed by converting into the frequency domain.
In FIG. 18, when the spectral envelope calculation in step S2 or the spectral envelope correction in step S5 requires a past sample, the above-described effective digit correction is performed before the step S2 or S5 as indicated by the broken line. (S2 'or S5').

The spectral envelope correction based on LPC decoded from the auxiliary information in step S6 (or S5 ′) in the loop of the flowchart shown in FIG. 16 and S5 in the loop of the flowchart shown in FIG. Since the correction in step S7 in both flowcharts is to reduce the waveform distortion of the original sound and the reproduced signal, there is no guarantee that the operation will be converged by repeating this operation by the loop, but FIGS. 7B and 9 will be described. Similarly, the divergence can be prevented by multiplying the linear prediction coefficient α _k of the inverse filter 434 and the linear prediction coefficients β _k and β _k ′ of the synthesis filters 435 and 438 by a constant γ ^k . However, γ is 0 <γ <1, and k is the order of the parameter.

When the configuration of FIG. 10 is used, it is only necessary to apply constants less than 1 to all of the linear prediction cepstrum coefficients Ca and Cb. Also in the case of iterative processing in FIGS. 16 and 18, it is also possible to set γ to a value close to 1 at the beginning of the iteration, and to reduce the estimation error by decreasing the value of γ as it converges.
16 and 18, step S3 is omitted, and steps S1, S2, S4, S5 (S5 '), S6 (S6'), and S7 are repeated only once or a predetermined number of times, and a corrected waveform ( Sample sequence) may be output. Further, as shown in the figure, when the number of repetitions exceeds the specified number in step S3, the corrected waveform missing correction unit 430 finally obtained at that time may be output.

The digit alignment unit 460 may be inserted immediately after the rearrangement unit 220 as indicated by a broken line in FIG. In this case, the process for adjusting the number of significant digits in the missing correction unit 430 is not necessary. In the missing information correction process shown in FIGS. 16 and 18, the amplitude correction in steps S5, S6 ′ and S6 may be omitted. In this case, the corresponding portions 438 ', 439', and 501 'are omitted in FIG. In some cases, correction of missing information using the decoded spectral envelope is omitted, and other corrections such as a method using the above-described low-pass filter or a method of correcting to match the spectral envelope coefficient of the past frame may be used. . In this case, the generation of the spectrum envelope or the auxiliary information of the average power is omitted in the encoder 10, and the missing correction unit 430 of the decoder 20 is different from the configuration shown in FIG.
Fifth Embodiment The encoding method for rearranging the significant digits of the samples in each frame described in FIG. 14 is the lossy encoding with a high compression ratio described in FIG. 1, and the error between the reproduced signal and the original signal. FIG. 19 shows a fifth embodiment applied to an encoding method combined with lossless compression of a signal. In the encoder 10 of this embodiment, the bit array conversion unit 162 rearranges the significant digits described in FIG. 15A or 15B. The auxiliary information generation unit 350 is configured in the same manner as the auxiliary information generation unit 350 in the embodiment of FIG. Other than that, the configuration is the same as that of the encoder 10 in FIG. Further, in the decoder 20 in the embodiment of FIG. 11, the decoder 20 is provided with a digit aligning unit 460 on the output side of the rearranging unit 220, and the missing correction unit 430 of FIG. The correction unit 430 is used, and the other configuration is the same as that of the decoder 20 in FIG. Therefore, detailed description will not be repeated.

  When the lossy quantization code and the error signal are used in this way, the decoder 20 may perform missing information correction on the error signal before the original sound synthesis. That is, for example, as shown in FIG. 20, when the reordering output from the rearrangement unit 220 is missing, it is supplied to the missing correction unit 430 to correct the missing information, and is supplied to the digit alignment unit 460. If not included, the output of the rearrangement unit 220 is directly supplied to the digit alignment unit 460, and the sample string aligned by the digit alignment unit 460 is supplied to the frame synthesis unit 250. The various correction methods described above can be used for correction by the missing correction unit 430. When the auxiliary information decoding spectrum envelope or the decoding average power is used, the decoding output of the auxiliary information decoding unit 450 is used, or the inverse quantum is used. When the parameter LPC representing the corresponding spectrum envelope is obtained by the conversion unit 230, this may be used. The adder 240 adds the reproduction error signal from the frame synthesizer 250 and the inverse quantized signal from the inverse quantizer 230. As indicated by a broken line in FIG. 20, the digit alignment unit 460 may be provided immediately after the rearrangement unit 220.

In the above description, the number of significant digits and transmission record unit data may be transmitted at least for each frame without being packetized by the encoder 10, and the decoder 20 may decode them using these.
Sixth Embodiment FIG. 21 shows embodiments of the encoder 10 and the decoder 20 according to the present invention. In this embodiment, instead of converting the input signal sample string divided into frames into a coordinate bit string and transmitting it in the embodiment of FIG. 6, the prediction error of the input signal sample string is converted into a coordinate bit string and transmitted. It is.

  The encoder 10 has a configuration in which a prediction error generation unit 370 including a sample register 371, a linear prediction unit 372, an integerization unit 373, and a difference circuit 374 is added to the encoder 10 in FIG. . Further, the decoder 2 has a configuration in which a synthesis filter 470 including a sample register 471, a linear prediction unit 472, an integerization unit 473, and an addition unit 474 is added to the decoder 20 in FIG. . An input signal sample sequence is supplied from the frame dividing unit 110 to the spectrum envelope calculation unit 351 of the auxiliary information generation unit 350 and the difference circuit 374 of the prediction error generation unit 370 for each frame. The input signal sample sequence is subjected to, for example, linear prediction analysis by the spectrum envelope calculation unit 351 for each frame, and a linear prediction coefficient LPC is obtained as a parameter representing the spectrum envelope. The spectrum envelope parameter LPC is encoded by the auxiliary information encoding unit 352.

For example, a predetermined number of a plurality of samples of the immediately preceding frame from the frame dividing unit 110 are supplied from the register 371 to the linear prediction unit 372, and the linear sequence based on the spectrum envelope parameter LPC from the spectrum envelope calculation unit 351 is supplied to these sample sequences. The prediction coefficient is multiplied and added to obtain a linear prediction value for each input sample. This linear prediction value is converted into an integer value by the integer converting unit 373, and the difference between the integer predicted value and the current sample from the frame dividing unit 110 is obtained as the prediction error signal sample Spe by the difference circuit 374.
The prediction error signal Spe obtained for each input sample is input to the rearrangement unit 160. In the rearrangement unit 160, the corresponding bit (digit) of the prediction error signal sample Spe (amplitude bit string) for each input sample is obtained for each frame. As described in FIG. 4A, they are arranged in the time direction. The equidistant bit string from the reordering unit 160 is losslessly compressed and encoded by the lossless encoding unit 150 by the transmission / recording unit dividing unit 310 for each data of the transmission unit or recording unit as necessary, and decoded by the header adding unit 320. A header is attached so that the transmission record unit data divided at times can be reconstructed as one frame, and is output to the output terminal 170 as a packet.

The encoded information (auxiliary information) of the spectral envelope parameter LPC from the auxiliary information encoding unit 350 is output from the header adding unit 320 as one packet or stored in the highest priority packet.
In the decoder 20, the packet 30 from the input terminal 200 is separated into auxiliary information and transmission recording unit data (including a polarity code string) in the separation unit 440, and the auxiliary information is supplied to the auxiliary information decoding unit 450. . The transmission recording unit data is supplied to the lossless decoding unit 210 when it is losslessly compressed. After the lossless decoding, the transmission recording unit data is supplied to the transmission recording unit integration unit 410. The transmission record unit integration unit 440 integrates transmission record unit data for one frame from a plurality of packets based on the packet number. In the rearrangement unit 220, the integrated data is converted into an equivalent bit string for one frame of an amplitude bit string, that is, a prediction error waveform.

  If the transmission record unit data is based on the amplitude bit string expressed in two's complement form and directly rearranged to the equidistant bit string, the bit array inverse conversion unit 221 starts from the bit array inverse conversion unit 221 as shown by a broken line in FIG. Are directly supplied as decoding samples to the synthesis filter 470, bypassing the missing correction unit 430. If no packet loss occurs, the rearrangement unit 220 obtains a prediction error signal Spe that is completely identical to the prediction error signal sample Spe input to the rearrangement unit 160 of the encoder 10. In the synthesis filter 470, a process reverse to the process of the prediction error generation unit 370 in the encoder 10 is performed. That is, a predetermined number of immediately preceding decoded samples are input to the linear prediction unit 472 from the register 471, and the linear prediction coefficient PLC decoded by the auxiliary information decoding unit 450 for each sample is input to the linear prediction unit 472, respectively. Multiplication is performed, and the sum of the multiplication results is obtained as a predicted value of the current decoded sample. The prediction value is converted into an integer value by the integerizing unit 463, and the sum of the integerized prediction value and the current prediction error signal from the rearrangement unit 220 is calculated by the adding unit 474, and the filter of the synthesis filter 470 is obtained. The output is supplied to the frame composition unit 250 and to the register 471. Accordingly, the digital signals are synthesized by the synthesis filter 470 and are combined with each other by the frame synthesis unit 250, and the signal input to the input terminal 100 of the encoder 10 is reproduced and output to the output terminal 260.

When a packet loss has occurred from the packet number of the input packet, this is detected by the loss detection unit 420, and the amplitude bit string from the rearrangement unit 220 is not directly supplied to the synthesis filter 470, but the loss correction unit 430. And the correction for the missing information is performed on the amplitude bit string (prediction error signal) and supplied to the synthesis filter 470.
The processing procedure and configuration for obtaining a prediction error waveform obtained by correcting missing information in the missing information correction missing correction unit 430 will be described below with reference to FIGS.

The missing prediction error waveform (sample frame output for one frame from the rearrangement unit 220) is input to the provisional waveform generation unit 431, and first, the provisional prediction error waveform in the frame is generated using only the determined bits ( S1). At this time, the missing bits are fixed to 0, for example, or set to an intermediate value within the range of missing information.
Next, the spectrum envelope of the provisional prediction error waveform is calculated by the spectrum envelope calculation unit 432 (S2). For example, the spectral envelope can be estimated by performing all-pole linear prediction analysis used in speech analysis on the provisional prediction error waveform. Since the spectrum envelope of the prediction error waveform generated by the prediction error generation unit 370 of the encoder 10 is substantially flat, this estimated spectrum envelope is obtained by the prediction error generation unit 370 of the encoder 10 as a provisional prediction error waveform. If it is the same as the original prediction error signal waveform, it is expected to be flat. However, if this provisional prediction error waveform is different from the original prediction error signal waveform, the spectrum envelope will not be flat. The flatness determination unit 433F determines whether or not the flatness is within the allowable range (S3). If the flatness is within an allowable range, the provisional prediction error waveform is output to the synthesis filter 470 as it is.

Determination of the flatness is, for example, linear prediction coefficients c _1, c ₂ were determined as a parameter of the spectral envelope in the spectral envelope calculator 432, ..., the values divided by the geometric mean of the arithmetic mean of c _M is at 0dB any For example, if it is completely flat, for example 3 dB or less, it is determined that it is almost flat. Alternatively, an LPC cepstrum coefficient is obtained as the spectrum envelope parameter, and if the sum of squares of the coefficient is equal to or less than a certain value, it is determined that the state is almost flat.
If it is determined that the spectrum envelope shape is significantly different from the flat state, first, the inverse characteristic of the estimated spectrum envelope is given to the provisional prediction error waveform (S4). More specifically, for example, the provisional prediction error waveform is passed through an all-pole linear prediction inverse filter (all zero type) 434 to flatten the spectrum. At this time, instead of completely flattening, the bandwidth of the spectral envelope characteristic may be expanded, that is, flattened by a prediction coefficient with a smoothed spectral envelope.

  However, since this flattened correction waveform may be inconsistent with a bit having a fixed amplitude obtained from a normally received packet, the correction unit 436 corrects it to a correct value (S5). For example, if the lower 4 bits of the amplitude value of 16-bit precision are unknown, there are 16 ranges of uncertainty in the values that each amplitude bit string (prediction error signal) can take. If there is an amplitude bit string out of the range, the value is corrected to the closest value, that is, the amplitude bit string having the maximum value of the lower 4 bits in this example. With this modification, all the bits whose amplitude values are determined are matched, and at the same time, a waveform having a spectrum envelope close to that of the original prediction error signal is reproduced. If necessary, the corrected waveform is used as a provisional prediction error waveform and the process returns to step S1 to repeat the correction.

Although the corrected prediction error (provisional prediction error) waveform is premised on an integer value, it is treated as a real number in the filter calculation, and the output value of the filter needs to be converted to an integer. In the case of the synthesis filter 470, the result is different depending on whether it is converted into an integer for each amplitude bit string (for each sample) or is converted into an integer later for each frame, but both are possible.
As described above, when creating correction sample string candidates using all combinations of possible values of missing bits, if the number of missing bits increases, the number of corrected amplitude bit string (waveform) candidates increases significantly, and the processing amount is reduced. May become realistic. The processing of the missing correction unit 430 to prevent such a problem and the functional configuration thereof will be described below.

  FIG. 24 shows an example of the processing procedure, and FIG. 25 shows an example of the functional configuration. First, the provisional waveform generation unit 431 reproduces a provisional prediction error waveform (provisional amplitude bit string) in a frame using only the confirmed bits input from the rearrangement unit 220 (S1). In the reproduction of the provisional prediction error waveform, missing bits are fixed to 0, for example, or set to an intermediate value that can be taken by missing bits. For example, if the lower 4 bits are missing, any of the levels from 0 to 15 is a correct value, but is set to 8 or 7.

  Next, the prediction coefficient LPC of the spectral envelope decoded from the received auxiliary information is set in the synthesis filter 502, and this provisional prediction error waveform is passed through the synthesis filter 502 to linearly predict and synthesize the original input signal waveform to the encoder 10. (S2). The spectrum envelope calculation unit 432 calculates the spectrum envelope of the combined waveform (S3). The error calculation unit 433 compares the calculated spectrum envelope with the spectrum envelope of the original sound (original input signal) received as auxiliary information, that is, the spectrum envelope decoded by the auxiliary information decoding unit 450, and an error is allowed. If it is within the range, the provisional prediction error waveform is output to the synthesis filter 470 as a corrected prediction error waveform (corrected amplitude bit string) (S4).

  In step S4, when the spectrum envelope shape of the provisional prediction error waveform is greatly different from the spectrum envelope shape decoded from the auxiliary information, that is, when the provisional prediction error is incomplete, first, the inverse characteristic of the calculated spectrum envelope is given to the provisional prediction error waveform. (S5). Specifically, the spectrum is flattened by passing the provisional prediction error waveform through an all-pole linear prediction inverse filter (all zero type) 434 in which parameters representing the spectrum envelope obtained in step S3 are set. In this case as well, the prediction coefficient obtained by expanding the bandwidth of the spectral envelope characteristic, that is, the smoothed coefficient, may be used instead of completely flattening.

Next, a characteristic of a decoded spectrum envelope is given to the flattened signal (S6). That is, a corrected prediction error waveform of the provisional prediction error waveform is created through the output of the inverse filter 434 through the all-pole synthesis filter 435 in which the parameter LPC representing the spectral envelope decoded from the auxiliary information is set. The resulting corrected prediction error waveform is close to the original prediction error waveform (signal).
Similar to the case shown in FIG. 22, this corrected prediction error waveform may be inconsistent with the already known amplitude bit, and is corrected to the correct value by the correction unit 436 (S7).

