JPH10260695A - Speech signal encoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To decrease the encoding quantity in a high frequency band and to perform more efficient encoding by handling frequency bands together as one bundled band and omitting the recording of a mantissa as to the bundled band. SOLUTION: Block data of respective audio blocks are formed of digital data on the frequency axis for each prescribed period obtained by the MDCT. The obtained frequency data obtain a signal indicating the intensity by 1000 fine bands having specific intervals. In this case, intensity signals in respective bands are handled together by approximately 3 to 12 bands in order to reduce the data amount while preventing deterioration of sound quality as to high frequencies at which auditory influence is small. Namely, bands of high frequency are bundled and handled together. Then audio blocks are reused and reproduced to expand a recording and reproduction time per frame.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio signal encoding apparatus used for digitally storing an audio signal and the like, and more particularly to an apparatus capable of performing highly efficient encoding.

[0002]

2. Description of the Related Art Conventionally, an analog information signal has been encoded and digitally recorded. For example, an audio signal is converted into digital data by a CD (Compact Discharge).
In c) and MD (Mini Disc), analog audio signals are digitally recorded.

Further, it has been attempted to use a semiconductor memory as an audio recording device in a telephone with an answering machine function and digitally record an audio signal in this memory. Audio recording using a semiconductor memory has been increasingly used because it is easy to record and reproduce and is also convenient for miniaturizing the device. Furthermore, as the capacity and cost of semiconductor memories have been increasing, the possibility of using semiconductor memories for recording music and the like has been proposed.

In such digital recording of an audio signal, when an analog audio signal is converted into a digital signal, encoding is performed so that the original sound can be reproduced as faithfully as possible and the amount of digital data to be recorded is as small as possible. It is desired to do. In particular, in the case of audio recording using a semiconductor memory, although the cost of the semiconductor memory has been reduced,
Since a memory having a large recording capacity is expensive, it is necessary to perform encoding as efficiently as possible to reduce the amount of data.

[0005] When encoding a speech signal, a pulse code modulation (PCM) which indicates the instantaneous value of the time-series speech signal as a digital value is used.
de Modulation) system is generally used. However, the PCM method has a low encoding efficiency, and requires a higher sampling frequency Fs in order to prevent deterioration of reproduced sound.
If the number of bits allocated to the sampling value is increased, the data amount increases and the recording time is not sufficient. For example, when the sampling frequency Fs is set to 8 KHz and 8-bit quantization is performed, the data amount becomes 64 Kbps and 32
Even if an Mbit memory is used, the recording time is 8 minutes and 44 seconds, and can be used only for simple voice memos.
Conversely, the data amount can be reduced by lowering the sampling frequency Fs and the number of allocated bits, but the sound quality during reproduction is significantly reduced.

Therefore, a technique has been proposed in which audio data is converted into data on the frequency axis by using a technique such as discrete cosine transform for encoding the audio signal, and is encoded as data for each frequency band. According to this method, the decrease in fidelity due to rounding of numerical values is small, and the influence of errors due to digitization can be reduced. In other words, when an analog audio signal is digitized at a predetermined sampling period and used as data on the time axis, if there is an error in each sampling, a specific frequency component will appear. Therefore, unnecessary frequency components are generated by the digital processing, which becomes noise and the sound quality is easily deteriorated. On the other hand, if the data is converted on the frequency axis, deterioration of sound quality can be minimized only by changing the balance of each frequency band. Therefore, by converting the digital data to data on the frequency axis, the number of data can be further reduced and highly efficient encoding can be performed.

[0007]

Here, in the encoding using such orthogonal transform, level data (or a difference thereof) is obtained for each frequency band. This reduces the number of bits allocated for each frequency band. However, in order to maintain sound quality, the number of bits allocated to the intensity level in each band cannot be reduced so much. Therefore, by reducing the number of bits for low-frequency and high-frequency bands that have little effect on sound quality,
The amount of data as a whole is reduced. But,
It is better that the data amount is as small as possible, and high-efficiency coding for further data reduction is required.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and has as its object to provide a signal coding apparatus capable of reducing the amount of codes in a high frequency band and performing more efficient coding.

