JPH08137494A - Sound signal encoding device, decoding device, and processing device - Google Patents

Abstract

PURPOSE: To provide a sound signal encoding device, a sound signal decoding device, and a sound signal processing device in which a sound signal is inputted, redundancy of a sound signal is reduced, the signal is encoded, encoded data in which redundancy is reduced is inputted and it is decoded to a sound signals. CONSTITUTION: An input sound signal is divided into plural frequency components by a band divider 11, and code quantity corresponding to each frequency band of input sound signals is allowed by a code quantity allotting device 14. Next, after the next coefficient is predicted from a coefficient column before it corresponds to a frequency component divided by the band divider 11 by an in-band signal predicting device 12, a frequency component is subtracted from a predicted value and a residual component is obtained, after the residual component obtained by the in-band signal predicting device 12 is quantized by a residual component encoding device 13 corresponding to code quantity allotted by the code quantity allotting device 14, the quantized residual component is encoded.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an acoustic signal coding device,
BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an acoustic signal decoding device and an acoustic signal processing device, and more particularly, to an acoustic signal that inputs an acoustic signal to reduce the redundancy of the acoustic signal for encoding, and to input encoded data with reduced redundancy to decode the acoustic signal. The present invention relates to a signal encoding device, an acoustic signal decoding device, and an acoustic signal processing device.

[0002]

2. Description of the Related Art Conventionally, as an acoustic signal encoding apparatus for efficiently encoding an acoustic signal by utilizing the redundancy of the acoustic signal, by predicting the subsequent waveform by utilizing the acoustic signal of the entire band, Some have reduced the amount of code to be transmitted or recorded. For example, in the DPCM system and the ADPCM system, there is a system in which the code amount is reduced by transmitting a difference value from the value up to immediately before the acoustic signal. In order to effectively use the masking effect of masking components of frequencies relatively close to human hearing, as a method of performing the above method by dividing it into a plurality of frequency bands, for example, ITU-T Recommendation G72.
In 2, after dividing the acoustic signal into four bands,
Some have used the ADPCM method for each component.

On the other hand, in recent years, a system that more positively utilizes the nature of human hearing has appeared. These methods utilize the masking effect to the limit by applying a precise model of the frequency masking effect. For example, I
SO / IEC 11172.3 International Standard (MPEG Audio)
Is to reduce the code amount by dividing the band of the acoustic signal into 32 to 384 bands and then allocating an optimum code amount to each band. In addition, ATSC (Advanced Television Systems Committee)
) Standard sound transmission (Dolby AC3) and MiniDis
The same applies to the c method and the like.

Further, in the past, regarding the audio signal encoding device, the audio signal decoding device and the audio signal processing device, there is disclosed in Japanese Unexamined Patent Publication No. 4-503136, "Short delay conversion coder for high quality audio, decoder and encoder. Decoder "has been reported. The encoder part of this is a signal sampling and quantizing device 101, a signal sampling buffer 102, an analysis window multiplier 103, a digital filter bank 104, a block floating point encoder 105, an adaptive bit allocation device 1.
06, homogeneous quantizer 107 and formatter 109
It is composed by. This signal is obtained by standardizing and quantizing wide band analog audio information into a sample block of a time domain signal by a signal sampling and quantizing device 101, and analyzing a time domain signal sample by an analysis window multiplication device 103 using an analysis window. • Weight blocks. The frequency subband is then defined in response to the analysis window weighted time domain signal sample blocks by the subband unit 108 having the adaptive bit allocation unit 106, and the coding accuracy is adaptive bit allocation unit 106. Generating subband information containing for each subband one or more digital words consisting of an adaptive portion and a non-adaptive portion determined by the formatting device 109; It has the advantage of high-quality digital encoding of wideband analog audio information by configuring it for assembly into a digital output having a format suitable for transmission or storage.