  The modified prediction error waveform is used as the provisional prediction error waveform in step S1, and the processes in and after step S2 are repeated. As shown by the broken lines in FIGS. 24 and 25, after step S4, the provisional prediction error waveform is synthesized using the decoded spectral envelope parameter (through synthesis filter 435 ") (S5 '), and the synthesized waveform is obtained. On the other hand, an inverse characteristic of the calculated spectral envelope may be given (through the inverse filter 434 ′) (S6 ′), and when a coefficient with an expanded bandwidth is not used, a waveform synthesized to obtain a prediction coefficient, that is, The output waveform of the synthesis filter 502 may be supplied to the inverse filter 434.

Next, another example of missing information correction processing using the decoded spectrum envelope will be described with reference to FIG.
Steps S1 to S4 and S7 are the same processes as steps S1 to S4 and S7 in the process shown in FIG. In this example, after step S4, the filter coefficient of one synthesis filter unit 435 ′ that combines the inverse filter 434 of the spectrum envelope calculated in step S3 and the synthesis filter 435 of the decoded spectrum envelope is calculated (S5), and step S6. Then, the corrected prediction error waveform is synthesized through the provisional prediction error waveform to the synthesis filter unit 435 ′.

The functional configuration of the process shown in FIG. 26 is reversed from the decoded spectrum envelope parameter LPC from the auxiliary information decoding unit 450 and the estimated spectrum envelope parameter α from the spectrum envelope calculation unit 432 as indicated by a broken line in FIG. The filter characteristic of one synthesis filter unit 435 ′ combining the filter 434 and the synthesis filter 438 is calculated by the synthesis spectrum envelope calculation unit 437, and the filter coefficient is set in the synthesis filter 438. The error waveform is passed.
For example, as described with reference to FIG. 10, the calculation of the filter coefficient of the synthesis filter 438 is performed by converting the linear prediction coefficient of the provisional prediction error waveform into the linear prediction cepstrum coefficient Ca by the coefficient conversion unit 437a and Is converted into a linear prediction cepstrum coefficient Cb by a coefficient conversion unit 437b, and these coefficients Ca and Cb are subtracted by a subtraction unit 437c to obtain Cb-Ca. The linear prediction coefficient may be used as the filter coefficient of the synthesis filter unit 435 ′.

In order to prevent the prediction error waveform from diverging in the iterative processing in the flowchart shown in FIG. 22, in the example shown in FIGS. 22, 24, and 26, the linear prediction coefficient α _k of the inverse filter 434 and FIGS. In the example shown in FIG. 4, there is a method in which the linear prediction coefficients β _k and β _k ′ of the synthesis filters 435 and 438 are multiplied by a constant γ of less than 1 (k is the parameter order). In the case shown in FIG. 10, it is only necessary to apply a constant less than 1 to all the linear prediction cepstrum coefficients. Even in the case of iterative processing in FIGS. 22, 24 and 26, it is also possible to set γ to a value close to 1 at the beginning of the iteration, and to reduce the estimation error by decreasing the value of γ as it converges. is there.

  In FIG. 22, Step S3 is omitted in Steps S3, 24 and 26, Steps S1, S2, S4 and S5 are performed only once in FIG. 22, and Steps S1, S2, S3, S5 in FIG. S6 and S7 may be repeated only once or a predetermined number of times to output a corrected prediction error waveform (amplitude bit string). Further, as shown in the figure, when the number of repetitions exceeds the specified number of times in step S3 in FIG. 22, step S4 in FIG. 24, the final corrected prediction error waveform may be output. The processing shown in FIGS. 22, 24, and 26 may be performed after conversion to the frequency domain. In this case, for example, the inverse filter process is a normalization process.

  The prediction error signal Spe often has a small amplitude. For this reason, when the prediction error signal Spe is expressed by a binary number of polarity and absolute value, for example, as described with reference to FIGS. 15A and 15B, the upper digits of each prediction error signal Spe (amplitude bit string) are all in the entire frame. It is often "0". Accordingly, the number of digits representing the maximum absolute value within one frame is obtained as the number of significant digits, that is, the maximum number of digits including “1” is the number of significant digits Fe, and the number of significant digits in the encoder 10 in FIG. The number detection unit 163 detects this significant digit Fe and the auxiliary information encoding unit 352 encodes it and outputs it together with the spectrum envelope parameter LPC. The rearrangement unit 160 converts the bits and polarity within the range 41 of the significant digit Fe. Only the code is converted into a coordinate bit string and output to the transmission recording unit division unit 310.

  In the decoder 20, the auxiliary information decoding unit 450 decodes the reproduction prediction error waveform composed of the amplitude bit string from the rearrangement unit 220 or the correction (correction) amplitude bit string from the missing correction unit 430 from the input packet. The digit alignment unit 460 performs digit alignment based on the number of digits Fe. When the reproduction prediction error waveform (amplitude bit string) input to the digit aligning unit 460 becomes, for example, the amplitude bit string of the number of digits Fe on the right side of FIG. 15B, for each amplitude bit string (2's complement expression), As shown on the left side of FIG. 15B, when the polarity code is positive ("0"), the bit width of the original amplitude bit string (bit width of the input prediction error sample Spe of the rearrangement unit 160 of the encoder 10) m "0" is added to the upper order by the number obtained by subtracting the effective number of digits Fe, and "1" is added to the upper order by the same number m-Fe for those with a negative polarity sign ("1").

  The reproduction prediction error waveform aligned in this way is supplied to the synthesis filter 470. By adding the number of significant digits to the auxiliary information in this way, the encoding efficiency can be improved. It is to be noted that the rearrangement unit 160 of the encoder 10 does not convert the prediction error signal (amplitude bit string) into a polarity code and an absolute binary value, and can convert the number of significant digits into a two's complement expressed amplitude bit string. Fe may be output as auxiliary information. Also in this case, as shown on the left side of FIG. 15B, the number of lower digits is the effective digit number Fe from the range in which only “0” or “1” is the same upper digit number in any amplitude bit string in the frame. Yes, the least significant bit of the number of significant digits Fe may be used as the polarity code.

  When the effective digit number Fe is used, the spectrum envelope calculation unit 432, the inverse filter 434, the synthesis filter 435, the synthesis filter 502, and the synthesis filter 435 ′ in the missing correction unit 430 (FIG. 25) of the decoder 20 perform analysis thereof. If the playback prediction error signal of the previous frame is required for processing or filtering, it is necessary to process the effective frame number Fe of the previous frame and the effective frame number of the current frame together. is there. For example, when the number of significant digits of the current frame is M bits (digits) larger than the number of significant digits of the previous frame, for example, the amplitude bit string of the previous frame is shifted to the lower M bits and the amplitude of the past frame The amplitude value of the bit string is reduced, and the number of significant digits of the previous past frame is made equal to the number of significant digits of the current frame. When the number of significant digits of the current frame is smaller than the number of significant digits of the previous frame by N bits (digits), for example, the amplitude bit string of the previous frame is shifted to N bits higher as a floating point, for example. Enlarge the amplitude value of the amplitude bit string of the previous frame to make the number of significant digits of the previous frame equal to the current significant digit, and then perform the processing by the above filters, or ignore the previous frame Process.

When aligning the number of significant digits, immediately before steps S2 and S4 in FIG. 22, immediately before steps S2, S3, S5 and S6 in FIG. 24, and immediately before steps S2, S3 and S6 in FIG. As indicated by the broken line, the effective number of digits may be corrected so that the effective number of digits is aligned with respect to the amplitude bit string of the previous frame required. In the missing correction unit 430 of FIGS. 23 and 25, in each unit that needs to align the number of significant digits, a process for aligning the number of significant digits may be performed on the amplitude bit string of the previous frame. As shown in the decoder 20 in FIG. 21, the rearrangement unit 220 may perform digit alignment on the amplitude bit string by the digit alignment unit 460. In this case, it is necessary to correct the number of significant digits described above. Disappear.
By the way, conventionally, high compression encoding of acoustic signals generally takes into account auditory characteristics from the viewpoint of increasing compression efficiency and minimizes auditory distortion. This is coding using auditory optimization, which minimizes auditory quantization distortion by using frequency masking based on a given number of bits. In the norm of waveform distortion minimization, in order to minimize the energy of quantization distortion regardless of hearing, quantization distortion is allocated regardless of the magnitude of the spectrum of the original sound.

In auditory optimization, the quantized distortion around the large spectral component of the original sound is masked by the original sound component and cannot be heard. However, the quantization distortion is reduced.
For example, when auditory characteristics are considered in lossy encoding in a combination of lossy encoding with high compression ratio and lossless encoding as in the embodiment of FIG. In 11, the waveform of the output of the inverse quantization unit 130 is particularly distorted in a portion of a relatively large spectral component. For this reason, the error signal between the locally decoded signal and the input sound signal of the original sound has a relatively large variation in amplitude, and therefore the encoding efficiency is sufficiently large even when lossless compression encoding is performed by performing the above-described rearrangement. Can not do it. An embodiment in which this point is improved will be described below with reference to FIG.

In the encoding device 10, the sample sequence of the acoustic signal input to the input terminal 100 is divided into frame units, for example, every 1024 samples by the frame dividing unit 110, and the acoustic signal of each frame unit takes into account auditory characteristics. Lossy compression encoding is performed by the auditory optimization encoding unit 13. This auditory optimizing encoding is compression encoding so that auditory distortion is minimized. For example, MPEG encoding can be used. The auditory optimization encoding unit 13 outputs an auditory optimization code Ina that is an irreversible code.
This auditory optimizing code Ina is locally decoded by the local decoding unit 14 to generate a local decoded signal. The local decoded signal of the code Ina can be generated by the auditory optimizing coding unit 13 by the analysis / synthesis coding method. This local decoded signal is deformed in the deforming unit 15 so that an error from the output acoustic signal of the frame dividing unit 110 becomes small. That is, the error between the local decoded signal (modified local signal) transformed by the transformation unit 15 and the acoustic signal output from the frame division unit 110 is calculated by the error calculation unit 16, and the energy of the error signal is desirably minimized. As described above, the locally decoded signal is deformed by the deforming unit 15.

( )^T は行列の転置を表わす。
誤差信号のエネルギーｄは下記となる。
ｄ＝(Ｘ−ＹＡ)^T(Ｘ−ＹＡ) (2)
ｄを最小とする変形パラメータＡは下記となる。 This deformation is performed by multiplying the local decoded signal by the deformation parameter generated by the deformation parameter generating unit 17 or by weighted addition of a plurality of samples to the local decoded signal. For example, the deformation parameter is generated as follows.
A p-order deformation parameter A (1 × p), an input acoustic signal X (1 × n) composed of n samples, and a local decoded signal matrix Y (p × n) are set as follows.
A = (a ₀ , a ₁ , ..., a _(p-1) ) ^T
X = (x ₀ , x ₁ ,..., X _(n−1) ) ^T

() ^T represents matrix transposition.
The energy d of the error signal is as follows.
d = (X−YA) ^T (X−YA) (2)
The deformation parameter A that minimizes d is as follows.

A = (Y ^T Y) ⁻¹ Y ^T X (3)
(Y ^T Y) is the autocorrelation matrix, Y ^T X can be approximated as the cross-correlation vector, deformation parameters a ₀ when the p = 1 is as normalized cross-correlation coefficient between X and Y in the energy of Y Become.
Further, the correlation coefficient b between U (= X−Y) and Y is obtained,
Z = XY-bY (4)
Z may be used as an error signal. That is, Z = X− (1 + b) Y≡X−a ₀ Y.

That is, in the deformation parameter generation unit 17, for example, as shown in FIG. 28A, the cross-correlation vector between the transposed matrix of the input acoustic signal X and the local decoded signal Y is calculated by the multiplication unit 171, and the autocorrelation matrix Y ^T is calculated by the multiplication unit 172. Y is calculated, and the division unit 173 divides the calculation result of the multiplication unit 171 by the calculation result of the multiplication unit 172 to generate the deformation parameter A. Alternatively, the multiplication unit 171 calculates the X and Y cross-correlation coefficients, the multiplication unit 172 calculates the Y energy, and the division unit 173 divides the obtained cross-correlation coefficient by the Y energy. The deformation parameter a ₀ may be obtained by performing the process.

Alternatively, as shown in FIG. 28B, X−Y = U is calculated by the subtracting unit 174, the cross-correlation coefficient b between U and Y is calculated by the multiplying unit 175, and 1 is added to b by the adding unit 176. The parameter a ₀ may be used.
The deformation parameter A, a ₀ or b is encoded and output from the deformation parameter generation unit 17 as a deformation parameter code Inm. Thus, the deformation parameter includes a cross-correlation component between the acoustic signal X and the local decoded signal Y, and the deformation unit 15 multiplies the local decoding signal Y by the deformation parameter or converts the local decoded signal Y by the deformation parameter A. Performs weighted addition or convolution.

If the energy of the error signal between the acoustic signal and the local decoded signal is minimized, the error signal is uncorrelated with the acoustic signal or the local decoded signal, but if the energy of the error signal is not minimized, an error occurs. Correlation occurs between the signal and the acoustic signal or the locally decoded signal. The correlation component is obtained, and the local decoded signal Y is transformed by equation (3) according to the correlation component, and equation (2) transformation parameter A is determined so as to minimize the energy of the error signal.
The generation of the deformation parameter A may be sequentially corrected every frame so that an appropriate value is initially given and the amplitude or energy of the error signal U becomes small. In this case, it is not necessary to output the deformation parameter code Inm.

The error signal from the error calculator 16 is losslessly encoded by the lossless encoder 18 and output as a lossless code Pne. The lossless encoding can be performed by a combination of bit rearrangement and entropy encoding such as Huffman encoding and arithmetic encoding as disclosed in Japanese Patent Laid-Open No. 2001-44847 described with reference to FIGS.
Since the local decoded signal is deformed by the deforming unit 15 so as to reduce the error from the acoustic signal, “0” in the error signal is compared with the error signal between the local decoded signal and the acoustic signal that is not deformed. The number of bits is large, so that lossless encoding can be performed with high efficiency.

The auditory optimization code Ina, the lossless code Ine, and, if necessary, the deformation parameter code Inm are combined by the combining unit 315 and output. As shown in FIG. 28B, the deforming unit 15 multiplies the local decoded signal Y by the modified parameter b to calculate a modified local decoded signal bY, and an error U between the modified local decoded signal bY and U = X−Y. -BY may be calculated by the error calculation unit 16 to obtain an error signal, which may be losslessly encoded. Since this error signal is substantially the same as the error signal Δ = X−a ₀ Y, obtaining an error signal between the acoustic signal and the modified local signal means that both of these methods and an equivalent method thereof are used. means. Obviously, the deformation parameter code Inm may be an encoding of b.

In the decoder 20, the set of codes for each frame input to the input terminal 200 is separated into an auditory optimizing code Ina and a lossless code Ine by a separating unit 440 and a modified parameter code Inm as necessary. The auditory optimization code Ina is irreversibly decoded by the auditory optimization code decoding unit 23 to generate a decoded signal. The decoding method is the same as that of the local decoding unit 14 in the encoder 10. This decoded signal is transformed by the transforming unit 24 by constant multiplication or weighted addition of a plurality of samples. This deformation is also the same processing as the deformation unit 15 in the encoder 10. When the deformation parameter code Inm is input, the deformation parameter generation unit 25 decodes the code Inm to generate the deformation parameter A or a ₀ or b. When the deformation parameter code Inm is not input, the deformation parameter A is obtained by successive correction using the decoded signal Y and the reproduced sound signal X by the same method as the deformation parameter generation unit 14 in the encoding device 10.