[0009]

In order to achieve the above object, the present invention relates to an encoding apparatus used for digitally storing an audio signal, which divides the audio signal into a number of frequency bands, The band level is represented by an exponent and a mantissa, and a plurality of frequency bands are collectively treated as one binding band for a higher frequency band above a certain frequency, and the recording of the mantissa for this binding band is omitted. It is characterized by doing.

[0010] The level of each frequency band is expressed using the exponent and the mantissa, and the recording of the mantissa is omitted in the binding band on the high frequency side, thereby reducing the amount of stored data. In the binding band on the high frequency side, the phase is not important, and since a plurality of fine frequency bands are bound, the frequency number difference between adjacent binding bands is large, so that the waveform synthesizing action is not important. Therefore, by omitting the recording of the mantissa of the binding band on the high frequency side, the data amount can be reduced without substantially affecting the sound quality. The exponent indicating a plurality of frequency bands can be expressed by encoding the exponent itself.However, by using a method of indicating the exponent of each frequency band and the binding band as difference data, further reduction of the data amount can be achieved. It becomes possible to plan.

Further, in the present invention, in the above-described audio signal encoding apparatus, the absolute value of the level of each frequency band to be bound is averaged to obtain the level of the bound band. On the high frequency side, when the level of the neighboring band is high, the masking effect that the sound of the other bands becomes difficult to hear is strong. Therefore, even if the binding band is expressed by the average value of the absolute values of the levels of the respective frequency bands to reduce the data amount, it is possible to slightly suppress the deterioration of the sound quality.

[0012]

Embodiments of the present invention (hereinafter referred to as embodiments) will be described below with reference to the drawings.

[Outline of Encoding Method of Information Signal] In the present embodiment, as an information signal, for example, an analog audio signal such as a human voice, a natural sound, or a musical instrument sound is to be encoded. Then, the audio signal is sampled at a predetermined frequency and converted into digital data. Further, the digital data on the time axis is converted into the above-mentioned discrete cosine transform (MDCT: Modified D).
It is transformed into data on the frequency axis using iscrete cosine transform.

FIG. 1 shows a hierarchical structure of audio recording data in the encoding system of the present embodiment.

In the encoding system of the present embodiment, the minimum unit of audio recording is an audio block, and block data of each audio block is formed from digital data on the frequency axis for every 10 ms period obtained by the above-mentioned MDCT. Is done. Specifically, first, frequency data obtained every 10 ms period by MDCT is DC to
A signal indicating the intensity of each of the 100 fine bands at 50 Hz intervals up to 5 KHz is obtained. In the present embodiment, in order to reduce the amount of data while preventing the sound quality from deteriorating, the intensity signal of each band is reduced to about 3 to 12 bands in a high-frequency range (for example, a frequency band of 1 KHz or more) that has little effect on hearing. Handle collectively. That is, a plurality of bands on the high frequency side are handled by binding the bands. In addition, 25 Hz or less is omitted, and a total of 100 fine bands are divided into 25 bands #
Compress the intensity signal into minutes. The band # indicates a hand for recording. Further, the spectrum intensity for each band # is indicated by an exponent and a mantissa as described later, and is data of one audio block.

As shown in FIG. 1, one audio block is composed of 4-bit initial data, a 56-bit difference index and a 20-bit mantissa, and the initial data is an index of a 75 Hz band spectral intensity. Is shown. The difference index is 2 following the 75 Hz band.
The spectrum intensities of 24 bands # are represented by expressing each spectrum intensity of 4 bands # as differential data and displaying 3 bands # with 7 bits. The mantissa represents the phase of each exponent, and the mantissa is not displayed for the binding band on the high frequency side.

The three audio blocks constitute one frame of audio information. In addition, one frame is set to a fixed length of 2 ⁿ bits (for example, n = 8, that is, 256 bits), so that a synchronization signal for recognizing the frame can be omitted. Also, 2 ⁿ frames (for example,
(n = 6, that is, 64 frames) constitutes one superframe.