On the other hand, the decoder part of this one is composed of a deformatter 112, a linearization device 113, a digital filter bank 114, a synthesis window multiplication device 115, a signal block duplication / addition device 116 and an analog signal generator 117. . It receives a coded signal to derive a subband block from the coded signal, and a deformatter 112 including an adaptive allocator provides an adaptive bit allocator 10 with a coding accuracy.
Reconstruct the digital word determined by 6. Next, a time domain signal sample block is generated in which each of the time domain signal sample blocks is generated in response to a corresponding one of the subband information blocks derived by the linearizer 113 and the digital filter bank 114. , The synthesis window multiplier 115 weights the time domain signal sample blocks by the synthesis window. Next, the signal block duplication / addition device 11
By removing the window weighting effect by 6, analog
A wideband analog audio information is encoded with digital information by recovering the digitized representation of the audio information and converting the digitized representation by the analog signal generator 117 into an analog format. It has the advantage of reproducing from the quantized signal according to the quantized input signal weighted in the analysis window.

[0006]

However, in these acoustic signal decoding apparatus and acoustic signal processing apparatus, since the acoustic signal is divided into the constant frequency interval and the constant time interval, the code amount is assigned, and therefore There is also a problem that the model also has to be defined as one that gives the sensitivity of human hearing at constant frequency intervals and constant time intervals.

Further, in the official listening test within the standardization group of ISO / IEC, when the acoustic signal is coded using only the prediction so that the method utilizing the auditory characteristics exhibits the highest performance, the signal redundancy is used. However, since it does not correspond to the masking characteristics of the auditory sense, there is a problem that the masking characteristics cannot be used precisely and the optimum coding cannot be performed.

Further, when controlling the code amount or the coding accuracy for each fixed band and time block by using the auditory characteristic as in the conventional case, the auditory characteristic as a function of the bandwidth and the time as described above is used. Need to be modeled. As described above, when the auditory characteristics are modeled, the auditory characteristics in the frequency space and the masking characteristics in the time axis direction are usually described by completely different models and then combined. However, this method is difficult because the temporal change of the acoustic signal and the distribution in the frequency space cannot be completely separated. For example, if the frequency resolution is increased in order to enhance the masking effect in the frequency space, the resolution on the time axis becomes lower based on the uncertainty principle, the time masking effect cannot be used effectively, and conversely the time resolution is increased. And there was a problem that the frequency resolution was lowered.

For example, in the case of prioritizing either time resolution or frequency resolution, the above-mentioned Japanese Patent Laid-Open No. 4-50313.
In the method described in Japanese Patent Laid-Open No. 6-128, when the time resolution is prioritized, 128 samples are set as one frame, and when the frequency resolution is prioritized, 256 samples are set as one frame. However, in such a method, there are only two choices of the time resolution and the frequency resolution, and only the frame unit can be selected. Therefore, the optimum time / frequency resolution according to the signal cannot be obtained. In addition, in the case of a signal that includes a component that changes rapidly with time in one band and a continuous component that has a fine frequency distribution in another band, optimal adjustment of time / frequency resolution in block units is not possible. There was a problem that I could not do it.

With respect to the fine structure of the auditory characteristics on the frequency axis of the acoustic signal, it is necessary to obtain the frequency masking effect of adjacent frequencies such as proximity masking by comparing the frequency components. There is a problem that complicated calculation is required for comparison calculation. Therefore, the acoustic signal encoding device of the present invention divides the acoustic signal into bands, predicts the frequency components within the bands using a linear predictor, and quantizes the residual component of the predicted value and the input value. By outputting the compressed code as a compressed code, wide area masking characteristics can be used by band division and code amount allocation between bands, and proximity masking effect can be efficiently used by controlling the filter characteristics of the predictor. By adaptively controlling the filter characteristics of the predictor to the characteristics of the acoustic signal, it is possible to optimally adjust the time resolution and frequency resolution capability. Furthermore, regarding the acoustic input within the range that the predictor can predict, , The output code can be significantly reduced, and in particular, a waveform containing a sine wave and many harmonic components can be predicted by superimposing the waveforms predicted in each band, and the overall code amount can be calculated. It has an object to provide an acoustic signal encoding apparatus capable of reducing.