The lossless code Ine is losslessly decoded by the lossless decoding unit 21 to reproduce the error signal Δ. The error signal Δ and the modified decoded signal aY are added by the adder 27 to reproduce the acoustic signal. This reproduced sound signal is supplied to the frame synthesizing unit 250 through the switching unit 28, and the reproduced sound signals for each frame are sequentially connected and output. When the missing packet detecting unit 420 detects a packet loss that can be corrected, the information correcting unit 480 performs the correction described above with reference to FIG.
When the reversible code Ine is not input or when an error signal Δ having a quality of a certain level or more cannot be obtained due to a part of the loss, this is detected by the loss detection unit 420, and the switching unit 28 is controlled by this detection output. Thus, the decoded signal is supplied from the auditory optimizing code decoding unit 23 to the frame synthesis unit 250 as a reproduced sound signal. For the final lossless encoding, the compression efficiency is higher when the quantization error energy is smaller, but when only the information of the auditory optimizing code Ina is obtained, it is better to use the quantization result that minimizes the auditory distortion. The quality is high for the bit rate. Even in a decoding device that does not include the lossless decoding unit 26 and the like, the auditory optimizing code Ina can be decoded into a reproduced digital signal.
Eighth Embodiment FIG. 29 shows an eighth embodiment of the present invention, in which parts corresponding to those in FIG. 27 are given the same reference numerals. The encoder 10 is different from the embodiment of FIG. That is, in this second embodiment, an error signal between the local decoded signal Y from the local decoding unit 14 and the acoustic signal X is calculated by the error calculating unit 15A, and the error signal is quantized error energy by the error minimizing encoding unit 15B. Is lossy compression encoded so that the error is minimized, and the error minimization code Inp is output as the lossy code.

  This error minimizing code Inp is locally decoded by the local decoding unit 15C, the local decoded signal and the local decoded signal from the local decoding unit 14 are added by the adding unit 15D, and an error between the added local decoding and the acoustic signal is determined. It is calculated by the error calculation unit 16 and supplied to the lossless encoding unit 18 as an error signal. As described above, since the acoustic signal is encoded by the two stages of the auditory optimizing encoding unit 13 and the error minimizing encoding unit 15B, the local decoded signal, that is, the added local decoded signal from the adding unit 15D is converted into the acoustic signal accordingly. Accordingly, the number of “0” bits in the error signal from the error calculation unit 16 increases. In the combining unit 315, the auditory optimizing code Ina, the error minimizing code Inp, and the lossless code Ine are combined and output.

In the decoder 20, the error minimizing code Inp is irreversibly decoded by the error minimizing code decoding unit 24A, and the decoded signal and the decoded signal from the auditory optimizing code decoding unit 23 are added by the adding unit 24B. The added decoded signal is supplied to the adding unit 27 through the switching unit 28 and added to the reproduction error signal from the lossless decoding unit 26. When the reversible code Ine is not input or when sufficient information for reproducing the error signal cannot be obtained, the switching unit 28 reproduces the decoded signal of the auditory optimizing code Ina instead of the added decoded signal from the adding unit 24B. The sound signal is supplied to the frame synthesis unit 250 as an acoustic signal. The error minimizing code decoding unit 45 and the adding unit 24B constitute a deforming unit 24.
Ninth Embodiment FIG. 30 shows a ninth embodiment of the present invention. The ninth embodiment differs from the seventh embodiment of FIG. 27 in that the configuration of the deforming unit 15 in the encoding device 10 is different and that an error minimizing encoding unit 19 is provided. Has also changed.

  In the encoder 10, the acoustic signal for each frame is irreversibly compressed and encoded by the error minimizing encoder 19 so as to minimize the quantization error energy, and the error minimizing code Inp is output as the irreversible code. This error minimizing code Inp is locally decoded by the local decoding unit 15E, and the local decoded signal and the local decoded signal from the local decoding unit 14 are weight averaged with emphasis on the former by the load averaging unit 15F. The error between the weighted average local decoded signal and the acoustic signal is calculated by the error calculation unit 16, and the error signal is supplied to the lossless encoding unit 18.

  The weight in the weighted average unit 15F is a fixed value that is statistically obtained in advance so that the error from the error calculation unit 16 becomes small. For example, about 0.8 to 0.9 for the local decoded signal of the error minimization code Inp, The value is about 0.1 to 0.2 for the local decoded signal of the auditory optimizing code. Alternatively, the modification parameter generation unit 17 may input both local decoded signals and the acoustic signal, solve the simultaneous equations so that the error signal from the error calculation unit 16 is minimized, and determine both weights. In this case, both weights are encoded and a weight code Inw is output. The local decoding unit 15E and the load averaging unit 15F constitute a deformation unit 15.

  The combining unit 315 combines and outputs the auditory optimizing code Ina, the error minimizing code Inp, and the lossless code Ine, and also combines the weight code Inw when the deformation parameter generating unit 17 is used. The auditory optimal encoding unit 13 and the error minimizing encoding unit 19 may use parameters such as spectrum envelope and power together. In such a case, as shown by a broken line, a common encoding unit 13A that encodes parameters common to both is provided in one encoding unit, in FIG. 30, in the auditory optimizing encoding unit 13, and as the common code Inc. In addition to the output, the extracted commonly used parameters are supplied to the other encoding unit, in this example, the error minimizing encoding unit 19 as necessary. Furthermore, the coding unit 13B in the auditory optimization coding unit 13 encodes the waveform information of the input acoustic signal by vector quantization in consideration of auditory characteristics, etc., and outputs the code Inar for error minimization coding. Similarly, the encoding unit 19A in the unit 19 may encode the waveform information of the acoustic signal by vector quantization so as to minimize the quantization error energy and output the code Inpr.

  In the decoder 20, the error minimizing code Inp separated by the separating unit 440 is irreversibly decoded by the error minimizing code decoding unit 24A, and the decoded signal is decoded from the auditory optimizing code decoding unit 23 by the weighted average unit 24C. The signal and the load addition biased to the former are performed, and the load addition decoded signal is added to the reproduction error signal from the lossless decoding unit 21 by the adding unit 27 and supplied to the frame synthesizing unit 250 through the switching unit 28. The load in the load average unit 24C is the same as the load in the load average unit 15F of the encoding device 10. When the weight (load) is determined by the deformation parameter generation unit 17 in the encoding device 10, the weight code Inw separated by the separation unit 440 is decoded by the weight decoding unit 29, and the decoded weight is the weighted average. To the unit 24C.

The decoded signal of the auditory optimizing code Ina and the decoded signal of the error minimizing code Inp are also supplied to another weight averaging unit 26, where a weighted average biased to the former (Ina decoded signal) is performed. For example, the weight in the weighted average is fixedly about 0.8 to 0.9 for the decoded signal of the code Ina and about 0.2 to 0.9 for the decoded signal of the code Inp. Or you may give the weight decoded by the weight decoding part 29 with respect to the weight average part 26 by the reverse relationship with the weighting by the weight average part 24C.
When the lossless code Ine is not input or when the lossless decoding unit 21 cannot obtain sufficient information for reproducing the error signal, the averaged decoded signal from the weighted average unit 26 is frame-synthesized as a reproduced sound signal through the switching unit 28. Supplied to section 250. When the common code Inw is separated by the separation unit 440, the common code Inw is decoded by the common decoding unit 22, and these common decoded signals are supplied to the auditory optimizing code decoding unit 23 and the error minimizing code decoding unit 24A. . These decoding units 23 and 24A are supplied with codes Inar and Inpr, respectively, so that a desired decoded signal can be obtained. The error minimizing code decoding unit 24A and the weighted average unit 24C constitute a deformation unit 24.
Tenth Embodiment FIG. 31 shows the tenth embodiment corresponding to FIG. 30 with the same reference numerals. In the tenth embodiment, the encoder 10 directly supplies the local decoded signal of the error minimizing code Inp from the local decoding unit 15E to the error calculating unit 16. Therefore, local decoding is not performed for the auditory optimizing code Ina, and the local decoding result for the error minimizing code Inp is used for error calculation in the error calculating unit 16. Also in the tenth embodiment, as in the ninth embodiment, a common parameter is extracted on the one hand in the auditory optimizing encoding unit 13 and the error minimizing encoding unit 19 and used for encoding both of them and the common code Inw and the partial codes Inar and Inpr may be output.

In the decoder 20, the decoded signal from the error minimizing code decoding unit 24 </ b> A is directly supplied to the adding unit 27 and added to the reproduction error signal from the lossless decoding unit 21. The switching unit 28 switches the decoded signal of the auditory sensation optimization code decoding unit 23 and the addition signal of the addition unit 27 and supplies them to the frame synthesis unit 250 as a reproduced sound signal. The operation when the common code Inc is separated by the separation unit 440 is the same as that described with reference to FIG.
Eleventh Embodiment FIG. 32 shows an eleventh embodiment of the invention corresponding to FIG. 30 with the same reference numerals. In the encoder 10 of the eleventh embodiment, the auditory optimizing encoder 13 in the third embodiment of FIG. 30 is connected to the output side of the error calculator 16, and the error minimizing code Inp from the local decoder 15E. The error between the local decoded signal and the acoustic signal is calculated by the error calculation unit 16, and the error signal is irreversibly compressed and encoded by the audibility optimization encoding unit 13 so as to minimize the audible distortion. Is output as an auditory optimizing code Ina. In the combining unit 315, the error minimizing code Inp, the auditory optimizing code Ina, and the lossless code Ine are combined and output.

  The auditory optimizing code Ina separated by the decoder 20 is irreversibly decoded by the auditory optimizing code decoding unit 23, and the decoded signal and the decoded signal of the error minimizing code Inp by the error minimizing code decoding unit 24A are obtained. The sum is added by the adder 24B. The decoding signal of the error minimization code Inp and the lossless decoding signal of the lossless code Ine, that is, the reproduction error signal are added by the addition unit 27 and supplied to the frame synthesis unit 250 through the switching unit 28 as a reproduction acoustic signal. When the error signal is not reproduced by the lossless decoding unit 21, the addition decoded signal from the adding unit 24B is supplied to the frame synthesizing unit 250 through the switching unit 28 as a reproduced acoustic signal.

  Also in this case, a quantization sequence that minimizes quantization error energy is used for lossless compression encoding by the lossless encoding unit 18, and when only information of irreversible compression encoding is obtained, a quantum that minimizes auditory distortion is obtained. The point of using the conversion result is the same as in the eighth to tenth embodiments. However, in this case, after the error minimizing encoding unit 19 performs error minimizing encoding, the perceptual distortion minimizing quantization is further performed on the local decoded signal by the auditory optimizing encoding unit 13, that is, multistage. Therefore, the number of encoded bits as a whole is increased, but the perceptual improvement is made accordingly. Even when the distortion energy is optimized, the first stage local decoded signal, that is, the error from the local decoding unit 15E is minimized. It may be more efficient than the locally decoded signal of the coded code Inp. That is, it is better to use the more efficient one for optimizing the strain energy.

  A method in this case will be described with reference to a broken line configuration shown in FIG. The code Ina from the auditory optimizing coding unit 13 is locally decoded by the local decoding unit 14, and the local decoded signal is added by the adding unit 31 to the local decoded signal of the code Inp from the local decoding unit 15E. An error between the signal and the acoustic signal is calculated by the error calculation unit 32, the magnitude of the error signal and the error signal from the error calculation unit 16 is compared by the size comparison unit 33, and the smaller one is selected by the selection unit 34. And supplied to the lossless encoding unit 18. At this time, a selection code Ins indicating which error signal has been selected is output.

In the decoder 20, the selection unit 41 is controlled by the selection code Ins separated by the separation unit 440, and when the error signal from the error calculation unit 16 is selected by the encoding device 10, the error from the error minimizing code decoding unit 24A. When the decoded signal is selected and supplied to the adding unit 27 and the encoder 10 selects the error signal from the error calculating unit 32, the added decoded signal from the adding unit 24B is selected and supplied to the adding unit 27. The
A specific example of the lossless encoding unit 18 in FIGS. This configuration is almost the same as the configuration of the rearrangement unit 160, transmission recording unit division unit 310, lossless encoding unit 150, and auxiliary information generation unit 350 in FIG. The error signal from the error signal 16 is supplied to the rearrangement unit 160 and the auxiliary information generation unit 350. The number of significant digits detection unit 353 in the auxiliary information generation unit 350 detects the number of digits representing the maximum value of the absolute value of the error signal for each frame as the number of significant digits Fe. Further, only the portion within the number of significant digits of the error signal is arranged in the same bit position in each sample (amplitude bit string) by the rearrangement unit 160 as described with reference to FIGS. 15A and 15B, etc. It is assumed to be a bit sequence.

The equidistant bit string from the rearrangement unit 160 is divided into transmission unit or recording unit data by the transmission / recording unit dividing unit 310. These divided transmission recording unit data are losslessly encoded by the lossless encoding unit 150 as necessary and output as an error code Ine.
On the other hand, the significant digit number Fe detected by the significant digit number detecting unit 353 is given to the auxiliary information encoding unit 352. In this example, the error signal sample sequence is further subjected to, for example, linear prediction analysis in the spectral envelope calculation unit 351 for each frame, and a parameter LPC expressing the spectral envelope is obtained as a linear prediction coefficient. Further, the power calculation unit 354 calculates the average power of the error signal for each frame. Alternatively, an error signal is input to the inverse filter 355 configured based on the linear prediction coefficient obtained by the spectrum envelope calculation unit 351 to flatten the spectrum envelope, and the average power of the flattened signal is obtained by the power calculation unit 356. It is done. The significant digit number Fe, the linear prediction coefficient LPC, and the average power are encoded by the auxiliary information encoding unit 352 with low bits of about 30 to 50 bits, for example, to be auxiliary information Inx. The auxiliary information Inx in which the number of significant digits, the spectrum envelope parameter, and the average power are encoded is supplied to the combining unit 315 (FIGS. 27 and 29 to 32), and transmission record unit data including a representative packet of each frame, for example, a polarity code. Is added to the stored packet or output as an independent packet.

  A specific example of the lossless decoding unit 21 of the decoder 20 in FIGS. 27 to 32 is shown in FIG. 34 together with the information correction unit 480. The auxiliary information Inx and the error code Ine are separated from the input packet Pe by the separation unit 440, the error code Ine is supplied to the lossless decoding unit 210, and the auxiliary information Inx is supplied to the auxiliary information decoding unit 450. The auxiliary information decoding unit 450 decodes the effective digit number Fe of the frame, the parameter LPC representing the spectral envelope and the average power, and supplies the effective digit number Fe to the digitizing unit 460, and the spectral envelope parameter and the average power are deleted. Supply to 430. Since the operation of each part has already been described in detail in the embodiment of FIG. 14, the description thereof is omitted here.