One frame includes audio block data of three audio blocks (AB) each consisting of 80 bits, and displays audio information for a minimum of 30 ms. Here, when there is a predetermined similarity between consecutive blocks, or when the exponential peak level of an audio block continues to be a predetermined value or less, recording of some audio blocks from the second block is omitted.

The number of uses of each of the three audio blocks is recorded by providing, for example, 15-bit mode data in the same frame. At the time of reproduction, the same audio block is used a plurality of times according to the number of times of use indicated in the mode data to repeat the reproduction.

By reusing and reproducing the audio block in this manner, the recording / reproducing time per frame can be extended up to 17.3 times (520 ms) at the maximum. In an audio signal or the like, a fixed approximate waveform often continues to some extent, and in such a case, recording of similar data can be omitted. Also,
When the audio level is low, the audio block can be replaced with the previous audio block.

In order to shorten the reproduction time, etc.
Depending on the setting, at the time of reproduction, an audio block recorded less than the number of repeats indicated in the mode data may be reproduced.

In one frame of 256 bits, 240 bits are allocated to three audio blocks AB0 to AB2, 15 bits are allocated as mode data, and the remaining 1 bit is allocated to a subcode. The subcode is composed of one superframe and represents constant information in 64 bits. Therefore, one superframe forms one unit of audio recording, but the superframe is related only to this subcode, and the recording of audio data itself is completed in frame units.

In the subcode for each superframe,
It is possible to record the date and time of audio recording, a phrase number corresponding to a header or track number, and the like.
Also, by setting the number of frames constituting one super frame to 2 ⁿ similarly to each frame (however, it is not necessary to make n coincide between the frame and the super frame), only the subcode is read at the time of reproduction, and the desired It is easy to find the super frame.

By encoding the audio signal according to the above-described method, the bit rate is 8.5 Kb in this embodiment.
Highly efficient encoding of ps or less can be performed. For example, 32
Since an audio of 3932 seconds (65 minutes 32 seconds) or more can be recorded in an M-bit memory, it is most suitable for IC audio recording using a semiconductor memory and audio recording on a floppy disk.

The specific encoding method and device configuration will be described below.

[MDCT] FIG. 2 conceptually shows conversion by the MDCT method. First, an analog audio signal is sampled at 10 KHz. In the MDCT method, digital data of 100 μs obtained by converting the digital data is divided every 10 ms, and the signal of each period is
It is converted into a frequency component signal by FT. In practice, to ensure the continuity of digital data for successive periods,
FFT is performed on the audio signal for a period of 20 ms, and the signal is set so as to overlap each other by 50% in each of the adjacent periods.
Obtain audio block data for each audio block. The following equation (1)
The window coefficient W (m) of the cosine wave is shown. This window coefficient is multiplied by the original signal x (m) of each period,
By performing an FFT analysis on the original signal x (m), a spectrum Z (k) is obtained for each of the fine bands represented by the following equation (2).

[0027]

(Equation 1)

(Equation 2) In the expressions (1) and (2), M is the number of samplings (200 in this embodiment), and k is a fine band number (1 to 100, but can be omitted from 1 to 79).

In this embodiment, 200 points of data x of the audio signal every 20 ms in the MDCT.
By calculating from (m) by the above equations (1) and (2), windowing with a cosine wave and FFT are performed simultaneously. The calculation is performed precisely with 16-bit precision.

[Bundling of Bands] By the above-described MDCT, D is applied to each audio block at intervals of 50 Hz.
Data on the phase and amplitude of 100 fine bands from C to around 5 KHz can be obtained. Here, the sound speed is 340 m / s
Considering that it is relatively slow, the wavelength of 1 KHz is 34c
m, which is equivalent to a value similar to the diameter of a human head. For this reason, the phase information is not important in a frequency band of about 1 KHz or more. Therefore, the recording of the mantissa necessary for expressing the phase for the band of 1 KHz or more is omitted. Also, originally, on the high frequency side, a so-called masking effect in which only frequency components exhibiting the highest intensity between adjacent bands can be heard. Therefore, in the high frequency range, even if the adjacent 2 to 12 bands are bundled, the deterioration of the sound quality is slight, and a plurality of bands are bundled and expressed by the average level of the intensity.