Therefore, the acoustic signal decoding apparatus of the present invention can synthesize an acoustic signal on the basis of encoded data whose output code is significantly reduced, and in particular includes a large number of harmonic components with a reduced overall code amount. An object of the present invention is to provide an acoustic signal decoding device capable of synthesizing waveform acoustic signals.

[0012]

According to the first aspect of the present invention,
In order to solve the above-mentioned problems, in an acoustic signal encoding device for inputting an acoustic signal and encoding while reducing the redundancy of the acoustic signal, a band divider that divides the input acoustic signal into a plurality of frequency components, A code amount assigner that assigns a code amount corresponding to each frequency band of the input acoustic signal, and after predicting the next coefficient from the previous coefficient sequence corresponding to the frequency component divided by the band divider, from the predicted value An in-band signal predictor for subtracting the frequency component to obtain a residual component, and a quantum for the residual component obtained by the in-band signal predictor corresponding to the code amount assigned by the code amount assigner And a residual component encoder which encodes the encoded residual component.

According to a second aspect of the present invention, in order to solve the above-mentioned problems, the band divider divides the input acoustic signal into components for each frequency, a polyphase filter bank, and the input acoustic signal for each frequency. A discrete Fourier transformer for separating the input acoustic signal into components for each frequency, a discrete Hartley transformer for separating the input acoustic signal into components for each frequency, a discrete cosine transformer for performing orthogonal transformation on the input acoustic signal to separate into components for each frequency, the input It is characterized in that it is configured by any one of modified discrete cosine converters that perform orthogonal transformation on an acoustic signal to separate it into frequency components.

In order to solve the above problems, the code amount assigner is configured to assign the code amount corresponding to the minimum audible level of the auditory model. In order to solve the above-mentioned problems, an invention according to a fourth aspect is an acoustic signal decoding apparatus for inputting encoded data with reduced redundancy and decoding it into an acoustic signal, wherein the input encoded data is used as a residual component. The residual component decoder to be decoded and the next coefficient is predicted from the previous coefficient sequence of each frequency component, and the predicted value is corrected by the residual component decoded by the residual component decoder to obtain the frequency component. An in-band signal predictor to be generated and a band synthesizer to synthesize an acoustic signal from each frequency component generated by the in-band signal predictor are provided.

In order to solve the above-mentioned problems, the present invention as set forth in claim 5 is an inverse polyphase filter bank in which the band synthesizer synthesizes an acoustic signal from each frequency component generated by the in-band signal predictor, An inverse discrete Fourier transformer that synthesizes an acoustic signal from each frequency component generated by the in-band signal predictor, an inverse discrete Hartley transformer that synthesizes an acoustic signal from each frequency component generated by the in-band signal predictor, An inverse discrete cosine transformer that synthesizes an acoustic signal by performing an inverse orthogonal transform from each frequency component generated by the in-band signal predictor,
It is characterized in that it is configured by any one of an inverse modified discrete cosine transformer that performs an inverse orthogonal transform on each frequency component generated by the in-band signal predictor to synthesize an acoustic signal.

In order to solve the above-mentioned problems, the invention according to claim 6 is the audio signal encoding device according to claim 1, wherein:
When quantizing the residual component in the residual component encoder in association with the code amount assigned by the code amount assigner, from the previous coefficient sequence of the pseudo frequency component pseudo corresponding to the frequency component After predicting the next coefficient, a local decoder for calculating the pseudo frequency component by subtracting the dequantized residual component from the predicted value is provided, and the in-band signal predictor is calculated by the local decoder. The residual component is obtained by subtracting the frequency component divided by the band divider from the predicted value, the residual component is quantized by the residual component encoder, and then encoded into encoded data. According to a fourth aspect of the present invention, there is provided the acoustic signal decoding device, wherein the encoded data is input and decoded into an acoustic signal.

In order to solve the above-mentioned problems, the invention according to claim 7 is constituted by combining the inventions according to claims 1 to 6.