  27 to 32, the lossless encoding unit 18 of the encoder 10 may be configured to further perform a prediction process on the error signal. An example of the configuration is shown in FIG. 35A. The error signal is given to the spectrum envelope calculation unit 351, and a linear prediction coefficient LPC representing the spectrum envelope is obtained. On the other hand, the error signal is given to the prediction error generation unit 370 to generate the prediction error signal Spe. For example, a plurality of error signal samples immediately before from the error calculation unit 16 are supplied from the register 371 to the linear prediction unit 372, and these samples are multiplied by the linear prediction coefficient LPC representing the spectrum envelope from the spectrum envelope calculation unit 371, These multiplication results are added to obtain a linear prediction value. This linear prediction value is converted to an integer value by the integerization unit 373, and a difference between the integer prediction value and the current sample of the error signal from the error calculation unit 16 is obtained by the difference circuit 374 to obtain a prediction error signal Spe. . The prediction error signal Spe is input to the rearrangement unit 160.

FIG. 35B shows a configuration example of the lossless decoding unit 21 in the decoder 20 when the configuration of FIG. 35A is applied to the lossless encoding unit 18 of the encoder 10. This configuration is the same as that of the rearrangement unit 220, missing correction unit 430, auxiliary information decoding unit 450, and synthesis filter 470 of the decoder 20 in FIG. The reproduced prediction error signal Spe from the information correction unit 480 is processed by the synthesis filter 470 in reverse to the processing of the prediction error generation unit 370 of the lossless encoding unit 18 in FIG. 35A. That is, a predetermined number of immediately preceding decoded samples are input from the register 471 to the linear prediction unit 472, and these samples are multiplied by weights based on the linear prediction coefficient LPC decoded by the auxiliary information decoding unit 450, respectively. Then, the sum of the multiplication results is obtained as the predicted value of the current decoded sample. This prediction value is converted to an integer value by the integerizing unit 473, and the sum of the integerized prediction value and the current prediction error signal from the information correcting unit 480 is calculated by the adding unit 474 to obtain a filter of the synthesis filter 470. It is an output, that is, a reproduction error signal.
Twelfth Embodiment Each of the above-described embodiments is a case of an encoder and a decoder for an input digital signal of one channel. Examples of applying these embodiments to a multi-channel signal are shown below. Several compression encodings for multi-channel signals have been developed, such as AC-3 (Dolby's Audio Coding) and AAC (Dolby's Advanced Audio Coding). Yes, but the waveform is quite different from the original sound. When a multi-channel signal is transmitted as it is as PCM (Pulse Code Modulation), the original sound is completely reproduced, but a large amount of information is required. In addition, sound cannot be reproduced from only a part of the encoded sequence, and therefore packet loss is greatly affected when packet transmission is performed on a PCM signal.

It is conceivable to increase the coding efficiency by mixing multi-channel signals and coding with a reduced number of channels. However, in this case, the original multichannel signal cannot be reproduced correctly.
In the twelfth embodiment, a plurality of digital signals are mixed and encoded as a smaller number of digital signals, the encoding efficiency is improved, and the original digital signals are faithfully reproduced as necessary. It is also possible to provide a multi-channel signal encoding method, decoding method, encoder, and decoder that enable selection of a wide range of bit rates and selection of a wide range of channels.

FIG. 36 shows a twelfth embodiment of an encoder and a decoder according to the present invention. In the encoder 10, when M is an integer of 2 or more, digital signals of M channels are input to the frame dividing unit 110 from terminals 100 _{1 to} 100 _M , and are divided into frame units, for example, every 1024 samples. Each of the divided digital signals is mixed by the channel mixing unit 30 into N-channel digital signals smaller than M. N is an integer of 1 or more, and N <M. The mixed N-channel signal is lossy or losslessly compressed and encoded by the encoding unit 120, and is output as the main code Im. This encoding is desirably high compression encoding. The main code Im is decoded by the local encoding unit 130 to generate a local decoded signal. The channel expansion unit 40 increases the number of channels of this local decoded signal to the original M channel local decoded signal. The local decoded signal of the code Im can be generated by the analysis unit coding method in the encoding unit 120.

The channel mixing may be, for example, the average of the first four channels of the eight channel signals and the average of the subsequent four channels as the right channel signal. Alternatively, the average of two-channel stereo signals may be a one-channel monaural signal.
In the channel expansion, the number of signals corresponding to the number of channels corresponding to the channel mixing is reversed. Here, the local code signal with the expanded number of channels loses information with respect to the input digital signal due to the decrease in the number of channels due to the channel mixing. Using the lost signal as an error signal, the M channel digital signal branched from the frame dividing unit 110 and the M channel local decoded signal are input to the error calculating unit 140 and calculated. The error signal is supplied to the array conversion unit 50, where the bit string is rearranged and compressed and encoded, and is output together with the auxiliary information code Inx as the error code Ine. Although the specific configuration and processing of the array changing unit 50 will be described later, it includes at least a rearranging unit 160 that performs conversion to the equidistant bit string described in FIGS. 4A, 15A, and 15B. The error code Ine from the array conversion unit 50, the auxiliary information Inx, and the main code Im from the encoding unit 120 are combined and output by the combining unit 315, and when the packet Pe is transmitted, the header 31 described in FIG. 4B. It is sent with.

In the decoder 20, the input packet Pe is separated into the main code Im, the error code Ine, and the auxiliary information Inx in this case by the separation unit 440. The main code Im is decoded by the decoding unit 60 by a decoding method corresponding to the encoding method of the encoding unit 120 in the encoder 10 to generate a decoded signal. The decoded signal is expanded by the channel expanding unit 80 from an N-channel decoded signal to an M-channel decoded signal.
The separated error code Ine is processed by the array inverse transform unit 70 to reproduce an M-channel error signal. Although the specific configuration and processing of the array inverse transform unit 70 will be described later, it includes at least the rearrangement unit 220 described in FIGS. 4A, 15A, and 15B, and reproduces an error signal composed of an amplitude bit string. Also, the auxiliary information Inx is decoded by the information correction unit 480, and when the loss detection unit 420 detects the packet loss, the information correction unit 480 corrects the reproduction error signal based on the auxiliary information.

The reproduced M channel error signal and the channel expanded M channel decoded signal are added by the adding unit 240 for each corresponding channel and supplied to the frame synthesizing unit 250 as an M channel reproduced digital signal. A digital signal is output by concatenating frames for each channel.
According to this configuration, if the main code Im, the error code Ine, and the auxiliary information Inx are input to the decoder 20 with a missing amount small enough to obtain a reproduced decoded signal and a decoded error signal, respectively, the original M channel digital signal is converted. Can be played faithfully. The encoding efficiency can be changed by selecting the number of channels to be reduced in the channel mixing unit 30 as necessary. Further, when a packet is not input or sufficient information for reproducing the error signal cannot be obtained, the M channel decoded signal from the channel expanding unit 80 is supplied to the frame synthesizing unit 250 as a reproduced digital signal. Can be obtained.

  In the embodiment of FIG. 36, channel mixing may be performed in multiple stages for decoding. An example will be described with reference to FIG. In FIG. 37, parts corresponding to those in FIG. In this example, the encoder 10 performs channel mixing in two stages, that is, a channel mixing unit 30 and a channel mixing unit 41. For example, an original digital signal of M = 8 channel stereo is channel-mixed with a stereo digital signal of N = 2 channel or N = 4 channel, and the stereo digital signal is further channel-mixed with a monaural digital signal of L = 1 channel. The channel mixing signal from the channel mixing unit 41 is encoded by the encoding unit 120, and the main code Im is output.

This main code Im is decoded by the local decoding unit 130, and the local decoded signal is channel-expanded by the channel expanding unit 42 to change from the L channel to the N channel. An error signal between the N-channel local decoded signal and the N-channel digital signal from the channel mixing unit 30 is calculated by the error calculation unit 43, and the error signal is irreversibly or reversibly encoded by the encoding unit 44, preferably high compression. Is output as a secondary code Ie.
The secondary code Ie is decoded by the local decoding unit 45 to create an N-channel local decoded signal, and this and the N-channel local decoded signal from the channel expanding unit 42 are added by the adding unit 46, The channel expansion unit 40 expands the local decoded signal to obtain an M channel addition local decoded signal. An error signal between the M channel addition local decoded signal and the M channel digital signal from the frame division unit 110 is calculated by the error calculation unit 140, and this error signal is supplied to the array conversion unit 50. The error code Ine, auxiliary information Inx, main code Im, and subcode Ie from the array conversion unit 50 are combined by a combining unit 315, packetized, and output.

  In the decoder 20, the main code Im and the sub code Ie separated by the separation unit 440 are decoded by the decoding units 60 and 61, respectively, and the error code Ine is rearranged by the array inverse conversion unit 70 to reproduce an error signal. The L channel decoded signal from the decoding unit 60 is expanded to an N channel decoded signal by the channel expanding unit 80, and the N channel decoded signal and the N channel decoded signal from the decoding unit 61 are added by the adding unit 62. The The N-channel addition decoded signal is channel-expanded to an M-channel decoded signal by the channel expanding unit 63, and the M-channel decoded signal and the M-channel error signal of the array inverse transform unit 70 are added by the adding unit 240. The reproduced digital signal is supplied to the frame synthesizing unit 250.

In this case, for example, a bit rate range of about 60 times from 8 kbit / s to 5 Mbit / s can be selected.
36 and 37, when the combining unit 315 generates the packet Pe, it is preferable that the priority of the packet including the main code Im is the highest. A packet including the secondary code Ie has the second highest priority after a packet including the main code Im.
As a specific example of the array conversion unit 50 used in the above-described embodiments of FIGS. 36 and 37, for example, the configuration of the lossless encoding unit 18 described in FIG. 33 can be used as it is. In that case, as a specific example of the array inverse transform unit 70 used in the decoder 20 of each embodiment of FIGS. 36 and 37, for example, the configuration of the lossless decoder 21 described in FIG.

Alternatively, as another specific example of the array conversion unit 50 used in the embodiments of FIGS. 36 and 37, for example, the configuration of the lossless encoding unit 18 described in FIG. 35A may be used as it is. In that case, as a specific configuration of the array inverse transform unit 70 in FIGS. 36 and 37, the configuration of the lossless decoding unit 21 in FIG. 35B can be used as it is.
Thirteenth Embodiment When a multichannel signal is transmitted as it is as PCM, the original sound is completely reproduced, but a large amount of information is required. In addition, sound cannot be reproduced from only a part of the encoded sequence, and therefore packet loss is greatly affected when a PCM signal is transmitted in packets.

In the thirteenth embodiment, an encoding method, a decoding method, an encoder, a decoder, a multi-channel digital signal that are not easily affected by information loss and that can select a degree of information compression within a relatively wide range. I will provide a.
The encoder 10 and the decoder 20 of the thirteenth embodiment will be described with reference to FIG.
In the encoder 10, M digital signals (M is an integer of 2 or more) are input to the frame dividing unit 110 through input terminals 100 _{1 to} 100 _M , and are sequentially divided, for example, into frame units each consisting of 1024 digital signals. The These digital signals for each of M frames are subjected to inter-channel orthogonal transform at an inter-channel orthogonal transform unit 190 to output an orthogonal transform signal. That is, samples at the same time of M digital signals are orthogonally transformed. Here, if reversible transform using integer values is used for inter-channel orthogonal transform, reversible encoding can be performed as a whole.

This orthogonal transform signal is input to the rearrangement unit 160. In rearrangement section 160, the same bit position (digit) representing the amplitude in each sample (amplitude bit string) for each component (for example, L + R and LR) of the orthogonal transform signal for each frame is arranged in the time axis direction over the frame. And a coordinate bit string.
The equidistant bit string from the rearrangement unit 160 is divided into transmission unit or recording unit data by the transmission / recording unit dividing unit 310. Each of the divided transmission recording unit data is losslessly encoded by the lossless encoding unit 150 as necessary, and the combining unit 315 can reconstruct the transmission recording unit data divided at the time of decoding as one frame. And a header is attached to the output terminal 170 and output to the output terminal 170 as a packet Pe. Note that the encoding of the lossless encoding unit 150 is the same as the lossless encoding of the prior art. When prioritizing packets Pe, for example, L + R is given first priority in descending order of the energy in each component of the orthogonal transform signal, and the polarity code and the corresponding transmission recording unit data are included on the MSB side. Is given high priority.

  In the decoder 20, the packet 37 from the input terminal 200 is separated into auxiliary information and transmission recording unit data (including a polarity code string) by a separation unit 440 when auxiliary information is included, as will be described later. The If the transmission record unit data is losslessly encoded, it is supplied to the lossless decoding unit 210. After the lossless decoding, the transmission record unit data is supplied to the transmission record unit integration unit 410. Based on the packet number, the transmission record unit integration unit 410 integrates transmission record unit data for one frame from one or more packets, and integrates each component of the orthogonal transform signal. The integrated data is converted by the rearrangement unit 220 from the equivalence bit string to one frame of the amplitude bit string, that is, one component sample string (waveform) of the orthogonal transform signal. Since the operations of the sorting units 160 and 220 have been described in detail with reference to FIGS. 4A, 15A, and 15B, they will not be described here. If no packet loss occurs, the rearrangement unit 220 obtains an orthogonal transform signal that is completely the same as the orthogonal transform signal input to the rearrangement unit 160 of the encoder 10.

In the inter-channel orthogonal inverse transform unit 290, each component of the input reproduced orthogonal transform signal is subjected to a transform reverse to the orthogonal transform in the inter-channel orthogonal transform unit 190 to reproduce the M channel digital signal. The M-channel digital signal reproduced for each frame in this way is output from output terminals 260 _{1 to} 260 _M connected in the order of frames for each channel in frame synthesizing section 250.
In this way, transmission and recording can be performed by orthogonally transforming M channel digital signals between channels, and at that time, by packetizing in units of transmission recording units, the larger the energy, the closer to the MSB. By increasing the priority, when the transmission capacity or recording capacity is small, a reproduction signal having a relatively high quality can be obtained by using only a packet having a high priority. Further, as will be described later, it is possible to reproduce a multi-channel digital signal of better quality by supplementing lost information when a packet is lost.

In the above description, the lossless encoding unit 150 and the lossless decoding unit 210 may be omitted.
14th Embodiment When real number transformation such as DFT or DCT is used for inter-channel orthogonal transform, the reproduced digital signal does not coincide with the original digital signal even if the orthogonal transform signal is inversely transformed between channels. Further, it may be required to output only a part of the orthogonal transform coefficient, that is, only the component having high energy, for example, only L + R as a packet. An embodiment of the invention corresponding to these is the fourteenth embodiment described below.

The fourteenth embodiment will be described with reference to FIG.
39 differs from FIG. 38 in that all or a part of the orthogonal transform signal from the inter-channel orthogonal transform unit 190, that is, a component with high power is losslessly encoded by the encoding unit 120, and the main code Im is Output. The local decoding unit 130 decodes the code from the encoding unit 120 to create a local decoded signal, and the local decoded signal is converted by the inter-channel orthogonal inverse transform unit 180 and converted by the inter-channel orthogonal transform unit 190 to perform a plurality of conversions. The digital local decoded signal of the channel is generated. The local decoded signal of the code Im can be generated by the analysis unit coding method in the encoding unit 120. An error calculation unit 140 calculates an error signal between the corresponding channel between the local decoded signal of each channel and each digital signal from the frame division unit 110. The calculated error signals are processed by the rearrangement unit 160, transmission recording unit division unit 310, and lossless encoding unit 150 in FIG.