FIG. 3 shows the band numbers of 25 bands # in total and the assigned frequency bands obtained by compressing 100 fine bands. The information of 25 Hz or less is omitted as shown in FIG. 3 in consideration of the overall performance, and a total of 78 fine bands from 75 Hz to 3975 Hz are compressed into 25 band intensity signals. Band numbers 0 to 15 (75 Hz to 825 Hz) are all recorded as fine bands, and band numbers 16 to 18 bind three continuous bands, respectively, and band numbers 19 to 21.
Then, six consecutive bands are bound. The band numbers 22 to 24 on the higher frequency side, which hardly affect the audibility even if they are bound, are represented by binding 12 consecutive bands. As described above, in the present encoding scheme, each band # as shown in FIG. 3 is set corresponding to the frequency, that is, the musical scale, with six bands per octave as a guide.

[Exponentialization of Spectrum Intensity Signal] In the encoding method of the present embodiment, the spectrum intensity signal in each of the 25 bands # is expressed using an index b and a mantissa a. FIG. 4 shows the spectrum intensity of each band as an index. Also, to further reduce the amount of data,
For a 75 Hz band # with band number 0, the index itself of the spectrum intensity of the band is represented by 4 bits,
This is used as initial data. The index of the spectrum intensity and the 4-bit initial data have a correspondence as shown in FIG. 5, and the index b (0 to 1) of the intensity of each band.
The corresponding initial data is determined according to the value of 5).

The remaining 24 bands # are coded by calculating the difference from the adjacent low band from the low band side.
Record this. FIG. 6 shows the correspondence between the change value of the index, which is the difference data, and the difference index code.
The amount of change in exponent between +2 is represented by five values from 0 to 4 using an exponent code. At this time, it is preferable to perform encoding so that a more accurate value can be calculated using a decoder so that errors are not accumulated. Also for the binding bands # of band number 16 or more, the index of the average value of each binding band # is shown, and the difference between the wide band bands is obtained.

Each difference data for 24 bands # of band numbers 1 to 24 is indicated by difference index data in FIG. 6, and is grouped into three bands #, and these three bands # are divided into seven based on the following equation (3). Expressed in bits. FIG. 7 shows the correspondence between the recording codes of the difference data for the three bands # and the five-value difference index codes E0, E1, and E2 for each band #.

[0034]

Equation 3] recording code ^{2 7 = 5 0 · E0 +} 5 1 · E1 + 5 2 · E2 ··· (3) Thus, the difference data as the 5 values, and by putting together the three bands #, 7-bit × (24 / 3), that is, 56 bits represent differential data for 24 bands, and band numbers 1 to 24 can be represented by 2.33 bits per band #.

The mantissa a represents the phase, and its coefficient is “+1”, “0”, “−” with respect to the exponent represented by b.
1 "or" +1 "or" -1 ".

As shown in FIG. 8, for the six low band fine bands # 0 to # 6, three values [−1,
The mantissa code [0, 1, 2] is assigned according to the mantissa of [0, +1]. As shown in FIG. 9, the three-valued mantissa code is
Bands # are grouped together, and mantissa codes M0 and M of each band #
1 and M2 are collectively indicated by 5 bits. As a result, a total of 1 band for low band 6 band # that has a large effect on sound quality
A ternary mantissa is indicated by 0 bits (5 bits × (6 bands # / 3)), and more accurate difference data is represented. In addition, for 10 bands # (band numbers 6 to 15 in FIG. 3) subsequent to the 6 low bands, 1 bit of only the polarity is recorded as a mantissa. In the 1-bit polarity display, “1” indicates +1 and “0” indicates −1.

For a band number 16 or higher, since a plurality of bands are bound, the frequency difference between adjacent bands # is large, and no waveform synthesizing action occurs between adjacent bands #. Therefore, in these bands #, the recording of the mantissa is omitted, and the mantissa to be recorded is (5 bits × 2) + (1 bit × 10 band #), for a total of 20 bits.

According to the above-described coding method, initial 4 bits indicating the lowest band, a difference index for 24 bands #, and mantissa data of 20 bits are obtained.
One audio block is composed of bits.