[0018]

According to the first aspect of the present invention, the band divider divides the input acoustic signal into a plurality of frequency components, and the code amount assigner assigns the code amount corresponding to each frequency band of the input acoustic signal. Next, after predicting the next coefficient from the previous coefficient sequence corresponding to the frequency component divided by the band divider by the in-band signal predictor, the frequency component is subtracted from the predicted value to obtain the residual component, After the residual component obtained by the in-band signal predictor is quantized by the difference component encoder in correspondence with the code amount assigned by the code amount assigner, the quantized residual component is encoded.

According to a second aspect of the present invention, in the first aspect of the present invention, the band divider is a polyphase filter bank for separating the input acoustic signal into frequency components, and the input acoustic signal into frequency components. Discrete Fourier transformer for separation, Discrete Hartley transformer for separating input acoustic signal into frequency components, Discrete cosine transformer for performing orthogonal transformation of input acoustic signal into frequency components, and orthogonal transformation of input acoustic signal , And a modified discrete cosine converter that separates each frequency component.

According to a third aspect of the present invention, in the first aspect of the invention, the code amount assigner is configured to assign the code amount corresponding to the minimum audible level of the auditory model. In the invention according to claim 4, the encoded data input by the residual component decoder is decoded into residual components, and the in-band signal predictor predicts the next coefficient from the previous coefficient sequence of each frequency component. , The residual component decoded by the residual component decoder corrects the prediction value to generate a frequency component. Next, the band synthesizer synthesizes an acoustic signal from each frequency component generated by the in-band signal predictor.

According to a fifth aspect of the present invention, in the fourth aspect of the present invention, the band synthesizer synthesizes an acoustic signal from each frequency component generated by the in-band signal predictor. Inverse Discrete Fourier Transform that synthesizes an acoustic signal from each frequency component generated by the in-signal predictor, Inverse Discrete Hartley Transform that synthesizes an acoustic signal from each frequency component generated by the in-band signal predictor, In-band signal Inverse Discrete Cosine Transformer that synthesizes acoustic signal by performing inverse orthogonal transform from each frequency component generated by the predictor, and synthesizes acoustic signal by performing inverse orthogonal transform from each frequency component generated by the in-band signal predictor , Or an inverse modified discrete cosine transformer.

According to a sixth aspect of the present invention, in the acoustic signal encoding apparatus according to the first aspect, the residual component encoder causes the residual component to correspond to the code amount assigned by the code amount assigner by the local decoder. When quantizing, the next coefficient is predicted from the previous coefficient sequence of the pseudo frequency component that pseudo corresponds to the frequency component, and then the dequantized residual component is subtracted from this predicted value to determine the pseudo frequency. After obtaining the component, the in-band signal predictor subtracts the frequency component divided by the band divider from the predicted value obtained by the local decoder to obtain the residual component, and the residual component encoder While the difference component is quantized and then encoded into encoded data, the acoustic signal decoding device according to claim 4 is configured so that the encoded data is input and decoded into an acoustic signal.

In the invention described in claim 7, claims 1 to 6 are provided.
It is configured by combining the described inventions.

[0024]

Embodiments of the present invention will be described below with reference to the drawings. (Embodiment 1) FIG. 1 is a system configuration diagram of an acoustic signal coding apparatus and an acoustic signal decoding apparatus according to an embodiment of the present invention.

The acoustic signal coding apparatus shown in FIG. 1 is a band divider 11 for inputting an input acoustic signal and outputting 64 frequency components, and an in-band signal predictor 12, 64 divided into 64 frequency blocks. Residual component encoder 13 divided into frequency blocks, code amount allocator 14 that inputs an input acoustic signal and outputs code amount allocation for each of 32 frequency components, and is divided into 64 frequency blocks. Formatter 1 that assembles output code data into a series of bit strings
It is composed of 5.