  The main code Im from the encoding unit 120 is divided into transmission recording units by the transmission recording unit dividing unit 310, and this is combined with the error code Ine from the array conversion unit 50 by the combining unit 315, and if necessary, the packet Pe Is output as Note that the array conversion unit 50 only needs to include at least the rearrangement unit 160 in FIG. 38, and the processing of the transmission recording unit division unit 310 and the lossless encoding unit 150 in FIG. 38 may not be performed. In this case, the main code Im is also not processed by the transmission recording unit dividing unit 330. When the combining unit 315 uses the main code Im and the error code Ine as a packet, the packet including the main code Im is given the highest priority.

In the decoder 20, the main code Im and the error code Ine are separated by the separation unit 440, the former is losslessly decoded by the decoding unit 60, and the decoded signal is transmitted between the channels of the encoder 10 by the interchannel orthogonal inverse transformation unit 290. The inter-channel orthogonal inverse transform is performed by the same processing as the orthogonal inverse transform unit 180.
On the other hand, the error code Ine is input to the array inverse transform unit 70, and the processing of the lossless decoding unit 210, transmission record unit integration unit 410, rearrangement unit 220, and information correction unit 480 of the decoder 20 in FIG. 38 is performed. The error signal sample sequence is reproduced. However, at least the processing in the rearrangement unit 220 may be performed, and the processing of the lossless decoding unit 210 and the transmission record unit integration unit 410 may not be performed corresponding to the encoder 10 side.

Channels corresponding to these reproduced error signals and the inverse transform signal from the inter-channel orthogonal inverse transform unit 30 are respectively added by an adder 240 to obtain a reproduced digital signal of each channel, which is a frame synthesis unit. Supplied to 250.
The original digital signal can be reproduced according to the fourteenth embodiment even if a part of the orthogonal transform component is omitted, the number of encoded bits is reduced, and efficient encoding is performed. Even if error signal components are not sent or sufficient information necessary for reproduction is not obtained, each signal subjected to inverse transformation processing from the interchannel orthogonal inverse transformation unit 30 is used as a digital reproduction signal to some extent. A quality reproduction signal can be obtained. In the array conversion unit 50 of the encoder 10, various processes can be omitted, but the process in the array reverse conversion unit 70 is also omitted accordingly. The encoding unit 120 is not limited to lossless encoding, and so-called irreversible high compression encoding may be performed. In that case, the decoding unit 60 of the decoder 20 also performs lossy decoding accordingly. As this high compression encoding method, the one described in the section of the prior art can be used.
Fifteenth Embodiment FIG. 40 shows a fifteenth embodiment of the present invention, in which the same reference numerals are assigned to the portions corresponding to FIG. Only the differences from the second embodiment will be described.

  In the encoder 10, a signal obtained by orthogonally transforming the M channel digital signal by the interchannel orthogonal transform unit 190 is encoded by the encoding unit 120 and output as the main code Im. The signal is locally decoded by the decoding unit 130 to be a local signal. The local decoded signal of the code Im can be generated by the analysis unit coding method in the encoding unit 120. The local signal is not subjected to the inter-channel orthogonal inverse transform here, and the error calculation unit 140 calculates each error signal corresponding to the inter-channel orthogonal transform signal from the inter-channel orthogonal transform unit 190. This error signal is supplied to the array conversion unit 50. Thereafter, the same processing as in the encoder of the fourteenth embodiment is performed.

In the decoder 20, the main signal of the main code Im decoded by the decoding unit 60 based on the difference from the fourteenth embodiment and the error signal reproduced by the array inverse conversion unit 70 correspond to each other by the adding unit 240. In addition, the inter-channel orthogonal inverse transform unit 290 performs inverse transform corresponding to the inter-channel orthogonal transform unit 190 of the encoder 10 to obtain a frame as a reproduced digital signal of each channel. Supplied to the combining unit 250.
Also in this case, if the error signal can be correctly reproduced, the multi-channel digital signal can be correctly reproduced. Similarly to the fourteenth embodiment, the encoding unit 120 selects only a component with high power in the inter-channel orthogonal transform signal or determines a component with high average power in advance and encodes only this component. It is also possible to improve the encoding efficiency. The encoding unit 120 may be lossy high compression encoding or lossless compression encoding.

As a specific configuration example of the array conversion unit 50 in FIGS. 39 and 40, for example, the configuration of the lossless encoding unit 18 shown in FIG. 33 can be used as it is. As a specific configuration example of the array inverse transform unit 70 corresponding to the array transform unit 50, the configuration of the lossless decoding unit 21 illustrated in FIG. 34 can be used as it is.
Alternatively, the lossless encoding unit 18 illustrated in FIG. 35A is used as a specific configuration example of the array conversion unit 50, and the lossless decoding unit 21 illustrated in FIG. 35B is used as a specific configuration example of the array inverse conversion unit 70 correspondingly. May be.
Sixteenth Embodiment Using the lossy compression code and the lossless code of the error signal as described above, the lossy compression code is also reproduced by the encoding / decoding method for realizing lossless encoding. Depending on the environment of the decoder or computer, the reproduced signal of the lossy compression code may contain a slight error. In such a case, even if the decoder combines the decoded signal of the lossy compression code and the error signal obtained by decoding the lossless code, the reproduced digital signal does not completely match the original digital signal.

This sixteenth embodiment uses lossy compression coding and lossless coding of its error signal, and in theory, a reproduced digital signal that perfectly matches the original digital signal regardless of the decoder or computer environment. An encoding method, a decoding method, an encoder, and a decoder that can obtain the above are provided, which will be described below with reference to FIG.
In FIG. 41, in the encoder 10, the digital signal from the input terminal 100 is divided by the frame dividing unit 110 in units of frames, for example, every 1024 samples, and the digital signal is encoded by the lossy encoding unit 120 for each frame. Output as main code Im. The main code Im is decoded by the local decoding unit 14 to generate a local decoded signal, and the local decoded signal is supplied to the fluctuation maximum digit number detecting unit 55 and the cutoff unit 56. The local decoded signal of the code Im can be generated by the irreversible encoding unit 120 by the analysis / synthesis encoding method. In the variable maximum digit number detection unit 55, the number of guaranteed digits of the decoded signal when the main code Im is decoded by the decoder in the frame or in one or more subframes in the frame, that is, the decoding accuracy of the decoder is determined. Detect the maximum number of digits that can vary based on.

For example, a decoder conforming to MPEG-4 audio is guaranteed to be within a range of ± 1 from the reference reproduction signal, regardless of the environment. Therefore, in consideration of the local decoding unit 14, there is a possibility that an error of ± 2 is included in the amplitude of the reproduction signal of various decoders. Depending on the amplitude of the local decoded signal, the binary representation value may change up to the number of upper digits due to the error of ± 1.
For example, the amplitude of the locally decoded signal is
8192 (0010000000000000 in binary)
In the case of
8191 (0001111111111111 in binary)
Or
8193 (0010000000000001 in binary)
It is thought that it is.

So with another decoder
8190 (0001111111111110 in binary)
From
8194 (0010000000000010 in binary)
It is thought that it becomes. In this case, even if the environment changes, the decoded value is guaranteed only in the upper two digits.
Here, the number of digits with which the binary representation of the amplitude of the reproduced (decoded) waveform matches up to the number of digits specified by a decoder in a different environment within a frame or one or more subframes is defined as the minimum value of accuracy. This is called the maximum number of digits that can be changed. The maximum number of digits when the above number 8192 is included is 14. In the numerical example, the minimum accuracy value is the upper two digits. In the case of a negative number, the same processing is performed on the absolute value. In addition, when the decoded amplitude is any of -2, -1, 0, 1, and 2, it is determined that all are set to 0 so that the polarity is not inverted. The variation maximum digit number detection unit 55 calculates the accuracy guaranteed for each sample of each local decoded signal for each frame or one or more subframes, that is, calculates the amplitude value guaranteed for decoding, and calculates the frame or one or more subframes. The minimum value of accuracy guaranteed for each subframe, that is, the maximum value of the number of digits that can be varied is obtained as shown in FIG.

The variation maximum digit number detection unit 55 calculates the number of digits that can vary depending on the environment in this way for each sample of the local decoded signal, and calculates the maximum value of the number of digits that can vary for each frame or more than one subframe in the frame. This is detected and output as a digit number code Ig, and the truncation unit 56 generates a guaranteed local decoded signal by truncating the local decoded signal below the numerical value of the maximum value of the fluctuation digits. The guaranteed local decoded signal is the line 11 below the oblique line in FIG.
An error signal between the guaranteed local decoded signal and the digital signal from the frame dividing unit 110 is calculated by the error calculating unit 16. This error signal becomes a signal of the portion below the guaranteed local decoding signal line 11 in FIG. This error signal is losslessly encoded by the lossless encoding unit 18 and output as an error code Ine. The main code Im, the error code Ine, and the digit code Ig are combined by the combining unit 315 and output from the output terminal 170.

In the decoder 20, the input code from the input terminal 200 is separated into a main code Im, a digit code Ig, and an error code Ine by a separation unit 440 for each frame. The main code Im is irreversibly decoded by the irreversible decoding unit 60 to generate a decoded signal. The number-of-digits code Ig is decoded by the number-of-digits decoding unit 81 to obtain the maximum number of digits of fluctuation. In the truncation unit 82, the portion of the decoded signal from the irreversible decoding unit 60 that is less than or equal to the maximum number of digits of the variation is discarded and output as a guaranteed decoded signal.
The error code Ine is losslessly decoded by the lossless decoding unit 21 to reproduce an error signal. The error signal and the guaranteed decoded signal from the cutoff unit 82 are added by the adding unit 240 and supplied to the frame synthesizing unit 250 as a reproduced digital signal. The frame synthesis unit 250 sequentially combines the reproduction signals for each frame and outputs them to the output terminal 260.

Thus, the guaranteed local signal generated by the encoder 10 is not affected by the worst total value of errors generated by the local decoder 14 and the decoder 20 and is generated by the decoder 20. The guaranteed decoded signal is theoretically completely identical to the guaranteed local decoded signal of the decoder. Therefore, if the error signal is reproduced correctly, ideally, the reproduced digital signal is completely identical to the original digital signal. Obtainable.
In the sixteenth embodiment, the digit code Ig may be an encoded minimum value of accuracy.
Seventeenth Embodiment The maximum number of digits of fluctuation detected by the fluctuation maximum digit number detection unit 55 is often about 3 or 4 bits, for example, when one sample is expressed in 16 bits. If there is a sample whose decoding guarantee accuracy is only the upper two digits, the minimum value of accuracy, that is, the maximum number of digits of fluctuation becomes large, the error signal amplitude increases, and the error code output from the lossless encoding unit 18 The number of bits of Ine increases and the encoding efficiency deteriorates. However, this is the case when the sample value of the locally decoded signal is 1 in only one place in the upper digits and all other digits are 0 as in the above numerical example, and this is rare. That is, excluding such exceptional samples, for example, as shown in FIG. 43A, the minimum value of accuracy is about 12 or 13 bits.

In the twenty-third embodiment, the minimum value of accuracy excluding exceptional samples for each frame or one or more subframes, that is, the maximum number of digits that can be changed, and the information (in the frame or subframe) of the exceptional samples Then, the decoding guarantee accuracy, that is, the maximum number of digits of fluctuation is encoded, respectively, to produce a guaranteed local decoded signal shown in FIG. 43A.
For example, as shown in FIG. 44 with the same reference numerals assigned to the portions corresponding to FIG. 41, in the encoder 10, the local decoded signal from the local decoding unit 14 is supplied to the exception detecting unit 57, and the fluctuation is exceptionally large. The number of digits to be obtained (exceptionally small guaranteed accuracy) is detected, the number of digits (or accuracy) and the sample position information are supplied to the fluctuation maximum digit number detection unit 55 and the truncation unit 56, and the digit that can be changed The number (or precision) and sample position information are encoded and output as an exception code Ige.

  The variation maximum digit number detecting unit 55 removes the exception sample from the local decoded signal, detects the maximum number of digits (or the minimum accuracy value) that can be changed in the removed local decoded signal, and cuts this out. Is encoded and output as a digit code Ig. The truncation unit 56 rounds off the numerical value of the variable number of exception samples from the local decoded signal, and rounds off the maximum number of variable digits of the other parts to obtain the guaranteed local decoded signal. This guaranteed local decoded signal is supplied to the error calculator 16. Further, the combining unit 315 synthesizes and outputs the exception code Ige in addition to the main code Im, the error code Ine, and the digit code Ig. Others are the same as the embodiment of FIG.

Exceptional samples are small in accuracy (predictable number of digits is large) for most samples, or small in precision (predetermined number of digits is large). ) Are selected in order, and in the combining unit 315, for example, as shown in FIG. 43B, for each frame (or subframe), a code indicating the number of exceptional samples, then the positional information of the exceptional samples and the number of digits that can vary (accuracy) ) Are sequentially arranged and output as a lump with a digit code Ig at the end.
In the decoder 20, the separated exception code Ige is decoded by the exception decoding unit 83 to obtain sample position information and the number of digits that can vary. In the truncation unit 82, the sample indicated by the sample position information from the exception decoding unit 83 in the decoded signal from the lossy decoding unit 60 is rounded down below the maximum variation digit number, and the other samples are subjected to digit number decoding. Less than the maximum number of digits that can be changed from the part 81 is rounded down, and a guaranteed local signal is generated. Other processes are the same as those in the embodiment of FIG.

According to the seventeenth embodiment, the information amount of the error signal in the encoder 10 is reduced, and the encoding efficiency in the lossless encoding unit 18 is improved accordingly.
As a specific configuration example of the lossless encoding unit 18 in FIGS. 41 and 44 described above, for example, the configuration of the lossless encoding unit 18 shown in FIG. 33 can be used as it is, as described above. In this case, the configuration of the lossless decoding unit 21 in the decoder 20 can use the configuration of the lossless decoding unit 21 shown in FIG. 34 as it is. Alternatively, the configuration of the lossless encoding unit 18 illustrated in FIG. 35A may be used as the configuration of the lossless encoding unit 18. In this case, the configuration shown in FIG. 35B can be used as it is as the configuration of the lossless decoding unit 21. ,
The above-described encoding method and decoding method of each embodiment according to the present invention can be described in a storage medium as a computer-executable program, and can be read and executed by a computer as necessary.

As described above, according to the encoding method and the decoding method according to the first aspect of the present invention, the corresponding bits of a series of samples are generated for each frame to generate a coordinate bit string and output as a packet. Even if a packet loss occurs in the transmission path, the corresponding bit loss in the sample sequence of the decoded frame does not cause much quality degradation.
According to the encoding method and the decoding method according to the second aspect of the present invention, the maximum number of digits that fluctuates based on the accuracy of the decoder for each frame or subframe according to the accuracy of the decoded signal guaranteed by the decoder. Detected and rounded down the maximum number of digits of fluctuation in the local decoded signal to make a guaranteed local signal, and the error signal between this and the original digital signal is losslessly encoded. By truncating below the maximum number of digits of variation, it is possible to obtain a guaranteed decoded signal that logically matches the guaranteed local signal of the encoder.

As described above, the encoding method, decoding method, encoder, and decoder according to the present invention can be expressed as follows.
Encoding method Method according to the first aspect: An encoding method for encoding a digital signal for each frame including a plurality of samples includes the following steps.
(a) Generate a plurality of pairs of reversible data of bit strings straddling digital signal samples for each frame at each bit position, or reversible data sets of irreversible data and error signals derived from the irreversible data. ,
(b) The above data set is output as a code.