[Repeat of Audio Block] When considered in units of 10 ms or 20 ms, the audio signal often has the same waveform, and even if the audio block is repeatedly reproduced a plurality of times, there is little deterioration in the reproduced sound quality. . Therefore, in the present embodiment, recording of a predetermined audio block is omitted, and the audio block recorded at the time of reproduction is repeatedly reproduced. In addition, by recording the number of repeats together with the audio block as mode data, recording of audio block data having the same or similar content or a low exponent level of the intensity is omitted, the recording data amount is reduced, and the recording / reproduction time is reduced. Can be extended.

FIG. 10 shows the structure of mode data to which 15 bits are assigned in one frame. In the present embodiment, the number of reuses (the number of repeats) is determined in consideration of the case where the audio block data AB is reused alone, the case where two continuous audio blocks are reused, and the case where they are combined. As shown in FIG. 10, "AB0", "AB0 to AB1", "AB1",
The number of repeats of each of “AB1 to AB2” and “AB2” is specified by 3 bits. As a result, it is possible to record and reproduce sound corresponding to a maximum of 0.52 seconds in one frame data of 256 bits.

The reuse of the audio block can be performed when all the bits match between the continuous audio blocks. In addition, when a predetermined similar relationship is observed, or when the index peak of the band intensity is constant. It can be executed in the following cases.

In the present embodiment, the following conditions (i) to (iii) are assumed as conditions for reusing an audio block, which condition is to be adopted, and a specific judgment criterion in that case. It can be set arbitrarily from outside during recording.

(I) All bits match between successive audio blocks; (ii) Index and mantissa from low band # to predetermined band # match between successive audio blocks; (iii) Index of audio block Peak below a certain value: FIG. 11 shows specific reuse conditions of these audio blocks, and also shows a 3-bit repeat mode associated with each condition. This repeat mode is
By recording in the subcode, the user can refer to this at the time of reproduction or the next recording, and can know the relationship between the repeat mode and the sound quality.

As a condition for reuse, the repeat mode “0”
By setting “all bits of the audio block match” corresponding to “00”, it is possible to eliminate the duplicate recording of the same continuous audio block without deteriorating the reproduction sound quality at all.

In practice, when the difference in data between audio blocks is small, even if the reproduction is performed using one of the audio blocks, the sound quality is hardly reduced. Therefore, a similar criterion for adjacent audio block data can be selected, and a similar criterion is set higher when a good sound quality is desired. Can be set to lower. Since the effect on the sound quality decreases as the frequency band is closer to the higher band # side, the similar criterion is seven levels corresponding to the repeat modes “001” to “111” in FIG. 11 (exact match “000”). If you include this, there will be a total of eight stages.) According to the stage set in this way, low band #
Whether the audio block is reused is determined by checking the match between the exponent and the mantissa of any of the bands # 0 to 23-11. Deterioration of the sound quality can be prevented if a condition up to band # on the higher frequency side is used as a condition, and if similarity is determined by matching up to a certain band #, the number of audio blocks to be omitted increases, so recording is performed. The time can be substantially extended.

The similar reference of data between audio blocks can be set not only at the time of audio recording but also at the time of reproduction. At the time of reproduction, it is fundamental to reproduce faithfully with the number of repeats indicated by the above-described mode data. . Therefore, in such a case, by setting a reference at the time of reproduction and reducing the number of times of repeat reproduction of the audio block, the speech speed is increased without changing the pitch, so that the reproduction time can be reduced without being difficult to hear. .

Referring to the peak level of the index of the recorded audio block, if the peak level is low, the actual number of repeats is reduced from the number of repeats specified by the mode data. Thereby, the reproduction time can be shortened. If the reference of the peak level is set low, the deterioration of the reproduced sound quality is extremely small, and if the reference of the peak level is set high, the sound quality is slightly lowered but the reproduction time can be shortened. For example, in the case of reproducing a voice recording of a conference or a lecture, it is often required to omit unnecessary periods during which the speaker or the speaker is not speaking and to shorten the reproduction time. Therefore, if the reference value of the peak level of the index of the audio block is set at the time of reproduction,
For example, when the speaker is interrupted and background noise continues, the reproduction of audio block data corresponding to the background noise can be omitted, and the reproduction time can be further shortened while preventing a decrease in reproduction sound quality. .

When recording audio while reusing audio blocks as described above, the recording time increases or decreases depending on the state of the input audio signal. However, in many cases, a recording device generally performs recording while assuming a period until the end of voice recording, and it is desirable that a recordable time be clear. Therefore, the function of reusing the audio block as described above changes the recordable time depending on the similarity, strength, and the like of the input voice, and thus it is sometimes difficult to use the function from the viewpoint of determining the recording time.

Therefore, the criterion value for audio block reuse is controlled according to the recording capacity consumption of a recording device such as a semiconductor memory. That is, the memory consumption is monitored, the recording capacity consumption is sequentially compared with the planned consumption determined by the planned recording period, and the recording time is changed by changing a reference value such as a similar reference or an exponent level reference. Can be adjusted. If the recording capacity consumption is large, relax the similarity and exponent level standards and increase the number of audio block repeats.If the consumption is small, set the above standard higher to ensure higher recording sound quality. Control automatically. As a result, it is possible to record sound with high sound quality within a desired time, and the usability as a sound recording device is improved.

[Subcode] A subcode in which one superframe is composed of 64 frames and a small number of bits (for example, one bit) is allocated in one frame is one information display unit in one superframe. And 64
It has enough information display capability. When one superframe is composed of 32 frames, two bits are allocated to each frame as a subcode.

FIG. 12 shows the structure of a 64-bit subcode. The first 12 bits are the phrase number, the next 3 bits are the above-described repeat mode, the recording date and time of 33 bits, and the last 16 bits are for error detection. CRC. The date and time of recording are recorded in binary format, and the year is indicated by the lower two digits of the Christian era using 7 bits. It should be noted that since the second is recorded every superframe, the superframe advances at intervals of about 1 to 30 seconds in each superframe. The phrase number data is used as a marking also represented as a track mark for cueing. The phrase numbers can be assigned consecutive numbers automatically or arbitrarily, and the maximum density is 64 phrases per Mbit. If encoded data is recorded in a large-capacity memory, the number of phrases to be added increases. However, since 12 bits are allocated and a number of about 4000 can be indicated, a sufficient information display capability is provided.

[Structure of Encoding Apparatus] FIG. 13 shows a schematic structure of a speech encoding processing apparatus for executing the above-described encoding processing. Unnecessary low-frequency and high-frequency signals are removed from the input analog audio signal in advance and supplied to the analog / digital (A / D) conversion unit 10. A / D converter 10
Samples the supplied analog audio signal at a sampling frequency Fs of 10 KHz and converts it into 12-bit digital data. The MDCT processing unit 12
The 12-bit digital data from the D conversion unit 10 is multiplied by a predetermined coefficient of MDCT, and a total of 100
The level and phase of the intensity signal for each of the fine bands are detected. Among the detected fine bands, the high-frequency bands having a band number of 16 or more are banded for every 3 to 12 bands by calculating the average value of the absolute value of the level of each band. Compress to

The exponent processing unit 14 encodes the level of the obtained 25-band # intensity signal into an exponent, a difference exponent, and a mantissa.

The ABR creating section 16 creates 80-bit recording audio block data ABR based on the obtained exponents, difference indices and mantissas, and outputs them to the synthesizing section 22 and the mode determining section 18, respectively. Mode determination unit 18
Determines whether to reuse the audio block based on the similarity between the audio blocks and the peak level of the exponent based on the audio block reuse condition set from outside. Further, the number of repeats of the predetermined audio block data obtained from the result of this determination is indicated as 15-bit mode data, which is output to the synthesizing unit 22.

The subcode generator 20 converts the recording time and, if a phrase number is assigned, the phrase number into binary data to generate a subcode.
Output to the combining unit 22.

The synthesizing unit 22 creates one frame of 256 bits with three audio blocks according to the subcode, the mode data, and the mode data, and outputs it to a recording device 24 such as a semiconductor memory as a bit stream.