Encoding method according to the second aspect: in the encoding method according to the first aspect, the step (a) converts the amplitude of each sample into a binary representation of a polar code and an absolute value in advance before lossless encoding. An encoding method comprising steps.
Encoding method according to the third aspect: in the encoding method according to the first aspect, the step (b) includes a step of storing the lossless code and generating and outputting a packet with the header information.
Encoding method according to the fourth aspect: In the encoding method according to the first aspect, the step (b) includes a step of assigning a high priority to the bit string in descending order from the polarity code and the absolute value.

Encoding method according to the fifth aspect: In the encoding method according to the first aspect, the step (a) includes:
(a-1) irreversibly encoding the original signal to generate lossy compressed information and a local decoded signal;
(a-2) irreversibly encoding an error signal between the local decoded signal and the original signal as the digital signal;
Including
The step (b) outputs the lossy compression information together with the lossless code.

Encoding method according to sixth aspect: In the encoding method according to the first aspect, a parameter expressing the spectral envelope of the digital signal is obtained, and the code generated by encoding the parameter is used as auxiliary information together with the lossless code. Output step.
Encoding method according to seventh aspect: In the encoding method according to the first aspect, the step (a) includes
(a-1) determining the number of digits representing the maximum absolute value of the amplitude of the digital signal in each frame as the number of significant digits;
(a-2) At least one bit sequence in which at least one corresponding bit of each sample is arranged in the time direction over the frame for a portion within the effective number of digits for each frame is a part of the lossless code. Generating as transmission record unit data;
Including
The step (b) includes a step of outputting the effective number of digits together with the lossless code for each frame.

Encoding method according to the eighth aspect: in the encoding method according to the seventh aspect, the step (b) stores and outputs the number of significant digits as auxiliary information or in any packet of the frame Includes steps.
Encoding method according to ninth aspect: in the encoding method according to the sixth aspect, the step (a),
(a-1) obtaining a linear prediction coefficient and a current prediction value from the past digital signal as the above parameters;
(a-2) subtracting the predicted value from the current sample to obtain a prediction error;
Including
In step (b), the linear prediction coefficient is encoded and output as auxiliary information together with the lossless code.

Encoding method according to the tenth aspect: The encoding method according to the first aspect further includes:
(0-1) generating an irreversible code by irreversibly encoding an input signal in consideration of auditory characteristics for each frame, and generating a local decoded signal;
(0-2) transforming the local decoded signal so as to reduce an error from the input signal, and generating a modified local decoded signal;
(0-3) generating an error signal between the input signal and the modified local decoded signal as the digital signal.

Encoding method according to the eleventh aspect: In the encoding method according to the tenth aspect,
The step (0-2) calculates a deformation parameter including a cross-correlation component between the digital signal and the local decoded signal, and outputs the deformation parameter code;
Multiplying or convolving the local decoded signal with the modified parameter to produce the modified local decoded signal.
Encoding method according to the twelfth aspect: in the encoding method according to the tenth aspect, the step (0-2) includes:
(0-2-1) multiplying the local decoded signal by a deformation parameter, or combining a plurality of samples with a deformation parameter to create the modified local decoded signal;
(0-2-2) generating the deformation parameter for each frame so that the energy of the error signal is reduced.

Encoding method according to the thirteenth aspect: in the encoding method according to the tenth aspect, the step (0-2) is:
(0-2-1) calculating an error of the locally decoded signal with respect to the digital signal to obtain an error signal;
(0-2-2) encoding the error signal so that the quantization error is minimized to generate an error minimized code, and generating a second local decoded signal for the error minimized code;
(0-2-3) adding the second local decoded signal and the local decoded signal to obtain the modified local signal.

Encoding method according to the fourteenth aspect: in the encoding method according to the tenth aspect, the step (0-2) is:
(0-2-1) encoding the digital signal so that a quantization error is minimized to generate an error minimizing code, and generating a second local signal for the error minimizing code;
(0-2-2) The second local signal and the local signal are subjected to a weighted average with emphasis on the former to obtain the deformed local signal.
Encoding method according to the fifteenth aspect: The encoding method according to the first aspect further includes:
(0-1) a step of irreversibly compressing the input signal for each frame in consideration of auditory characteristics and outputting a first irreversible code;
(0-2) irreversibly compression-encoding the digital signal to generate a second irreversible code and generating a local signal for the second irreversible code;
(0-3) obtaining an error signal between the local signal and the digital signal as the digital signal;
Including.

Encoding method according to the sixteenth aspect: The encoding method according to the first aspect further comprises:
(0-1) compressing and encoding the input signal for each frame so as to minimize the quantization error to generate an error minimizing code, and generating a first local signal for the error minimizing code;
(0-2) generating an error signal as the digital signal with respect to the input signal of the first local signal;
(0-3) generating a lossless code by irreversibly compressing the error signal in consideration of auditory characteristics.

Encoding method according to the seventeenth aspect: The encoding method according to the first aspect further comprises:
(0-1) Mixing an M-channel input signal into an N-channel signal, and M is an integer of 2 or more, N is an integer of 1 or more smaller than M, and the N-channel local part for the main code Generating a decoded signal; and
(0-2) The N-channel mixed signal is encoded to generate a main code,
(0-3) expanding the N-channel local decoded signal to an M-channel local decoded signal;
(0-4) obtaining an error signal between each M channel local decoded signal and the corresponding M channel input signal as the digital signal;
Including
The step (b) includes a step of outputting the main code together with the lossless code.

Encoding method according to the eighteenth aspect: The encoding method according to the sixteenth aspect further comprises:
(0-1) mixing the N-channel mixed signal into one or more L-channel mixed signals less than N;
(0-2) encoding the mixed signal of the L channel to generate a main code and generating a channel local signal for the main code;
(0-3) expanding the L channel local decoded signal to an N channel local decoded signal;
(0-4) generating a difference between the N-channel local decoded signal and the N-channel mixed signal as a first error signal;
(0-5) encoding the first error signal to generate a slave code, and generating an N-channel local error signal for the slave code;
(0-6) adding the N-channel local decoded signal and the N-channel local decoded error signal to generate an N-channel added local signal;
(0-7) Channel expanding the N channel addition local decoded signal to generate an M channel addition local signal;
(0-8) generating a difference between the M channel addition local signal and the M channel digital signal as a second error signal;
Including
The step (b) outputs the sub code together with the main code.

Encoding method according to the nineteenth aspect: The encoding method according to the first aspect further comprises:
An M-channel input signal is orthogonally transformed between channels to include an orthogonally transformed signal as the digital signal, M is an integer of 2 or more, and the above steps (a) and (b) are executed for each of the M channels. To do.
Encoding method according to twentieth aspect: The encoding method according to the first aspect further includes:
(0-1) a step of orthogonally transforming an input signal of M channels to generate an orthogonally transformed signal, and M is an integer of 2 or more,
(0-2) encoding at least a part of the orthogonal transform signal to generate a main code, and generating a local decoded signal for the main code;
(0-3) inter-channel orthogonal inverse transform of the local decoded signal to generate an M channel local decoded signal;
(0-4) obtaining an error signal between each M channel local decoded signal and the corresponding M channel input signal as the digital signal;
Including
The step (b) outputs the sub code together with the main code.

Encoding method according to the twenty-first aspect: The encoding method according to the first aspect further includes:
(0-1) a step of orthogonally transforming an input signal of M channels to generate an orthogonally transformed signal, and M is an integer of 2 or more,
(0-2) encoding at least a part of the orthogonal transform signal to generate a main code, and generating a local decoded signal for the main code;
(0-3) obtaining a difference between each locally decoded signal and the corresponding orthogonal transform signal as an error signal;
Including
The step (b) outputs the main code together with the lossless code.

Encoding method according to the twenty-second aspect: In the encoding method according to the first aspect,
(a-1) lossy encoding of a digital signal, generating an irreversible code, generating a local decoded signal for the irreversible code,
(a-2) The maximum number of digits in which the decoded signal for the lossy code can vary for each frame or one or more subframes in the frame is detected from the decoded signal, and the number of digits indicating the maximum number of variations Output the sign,
(a-3) The above-mentioned local decoded signal is generated by rounding off the ab sentence below the maximum number of digits of the fluctuation in the digital representation,
(a-4) Create an error signal between the accuracy-determined local decoded signal and the digital signal,
(a-5) The error signal is losslessly encoded and a lossless code is output.
A decoding method according to the twenty-third aspect: a decoding method for reproducing a sample of a digital signal for each frame,
(a) Multiple sets of reversible data of bit strings obtained by decoding input codes and straddling digital signal samples at their respective bit positions, or reversible data sets of error signals derived from irreversible data and the irreversible data A step of generating
(b) regenerating a digital signal based on the plurality of sets of data;
Including.

Decoding method of the twenty-fourth aspect: In the decoding method of the twenty-third aspect,
The step (b) includes the step of converting the sample string from a binary representation of a polarity code and an absolute value to a two's complement representation.
A decoding method according to a twenty-fifth aspect: In the decoding method according to the twenty-third aspect,
The step (b) includes a step of smoothing the digital signal and performing the digital signal.
Decoding method of the twenty-sixth aspect: In the decoding method of the twenty-third aspect, in step (b), the auxiliary information is decoded to obtain a spectral envelope, and the spectral envelope of the digital signal approaches the decoded spectral envelope Thus, the step of correcting the digital signal is included.

Decoding method of the twenty-seventh aspect: In the decoding method of the twenty-sixth aspect, the step (b)
(b-1) substituting provisional samples for the missing or error bits;
(b-2) calculating a spectral envelope of the provisional sample;
(b-3) flattening the spectral envelope of the provisional sample according to the characteristics of the calculated spectral envelope or its modified spectral envelope;
(b-4) reconstructing samples using the decoded spectral envelope or a modified spectral envelope thereof with respect to the flattened spectral envelope;
Including.

The decoding method according to the twenty-eighth aspect: in the decoding method according to the twenty-sixth aspect, the step (b) comprises:
(b-1) substituting a temporary sample for the missing or erroneous bit;
(b-2) calculating a spectral envelope of the provisional sample;
(b-3) calculating one spectral envelope combining the decoded spectral envelope or the modified spectral envelope according to the calculated spectral envelope or the inverse characteristic of the modified spectral envelope;
(b-4) reconstructing the temporary sample sequence using the one spectral envelope;
(b-5) correcting an erroneous correction in the reconstructed sample sequence;
The decoding method characterized by including these.

The decoding method according to the twenty-ninth aspect: in the decoding method according to the twenty-seventh aspect, the step (b-3) converts the calculated spectral envelope into a linear prediction cepstrum coefficient Ca and linearly predicts the decoded spectral envelope. A step of obtaining the one spectral envelope by obtaining a difference Cb-Ca of these linear prediction cepstrum coefficients by converting into the cepstrum coefficient Cb or using the decoded spectral envelope for the auxiliary information is included.
The decoding method according to the thirtieth aspect: In any one of the decoding methods according to the twenty-seventh, twenty-eighth or twenty-ninth aspect,
(b-5) Check whether the error between the spectrum envelope of the tentative sample and the decoded spectrum envelope is within a predetermined value. If it is within the predetermined value, the tentative sample at that time is regarded as a corrected sample. If not within the predetermined value, the above steps (b-2), (b-3) and (b-4) are repeated.

Decoding method according to the thirty-first aspect: in the decoding method according to the twenty-third aspect, the step (a) comprises irreversibly decoding the input irreversible encoded information, Adding the signals.
Decoding method according to the thirty-second aspect: in the decoding method according to the twenty-third aspect, the step (a) includes decoding the transmission record unit data of the lossless code and at least one of the one frame based on header information Generating at least one decoded bit string of bit positions, wherein the step (b) detects missing or error bits in the transmission record unit data, and is input to the sample string for each frame. A step of performing alignment according to the number of significant digits.

Decoding method according to the thirty-third aspect: in the decoding method according to the thirty-second aspect, if the step (b) is greater than the number of significant decoding digits of the previous frame, After reducing the amplitude value of the sample of the previous frame and aligning the number of significant digits of the previous frame with the number of significant digits of the current frame, perform correction processing,
If the number of significant digits of the current frame is smaller than the number of significant digits of the previous frame, the amplitude value of the sample of the previous frame is expanded and the number of significant digits of the previous frame is After aligning with the effective number of digits, correction processing is performed.

A decoding method according to a thirty-fourth aspect: in the decoding method according to the twenty-third aspect, the sample is a sample of a prediction error signal;
(c) correcting the prediction error signal based on the spectral envelope of the prediction error signal for missing or erroneous data;
(d) decoding input auxiliary information to obtain linear prediction information;
(e) synthesizing a reproduction original signal from the prediction error signal signal and a past sample of the reproduction original signal based on the linear prediction;
Including.

Decoding method according to the thirty-fifth aspect: in the decoding method according to the thirty-fourth aspect, the step (c) comprises:
(c-1) substituting a provisional sample for the missing or erroneous bit;
(c-2) transcribing the spectral envelope of the provisional sample;
(c-3) determining the flatness of the spectral envelope, and if the flatness is within a predetermined value, the provisional sample as the corrected prediction error signal;
(c-4) If the flatness is not within a predetermined value, the provisional sample is normalized by the spectral envelope or its modified spectral envelope to obtain a flattened signal;
(c-5) The above steps (c-1) to (c-4) are repeated using the normalized signal as the provisional sample.

Decoding method according to thirty-sixth aspect: in the decoding method according to the thirty-fourth aspect, the step (c) comprises
(c-1) substituting a provisional sample for the missing or erroneous bit;
(c-2) filtering the temporary sample using the decoded linear prediction coefficient to generate a synthesized signal;
(c-3) calculating a spectral envelope of the composite signal;
(c-4) normalizing the provisional error signal with the calculated spectrum envelope or its modified spectrum envelope to create a flattened spectrum signal;
(c-5) reconfiguring a prediction error waveform by filtering the signal having the flattened spectrum using the decoded linear prediction coefficient;
Including.

A decoding method according to a thirty-seventh aspect: in the decoding method according to the thirty-fourth aspect, the step (c) comprises:
(c-1) substituting a provisional sample for the missing or erroneous bit;
(c-2) filtering the temporary sample using the decoded linear prediction coefficient to generate a synthesized signal;
(c-3) calculating a linear prediction coefficient of the synthesized signal;
(c-4) calculating a linear prediction coefficient that combines the inverse characteristic of the calculated linear prediction coefficient or the linear prediction coefficient whose band is expanded, and the decoded linear prediction coefficient or the linear prediction coefficient whose band is expanded; ,
(c-5) generating a prediction error signal by filtering the provisional sample using the combined linear prediction coefficient;
Including.

The decoding method according to the thirty-eighth aspect: in the decoding method according to the thirty-sixth or thirty-seventh aspect, further checks whether or not an error between the calculated linear prediction coefficient and the decoded linear prediction coefficient is within a predetermined value. If it is within the range, the provisional sample at that time is used as a corrected prediction error waveform, and if it is not within the predetermined value, the synthesized signal is used as the provisional sample and the above steps (c-1) to (c-5) are performed. repeat.
The decoding method according to the thirty-ninth aspect: the decoding method according to the thirty-first aspect, further comprising the following steps:
(c) modifying the local decoded signal so as to reduce an error between the local decoded signal and the digital signal, and generating a modified signal;
(d) synthesizing the deformation signal and the error signal to update the reproduction signal;
A decoding method according to a 40th aspect of the invention, comprising: a decoding method according to the 39th aspect, wherein the step (c) comprises:
(c-1) decoding a deformation parameter code to obtain a deformation parameter;
(c-2) multiplying or decoding the decoded modified parameter with respect to the decoded signal to obtain the modified decoded signal;
Including.