[Recording in Recording Device] Output data from the synthesizing unit 22 is recorded in the memory in frame units. By setting one frame to 2 ⁿ bits as described above,
The frame can be identified by looking at the address of the memory.
FIG. 14 shows the relationship between the address of the 1-Mbit memory and the frame. In FIG. 14, the lower 8 bits of the memory address correspond to the bit number in the frame, the upper 6 bits indicate the frame number, and the upper 6 bits indicate the superframe number.

When one frame has 256 bits, each frame data is stored in the lower n = 8 bits of the address. That is, from the first bit to the last bit of each frame, the lower n bits of the memory address “00...
00 (all 0) ”to“ 11... 11 (all 1) ”. As a result, if the frame number of the address is specified, the corresponding one frame data can be immediately identified, and the synchronization signal for identifying the frame becomes unnecessary.

[Reproduction] In reproducing the audio data recorded in the recording device 24, the decoder decodes the exponent and the mantissa of the 25th band # for each read audio block of one frame, and further decodes each band # Inverse DCT is performed using the signal strength to inversely convert data on the frequency axis into digital data on the time axis. For the subcode, an error is checked using the CRC data, the phrase number, the repeat mode, and the recording date and time are decoded and output.

As described above, when the audio block reuse condition is set at the time of reproduction, the decoder determines the number of repeat reproductions of the audio block based on the setting, and reuses the audio block. .

[0061]

According to the present invention, when an audio signal is divided into a large number of bands and each band is represented by an exponent and a mantissa and encoded, a plurality of bands on the high frequency side which have little effect on the sound quality are bound. The recording of the mantissa of the binding band is omitted. As a result, the amount of stored data can be reduced without deteriorating sound quality, and high coding efficiency is realized.

Further, by expressing the binding band by the average value of the absolute values of the respective bands, it is possible to further reduce the amount of stored data without deteriorating the sound quality.

[Brief description of the drawings]

FIG. 1 is a diagram showing a data structure in an encoding method of the present invention.

FIG. 2 is a diagram conceptually showing an MDCT process of the present invention.

FIG. 3 is a diagram showing 25 bands # obtained by partially binding and compressing the fine bands of the present invention, and a frequency range assigned to each band #.

FIG. 4 is a diagram showing the spectrum intensity of each band obtained by the MDCT of the present invention as an index.

FIG. 5 is a diagram showing a correspondence between an index of spectrum intensity and initial data according to the present invention.

FIG. 6 is a diagram showing a correspondence between an index difference and a difference index code according to the present invention.

FIG. 7 is a diagram showing a correspondence relationship between a 7-bit recording code representing a difference exponent code for three bands # of the present invention and a difference exponent code of each band #.

FIG. 8 is a diagram showing the correspondence between mantissas and mantissa codes according to the present invention.

FIG. 9 shows 5 representing a mantissa code for three bands # of the present invention.
FIG. 4 is a diagram showing a correspondence relationship between a bit and a recording code.

FIG. 10 is a diagram showing a configuration of mode data of the present invention.

FIG. 11 is a diagram showing a reuse condition of an audio block and a 3-bit repeat mode corresponding to the condition according to the present invention.

FIG. 12 is a diagram showing the structure of a 64-bit subcode according to the present invention.

FIG. 13 is a diagram showing a configuration of a speech encoding processing device of the present invention.

FIG. 14 is a diagram showing addresses of a memory for storing encoded audio data of the present invention.

[Explanation of symbols]

10 A / D conversion unit, 12 MDCT processing unit, 14 exponentiation processing unit, 16 ABR creation unit, 18 mode determination unit, 20 subcode creation unit, 22 synthesis unit, 24 recording device.

Claims (2)

[Claims] An encoding device used for digitally storing an audio signal, wherein the audio signal is divided into a number of frequency bands, and the level of each frequency band is represented by an exponent and a mantissa, and a constant frequency is used. Regarding the frequency band of the above high frequency signal,
An audio signal encoding apparatus, wherein a plurality of frequency bands are collectively treated as one binding band, and recording of a mantissa for the binding band is omitted. 2. The audio signal encoding apparatus according to claim 1, wherein, for the binding band, an absolute value of a level of each frequency band to be bound is averaged to obtain a level of the binding band. Encoding device.