The decoding method according to the forty-first aspect: in the decoding method according to the thirty-ninth aspect, the step (c) comprises:
(c-1) generating the modified signal by multiplying or convolving the local decoded signal with at least one modified parameter;
(c-2) generating the deformation parameter so that the energy of the error signal between the deformation signal and the reproduction signal is reduced;
Including.

The decoding method according to the forty-second aspect: the decoding method according to the twenty-third aspect, further comprising:
(c) irreversibly decoding the input error minimizing code to reproduce the first local decoded signal;
(d) adding the digital signal and the first local decoded signal to reproduce the first digital signal;
(e) irreversibly decoding a second irreversible code in which auditory characteristics are taken into account to reproduce a second local decoded signal;
(f) outputting the first digital signal or the second digital;
Including.

Decoding method according to the 43rd aspect: in the decoding method according to the 39th aspect, the step (c) comprises:
(c-1) decoding the error minimizing code to obtain a second local decoded signal;
(c-2) adding the second local decoded signal to the local decoded signal to generate the modified signal;
Including.
Decoding method of the forty-fourth aspect: in the decoding method according to the thirty-ninth aspect,
(c-1) decoding the error minimizing code to produce a second local decoded signal;
(c-2) A step of weighted averaging the second local decoded signal and the local decoded signal with emphasis on the former to produce the modified decoded signal;
Including.

Decoding method according to the 45th aspect: in the decoding method according to the 23rd aspect, the sample is a sample of an error signal of M channels, M is an integer of 2 or more, and the decoding method further includes:
(c) decoding an input main code to obtain an N-channel decoded signal; N is an integer greater than or equal to 1 less than M;
(d) expanding the N-channel decoded signal to an M-channel decoded signal;
(e) adding the M channel error signal and the M channel decoded signal to reproduce an M channel digital signal;
Including.

A decoding method according to a 46th aspect: in the decoding method according to the 23rd aspect, the sample is a sample of an error signal of M channels, M is an integer of 2 or more, and the decoding method further includes:
(c) decoding an input main code to obtain an L channel decoded signal, and L is an integer of 1 or more,
(d) expanding the L channel decoded signal to generate an N channel decoded main signal; N is an integer greater than L and less than M;
(e) decoding the slave code to obtain an N-channel decoded slave signal;
(f) adding the N-channel decoded main signal and the N-channel decoded slave signal to generate an N-channel added signal;
(h) expanding the N channel addition signal to generate an M channel addition signal;
(i) adding the M channel error signal and the M channel addition signal to obtain an M channel reproduction digital signal;
Including.

47. Decoding method according to the 47th aspect: the decoding method according to the 23rd aspect, wherein the sample is a sample of a plurality of channels, and the decoding method further comprises:
(c) performing a multi-channel orthogonal inverse transform on the plurality of channel sample sequences to reproduce a plurality of channels of digital signals;
including.
A decoding method according to a forty-eighth aspect: in the decoding method according to the twenty-third aspect, the sample is a sample of an M channel error signal, M is an integer of 2 or more, and the decoding method further includes:
(c) decoding the input main code to obtain a locally decoded signal;
(d) performing an interchannel orthogonal inverse transform on the M channel local decoded signal to obtain an M channel decoded signal;
(e) adding the M channel decoded signal and the M channel error signal to reproduce an M channel digital signal;
Including.

The decoding method according to the forty-ninth aspect: in the decoding method according to the twenty-third aspect, the sample is a sample of an error signal of M channels, M is an integer of 2 or more, and the decoding method further includes:
(c) decoding an input main code to obtain an M channel local decoded signal;
(d) adding the M channel local decoded signal and the M channel error signal to obtain an M channel addition signal;
(e) performing an interchannel orthogonal inverse transform on the M channel addition signal to reproduce an M channel digital signal;
Including.

Decoding method according to the fifty-first aspect: the decoding method according to the twenty-third aspect, further comprising the following steps:
(c) lossy decoding the lossy code to generate a local decoded signal,
(d) Decode the maximum digit code to obtain the maximum number of digits of fluctuation,
(e) The above-mentioned decoded signal is rounded down below the maximum number of digits of the above-mentioned variation in the digital representation to generate an accuracy guarantee signal,
(f) The reproduction signal is generated by adding the digital signal and the accuracy guarantee signal. ,
Decoding method according to the fifty-first aspect: in the decoding method according to the fifty-first aspect, the step (d) comprises decoding the exception code, obtaining the exception value and sample position information, Assigning the above exceptional value to the sample position,
Encoder according to the 52nd aspect of the encoder: an encoder that encodes a digital signal for each frame composed of a plurality of samples,
Means for generating a plurality of sets of reversible data of bit strings straddling the digital signal samples for each frame at each bit position, or a set of reversible data of error signals derived from the irreversible data and the irreversible data for each frame When,
An output means for outputting the code set of the data;
Including.

Encoder according to a 53rd aspect: an encoder according to the 52nd aspect, wherein the amplitude of each sample is converted into a binary representation of polarity and absolute value, and the converted sample is supplied to the rearrangement unit Includes a value converter.
Encoder according to a 54th aspect: in the encoder according to the 53rd aspect, the combining unit assigns a packet priority to the bit string in the order of polarity and absolute value.
The encoder according to the 55th aspect: the encoder according to the 52nd or 53rd aspect, wherein the original signal is irreversibly encoded and outputs the irreversible encoded information and the local decoded signal; and the local And a differentiator for obtaining and outputting an error signal between the decoded signal and the original signal as the digital signal.

Encoder according to the 56th aspect: auxiliary information generation for encoding the parameter expressing the spectral envelope of the digital signal and outputting it as the auxiliary information together with the packet in the encoder according to any one of the 52nd, 53rd or 54th aspect Has a part.
57th aspect encoder: The encoder according to the 52nd aspect further determines and outputs, as the effective number of digits, the number of digits representing the maximum absolute value of the digital signal for each of the divided frames. An auxiliary information generation unit is included, and the rearrangement unit generates the equivalence bit string for each frame only within the number of significant digits.

58th aspect encoder: The encoder according to the 52nd aspect further includes:
A spectral envelope calculation unit for obtaining a linear prediction coefficient representing the spectral envelope of the input digital signal for each frame;
An auxiliary information generation unit that encodes the linear prediction coefficient to obtain auxiliary information;
A prediction unit that obtains an integerized prediction value of the current input signal from a past digital signal and a linear prediction coefficient for each frame;
A prediction error generator for subtracting the predicted value from the current input digital signal to obtain a prediction error signal;
Including.

The encoder according to the 59th aspect: the encoder according to the 58th aspect further includes an effective digit number detection unit that obtains the number of digits representing the maximum value of the absolute value of the amplitude bit string for each frame as the effective number of digits; Means for encoding the number of significant digits and adding it to the auxiliary information, and the rearrangement unit rearranges the amplitude bit string of the sample string into a coordinate bit string only within the range of the number of significant digits.
The encoder according to the 60th aspect: the encoder according to the 52nd aspect further comprises:
An irreversible encoding unit that generates an irreversible code by irreversibly encoding an input digital signal for each frame in consideration of auditory characteristics;
A modified unit that receives the local decoded signal and transforms the local decoded signal so as to reduce an error from the input signal, thereby generating a modified local decoded signal;
An error calculator that receives the input signal and the modified local signal and generates an error signal between them as the digital signal;
Including.

The encoder according to the 61st aspect: the encoder according to the 52nd aspect, wherein the digital signal is an M-channel digital signal, M is an integer of 2 or more, and the encoder further includes:
An M channel digital signal is input for each frame, a channel mixing unit that mixes these M channel digital signals and outputs an N channel mixed signal, M is an integer of 2 or more, and N is smaller than M 1 An integer greater than or equal to
An encoding unit that receives the N-channel mixed signals and encodes them to generate a main code and an N-channel local signal for the main code;
A channel expansion unit that receives the N-channel local decoded signal, expands these channels, and outputs an M-channel local decoded signal;
An error calculating unit for calculating an error signal between the M channel local decoded signal and the M channel digital signal as the digital signal;
The output means outputs the main code together with the lossless code.

The encoder according to the 62nd aspect: the encoder according to the 52nd aspect further comprises:
An inter-channel orthogonal transform unit that performs inter-channel orthogonal transform on an M-channel input signal and outputs an orthogonal transform signal; and M is an integer of 2 or more,
An encoding unit that generates a main code by encoding at least a part of the signal of the orthogonal transform signal, and generates a local decoded signal for the main code;
An inter-channel orthogonal inverse transform unit for generating an M-channel local decoded signal by performing an orthogonal inverse transform between channels on the local decoded signal;
An error calculating unit that generates an error signal as the digital signal between each of the M channel local decoded signals and the corresponding one of the M channel digital signals;
The output means outputs the main code together with the lossless code.

The encoder according to the 63rd aspect: the encoder according to the 52nd aspect, wherein M is an integer of 2 or more, and the encoder further comprises:
An inter-channel orthogonal transform unit that performs inter-channel orthogonal transform on an M-channel input signal to generate an orthogonal transform signal; and M is an integer of 2 or more,
An encoding unit that encodes at least a part of the orthogonal transform signal to generate a main code and generates a local decoded signal for the main code;
An error calculator that calculates an error signal between each of the locally decoded signals and the corresponding orthogonal transform signal as the digital signal;
The output means outputs the main code together with the lossless code.

64th aspect encoder: In the encoder according to the 52nd aspect,
An irreversible encoding unit that irreversibly encodes an input signal and outputs an irreversible code;
A local decoding unit that decodes the lossy code and outputs a local decoded signal;
For each frame or one or more subframes in the frame, the maximum number of digits that the decoded signal for the lossy code can vary depending on the environment is detected from the local decoded signal, and a digit code indicating the maximum number of digits of the variation is output. A variable maximum digit number detection unit,
The local decoded signal, a truncation unit that generates a guaranteed local decoded signal by truncating a portion below the maximum number of digits of the fluctuation,
An error calculation unit for creating an error signal between the guaranteed local decoded signal and the digital signal;
A lossless encoding unit that losslessly encodes the error signal and outputs a lossless code;
Including.
Decoder according to the 65th aspect: a decoder that reproduces a sample of a digital signal for each frame of a plurality of samples;
Decodes the input code and generates a reversible set of reversible data of the bit string straddling digital signal samples at each bit position, or a reversible set of error signals derived from the irreversible data and the irreversible data. Data generation means for
Means for reproducing a digital signal from the data set;
Including.

Decoder according to the 66th aspect: a decoder according to the 65th aspect, wherein a two's complement conversion unit converts each of the samples from a binary representation of a polarity code and an absolute value into a two's complement representation to obtain the sample. Have.
67th aspect decoder: The decoder according to the 65th or 66th aspect further includes a missing correction unit that estimates a missing bit string from known information and corrects the sample string.
68th aspect decoder: In the 67th aspect decoder, the missing correction section is a low-pass filter for smoothing the input.

The decoder according to the 69th aspect: The decoder according to the 67th aspect further comprises:
An auxiliary information decoding unit that decodes the input auxiliary information to obtain a spectrum envelope;
The missing correction unit corrects the sample sequence so that the spectrum envelope of the corrected sample sequence approaches the decoded spectrum envelope.
70th aspect decoder: In the 69th aspect decoder, the missing correction section comprises:
A provisional sample generation unit that generates a provisional sample string in which the missing or error bits are replaced across samples;
A spectrum envelope calculation unit for calculating a spectrum envelope of the provisional sample sequence;
An inverse filter that normalizes the spectral envelope of the provisional sample sequence with the decoded spectral envelope or a modified spectral envelope thereof;
For the flattened spectral envelope, a synthesis filter that reconstructs a corrected sample sequence using the decoded spectral envelope or a modified spectral envelope thereof and the flattened spectral envelope;
Including.

The decoder according to the 71st aspect: In the decoder according to the 69th aspect, the missing correction section comprises:
A provisional waveform generator that reproduces a provisional sample string in which an appropriate value is substituted for the missing or error bit;
A spectrum envelope calculation unit for calculating a spectrum envelope of the provisional sample;
A composite spectral envelope calculation unit that calculates the inverse characteristic of the calculated spectral envelope or its modified spectral envelope and one spectral envelope linear prediction coefficient obtained by synthesizing the decoded spectral envelope or its modified spectral envelope;
A synthesis filter that reconstructs the provisional sample sequence using the one synthesized spectral envelope;
Including.

A decoder according to the 72nd aspect: in the decoder according to the 71st aspect, the combined spectrum envelope calculation unit includes a first coefficient conversion unit that converts the calculated spectrum envelope into a linear prediction cepstrum coefficient Ca, and the decoded spectrum envelope. A second coefficient conversion unit for converting the linear prediction cepstrum coefficient Cb into a linear prediction cepstrum coefficient Cb, a subtraction unit for obtaining the difference Cb-Ca between the linear prediction cepstrum coefficients, and inversely converting the difference Cb-Ca into a linear prediction coefficient, And an inverse transform unit for obtaining an envelope linear prediction coefficient.
Decoder according to the 73rd aspect: In the decoder according to the 70th or 71st aspect, an error between a spectrum envelope of the provisional sample sequence or the modified sample sequence and the decoded spectrum envelope is obtained, and the error is calculated. If it is within the predetermined value, the provisional or modified sample string at that time is regarded as a corrected sample string, and if not within the predetermined value, the provisional or modified sample string is regarded as the inverse filter or the above An error calculation unit for supplying to the synthesis filter is included.

Decoder according to the 74th aspect: a decoder according to the 73rd aspect, wherein the input lossy encoding is irreversibly decoded to generate a local decoded signal, and the digital signal is included in the local decoded signal And an adding unit that supplies the added sample sequence to the missing correction unit.
The decoder of the 75th aspect: The decoder of the 65th aspect further comprises:
An auxiliary information decoding unit that decodes the input auxiliary information to obtain the number of significant digits of the frame;
A digit alignment unit that performs alignment according to the number of significant decoding digits for the sample of the frame or the corrected sample,
Including.

Decoder according to the 76th aspect: In the decoder according to the 75th aspect, the auxiliary information decoding unit also decodes average power, and the missing correction unit uses the decoded average power to determine the amplitude of the sample. It is to correct.
A decoder according to the 77th aspect: in the decoder according to the 65th aspect, the linear prediction coefficient of the sample is a linear prediction coefficient of a prediction error signal, and the decoder further decodes the input auxiliary information to obtain a linear prediction coefficient. An auxiliary information decoding unit for obtaining a prediction coefficient;
The decoding original signal is synthesized from the prediction error signal and past samples of the decoding original signal using the linear prediction coefficient, and the error or missing data of the prediction error signal is corrected based on the spectral envelope of the prediction error signal. Correction section,
Including.

A decoder according to a 78th aspect: in the decoder according to the 77th aspect, the missing correction unit includes:
A provisional waveform generator that reproduces provisional samples to replace the missing or error bits;
A spectrum envelope calculation unit for obtaining a spectrum envelope of the provisional sample;
Determining the flatness of the spectrum envelope, determining whether or not the flatness is within a predetermined value, and if the flatness is within the predetermined value, a flatness determination unit that uses the provisional sample as the prediction error signal;
If the flatness is not within a predetermined value, the provisional sample is normalized by the spectrum envelope or its modified spectrum envelope to obtain a normalized signal,
Including.

The decoder according to the 79th aspect: in the decoder according to the 77th aspect, the missing correction section comprises:
A provisional waveform generation unit for substituting provisional samples for the missing or error bits;
A first synthesis filter that filters the provisional samples using the decoded linear prediction coefficients to generate a synthesized signal;
A spectrum envelope calculation unit for calculating a spectrum envelope of the synthesized signal;
Synthetic spectral envelope calculation unit that calculates one linear prediction coefficient by combining the inverse characteristic of the calculated linear prediction coefficient or the linear prediction coefficient whose band is expanded and the decoded linear prediction coefficient or the linear prediction coefficient whose band is expanded. When,
A combined second filter that filters the provisional samples to generate a prediction error signal using the combined linear prediction coefficients;
Including.

80th aspect decoder: In the 65th aspect decoder,
A deformation unit that receives the local decoded signal and generates a modified signal by modifying the local decoded signal so that an error between the local decoded signal and the digital signal is reduced;
An adder for synthesizing the deformation signal and the error signal to update the reproduction signal;
Including.
A decoder according to the 81st aspect: in the decoder according to the 65th aspect,
A decoding unit that decodes the main code and outputs an N-channel decoded signal; and M is an integer of 2 or more, N is an integer of 1 or more smaller than M,
A channel expansion unit for expanding the N-channel decoded signal to an M-channel decoded signal;
The missing detection unit that detects a missing bit and outputs a missing signal, and the information correction unit includes a missing information correction unit that adds missing information to the error signal corresponding to the missing bit detected by the missing detection unit. And
An adder for adding the M channel error signal or correction error signal and the M channel decoded signal to output an M channel digital signal;
Including.

Decoder according to the 82nd aspect: In the decoder according to the 65th aspect, the M rearrangement units are provided to output an error signal of M channels, M is an integer equal to or greater than 2, and the decoder further includes ,
A decoding unit that decodes the main code and outputs a locally decoded main signal;
An inter-channel orthogonal inverse transform unit that performs an inter-channel orthogonal inverse transform on the local decoded main signal to generate an M-channel decoded signal;
An adder for adding the M channel decoded signal and the M channel error signal to output a reproduced M channel digital signal;
Including.

The decoder according to the 83rd aspect: in the decoder according to the 65th aspect,
A decoding unit for decoding an input main signal and outputting a decoded main signal;
An adder for adding the decoded main signal and the M channel error signal to output an M channel addition signal;
An interchannel direct inverse transform unit that receives the M channel addition signal, performs an inverse transform between the channels, and outputs an M channel decoded signal;
Including.

The decoder according to the 84th aspect: in the decoder according to the 65th aspect,
An irreversible decoding unit that irreversibly decodes the irreversible code to generate a local decoded signal;
A digit decoding unit that decodes the maximum digit code and outputs the maximum number of fluctuations;
A truncation unit that truncates a portion of the local decoded signal that is less than or equal to the maximum number of digits of the fluctuation and outputs an accuracy guarantee signal;
A lossless decoding unit that reversibly decodes the input lossless code and reproduces an error signal;
An adder for adding the digital signal and the accuracy guarantee signal to output a reproduced digital signal;
Including.

Encoding program according to the 85th aspect: An encoding program for executing the encoding method according to the first to the twenty-second aspects on a computer.
A decoding program according to the 86th aspect: a decoding program for executing the decoding methods according to the 23rd to 51st aspects on a computer.

The figure which shows the function structure of the encoder and decoder which combined the conventional irreversible quantization. The figure for demonstrating the process of the rearrangement part 160 in FIG. The figure which shows the function structure of each Example 1 of the encoder of this invention, and a decoder. A is a diagram showing an example of processing of the rearrangement unit 160, and B is a diagram showing an example of a packet format. The figure which shows the comparative example of the spectrum envelope of an original sound and a distortion component in relation to the process of the missing correction | amendment part 430. FIG. The figure which shows the function structure of each 2nd Example of the encoder and decoder of this invention. 6 is a flowchart showing an example of a processing procedure using auxiliary information in a missing correction unit 430. The figure which shows the function structural example corresponding to FIG. 12 is a flowchart showing another example of a processing procedure using auxiliary information in the missing correction unit 430. The figure which shows the function structure of the specific example of the synthetic | combination spectrum envelope calculation part 437. The figure which shows the function structure of each 3rd Example of the encoder and decoder of original sound analysis synthetic | combination correction by this invention. The figure which shows the function structure of the deformation | transformation of 3rd Example of the decoder of the error analysis by this invention, and pre-combination correction | amendment. The figure which shows the result of the computer simulation for demonstrating the effect of this invention. The figure which shows the function structure of each 4th Example of the encoder and decoder of this invention. FIG. 7A is a diagram illustrating a processing example of the rearrangement unit 160, and B is a diagram illustrating a processing example of the rearrangement unit 160 with respect to a two's complement value. 6 is a flowchart showing an example of a processing procedure using auxiliary information in a missing correction unit 430. FIG. 17 is a diagram illustrating a functional configuration example corresponding to FIG. 16 of a missing correction unit 430; 12 is a flowchart showing another example of a processing procedure using auxiliary information in the missing correction unit 430. The figure which shows the function structure of each 5th Example of the encoder and decoder of the correction | amendment after original sound analysis synthetic | combination by this invention. The figure which shows the function structure of the deformation | transformation of 5th Example of the decoder of the error analysis by this invention, and a pre-combination correction | amendment. The figure which shows the function structure of each 6th Example of the encoder and decoder of this invention. 5 is a flowchart showing an example of a processing procedure that does not use auxiliary information in the missing correction unit 430. The figure which shows the function structural example corresponding to FIG. 6 is a flowchart showing an example of a processing procedure using auxiliary information in a missing correction unit 430. The figure which shows the function structural example corresponding to FIG. 12 is a flowchart showing another example of a processing procedure using auxiliary information in the missing correction unit 430. The figure which shows the function structural example of 1st Example of this invention. FIG. 28A is a diagram illustrating a specific example of the deformation parameter generation unit 17 in FIG. 27, and B is a diagram illustrating another specific example of the deformation parameter generation unit 17 in FIG. 27. The figure which shows the function structural example of 2nd Example of this invention. The figure which shows the function structural example of 3rd Example of this invention. The figure which shows the function structural example of 4th Example of this invention. The figure which shows the function structural example of 5th Example of this invention. FIG. 3 is a diagram illustrating a functional configuration example of a lossless encoding unit 18. The figure which shows the function structural example of the lossless decoding part. A is a diagram illustrating an example of a functional configuration of a prediction error generation unit, and B is a diagram illustrating an example of a functional configuration of a prediction synthesis filter. The figure which shows the function structural example of 12th Example of this invention. The figure which shows the function structural example of the modification of 12th Example of this invention. The figure which shows the function structural example of 13th Example of this invention. The figure which shows the function structural example of 14th Example of this invention. The figure which shows the function structural example of 15th Example of this invention. The figure which shows the function structural example of 16th Example of this invention. The figure which shows the example of the guarantee local decoding signal 11. FIG. A is a figure which shows the other example of the guarantee local decoding signal 11, and B is a figure which shows the example of the code group of the exception code Ige and the digit code Ig. The figure which shows the function structural example of 17th Example of this invention.

Claims (20)

A decoding method for reproducing a sample of a digital signal every frame,
(A) decoding an input code which is a code related to the amplitude of the digital signal to generate a plurality of sets of reversible data of bit strings straddling samples of the provisional digital signal at their respective bit positions;
(B) inserting an arbitrary value into a plurality of sets of missing bits of the reversible data and reproducing it as a provisional digital signal;
(C) decoding auxiliary information, which is information relating to the spectral envelope of the digital signal, to obtain a decoded spectral envelope;
(D) correcting the provisional digital signal to obtain a decoded digital signal so that a spectrum envelope of the provisional digital signal approaches the decoded spectrum envelope;
A decoding method characterized by comprising: The decoding method according to claim 1, comprising:
Step (d) above is
(D-1) obtaining a spectral envelope of the provisional digital signal;
(D-2) obtaining a decoded digital signal by giving the provisional digital signal the inverse characteristic of the spectrum envelope of the provisional digital signal and the characteristic of the decoded spectrum envelope;
The decoding method characterized by comprising by these. A decoding method according to claim 2, comprising:
The step (d-2)
(D-2-1) converting the spectral envelope of the provisional digital signal into a linear prediction cepstrum coefficient Ca;
(D-2-2) converting the decoded spectral envelope into a linear prediction cepstrum coefficient Cb;
(D-2-3) converting the linear prediction cepstrum coefficient Cb from the linear prediction cepstrum coefficient Cb into a linear prediction coefficient;
(D-2-4) obtaining a decoded digital signal by performing linear prediction synthesis using the linear prediction coefficient on the provisional digital signal;
The decoding method characterized by comprising by these. The decoding method according to claim 2 or 3, wherein
The step (d-2)
If the error between the decoded spectral envelope and the spectral envelope of the provisional digital signal is within a predetermined value, the provisional digital signal is a decoded digital signal;
Further including
A decoding method characterized by the above. A decoding method according to any one of claims 1 to 4,
If the bits in the decoded digital signal are different from the bits obtained from the plurality of sets of data, replacing the bits in the decoded digital signal with bits obtained from the plurality of sets of data;
Further including
A decoding method characterized by the above. A decoding method for reproducing a sample of a digital signal every frame,
(A) decoding an input code which is a code related to a prediction error of the amplitude of the digital signal to generate a plurality of sets of reversible data of bit strings straddling samples of a provisional prediction error waveform at their respective bit positions;
(B) inserting an arbitrary value into a plurality of sets of missing bits of the reversible data to reproduce as a provisional prediction error waveform ;
(C) decoding auxiliary information , which is information relating to the spectral envelope of the digital signal, to obtain a decoded spectral envelope;
(D) as the spectral envelope of the decoded decoded digital signal approaches the decoded spectrum envelope, to obtain a prediction error waveform by correcting the provisional prediction error waveform,
(E) obtaining the decoded digital signal by giving the characteristic of the decoded spectral envelope to the prediction error waveform ;
A decoding method characterized by comprising: The decoding method according to claim 6, wherein
Step (d) above is
(D-1) obtaining a provisional decoded signal by giving the characteristics of the decoded spectrum envelope to the provisional prediction error waveform ;
(D-2) obtaining a spectrum envelope of the provisional decoded signal;
(D-3) obtaining a prediction error waveform by giving the provisional decoded signal the inverse characteristic of the spectrum envelope of the provisional decoded signal;
The decoding method characterized by comprising by these. The decoding method according to claim 6, wherein
Step (d) above is
(D-4) obtaining a spectrum envelope of the provisional prediction error waveform ;
(D-5) a step of obtaining a prediction error waveform giving an inverse characteristic of a spectral envelope of the provisional prediction error waveform to the provisional prediction error waveform,
The decoding method characterized by comprising by these. The decoding method according to claim 8, comprising:
The step (d-5)
If the flatness of the provisional prediction error waveform is within a predetermined value, the step of setting the provisional prediction error waveform as a prediction error waveform ;
Further including
A decoding method characterized by the above. A decoder for reproducing a sample of a digital signal every frame,
(A) means for decoding the input code, which is a code related to the amplitude of the digital signal, to generate a plurality of sets of reversible data of bit strings straddling samples of the provisional digital signal at their respective bit positions;
(B) means for inserting an arbitrary value into a plurality of sets of missing bits of the reversible data and reproducing it as a provisional digital signal;
(C) means for decoding the auxiliary information, which is information relating to the spectral envelope of the digital signal, to obtain a decoded spectral envelope;
(D) means for correcting the provisional digital signal to obtain a decoded digital signal so that a spectrum envelope of the provisional digital signal approaches the decoded spectrum envelope;
A decoder comprising: The decoder according to claim 10, comprising:
Said means (d)
(D-1) means for obtaining a spectral envelope of the provisional digital signal;
(D-2) means for obtaining a decoded digital signal by giving the provisional digital signal the inverse characteristic of the spectrum envelope of the provisional digital signal and the characteristic of the decoded spectrum envelope;
A decoder comprising: The decoder according to claim 11, comprising:
The means (d-2)
(D-2-1) means for converting the spectral envelope of the provisional digital signal into a linear prediction cepstrum coefficient Ca;
(D-2-2) means for converting the decoded spectral envelope into a linear prediction cepstrum coefficient Cb;
(D-2-3) means for converting a value obtained by subtracting the linear prediction cepstrum coefficient Ca from the linear prediction cepstrum coefficient Cb into a linear prediction coefficient;
(D-2-4) means for obtaining a decoded digital signal by performing linear prediction synthesis using the linear prediction coefficient on the provisional digital signal;
A decoder comprising: The decoder according to claim 11 or 12, comprising:
The means (d-2)
When the error between the decoded spectral envelope and the spectral envelope of the provisional digital signal is within a predetermined value, the provisional digital signal as a decoded digital signal;
Further including
A decoder characterized by that. A decoder according to any of claims 10 to 13,
Means for replacing the bits in the decoded digital signal with the bits obtained from the plurality of sets of data if the bits in the decoded digital signal are different from the bits obtained from the plurality of sets of data;
Further including
A decoder characterized by that. A decoder for reproducing a sample of a digital signal every frame,
(A) means for decoding the input code, which is a code related to the prediction error of the amplitude of the digital signal, and generating a plurality of sets of reversible data of bit strings straddling samples of the provisional prediction error waveform at their respective bit positions;
(B) means for inserting an arbitrary value into a plurality of sets of missing bits of the reversible data and reproducing it as a provisional prediction error waveform ;
(C) means for decoding the auxiliary information , which is information relating to the spectral envelope of the digital signal, to obtain a decoded spectral envelope;
(D) as the spectral envelope of the decoded decoded digital signal approaches the decoded spectral envelope, means for obtaining a prediction error waveform by correcting the provisional prediction error waveform,
(E) means for obtaining the decoded digital signal by giving the characteristic of the decoded spectral envelope to the prediction error waveform ;
A decoder comprising: The decoder according to claim 15, comprising:
Said means (d)
(D-1) means for providing a provisional decoded signal by giving the characteristics of the decoded spectrum envelope to the provisional prediction error waveform ;
(D-2) means for obtaining a spectrum envelope of the provisional decoded signal;
(D-3) means for obtaining a prediction error waveform by giving the provisional decoded signal the inverse characteristic of the spectrum envelope of the provisional decoded signal;
A decoder comprising: The decoder according to claim 15, comprising:
Said means (d)
(D-4) means for obtaining a spectrum envelope of the provisional prediction error waveform ;
(D-5) means for obtaining a prediction error waveform giving an inverse characteristic of a spectral envelope of the provisional prediction error waveform to the provisional prediction error waveform,
A decoder comprising: The decoder according to claim 17, comprising:
The above means (d-5)
If the flatness of the provisional prediction error waveform is within a predetermined value, the provisional prediction error waveform as a prediction error waveform ,
Further including
A decoder characterized by that.   A decoding program for causing a computer to execute each step of the decoding method according to claim 1.   A computer-readable recording medium on which the decoding program according to claim 19 is recorded.

Images