FIG. 2 is a block diagram corresponding to each frequency block of the acoustic signal coding apparatus shown in FIG. Each of the 64 frequency blocks processes 1/64 band of each input signal, and is configured to process the entire band in total. First, FIG. 2 shows the system configuration of the audio signal encoding device.
Refer to the acoustic signal encoding device shown in FIG. 4, an example of an in-band signal predictor using the linear prediction method shown in FIG. 3, and an example of an in-band signal predictor using the lattice variable coefficient digital filter shown in FIG. And explain.

As shown in FIG. 2, the acoustic signal coding apparatus is composed of a band divider 11, an in-band signal predictor 2, a residual component coder 3, a code amount allocator 14 and a formatter 15. Here, the input acoustic signal is 16 bi
It is a linear PCM code of an acoustic signal at a constant sampling rate like the linear PCM signal of t. However, it goes without saying that the accuracy of this code does not have to be limited to 16 bits, and if the accuracy is higher (for example, 20 bits), better performance can be obtained as a whole. The input audio signal shown in FIG. 2 is monaural 1c.
Although it is an input signal of h, it goes without saying that it can be applied to many channels.

The band divider 11 is a polyphase filter bank PFB (Poly) for separating an input acoustic signal into components for each frequency.
phase filter bank), a discrete Fourier transformer DFT (Discrete Fouri) that separates the input acoustic signal into frequency components
or Transform), a discrete Hartley transformer DHT (Discrete HartleyTr) that separates the input acoustic signal into frequency components
ansform), a discrete cosine transformer DCT (Discrete) for orthogonally transforming an input acoustic signal to separate it into frequency components.
Cosine Transform), a modified discrete cosine transformer M that performs orthogonal transformation on the input acoustic signal to separate it into frequency components
It is composed of any one of DCT (Modified Discrete Cosine Transform) and the like, and has a function of dividing into a plurality of frequency components without losing the information amount of the input acoustic signal.

MDCT will now be described as an example.
In MDCT, an input signal is processed one frame (L sample) at a time and decomposed into L frequency components. L is usually 2 ^N
It is set to a value (an integer power of 2). Here, L is 64
And Next, the frequency component is obtained as follows. Input: x [m, i] m = 0,1, ..., ∞ (frame number) i = 0, ..., L-1 (frame sample number)

[0030]

[Equation 1]

For i = 0, ..., L-1 MDCT coefficient: M [i, k] = cos ((2k + 1) * (2i + n ₀ ) * (π / 2) * (1 / L )) (Equation 2) n ₀ = (L + 1) / 2 For example, when the input x [m, i] is a sine wave having a certain frequency, the output concentrates on the following components by this conversion.

Frequency coefficient (fsc / L) * (1/2) * 0.5 c [m, 0] (fsc / L) * (1/2) * 1.5 c [m, 1] (fsc / L) * ( 1/2) * 2.5 c [m, 2] ・ ・ ・ ・ ・
(Fsc / L) * (1/2) * (L-0.5) c [m, L] That is, the effect of dividing the frequency band into L equal parts was obtained by converting using MDCT. In addition,
If L = 64, the frequency band will be divided into 64 equal parts.

The in-band signal predictor 2 shown in FIG. 2 corresponds to one of the 64 blocks of the in-band signal predictor 12 shown in FIG. The in-band signal predictor 2 inputs each frequency component divided by the band divider 1, predicts the next coefficient from the previous coefficient sequence of the same frequency component, and predicts the predicted value and the actual input value. Find the difference. Here, the relationship between the predicted value y [m, k], the frequency component c [m, k], and the residual component e [m, k] is shown.

Y [m, k] = Pred {c [m-1, k], c [m-2, k], ....,} (Equation 3) e [m, k] = y [m , k] -c [m, k] (Formula 4) Note that Pred {...} shown in (Formula 3) is a frequency component y [m, k].
It is a function that predicts the next value from the past value column of.
Therefore, Pred {c [m-1, k], c [m-2, k], ...} has a slight error e
Although [m, k] occurs, a value close to the value of c [m, k] can be predicted.

Here, an example of the Pred function using the linear prediction method is shown in (Equation 5). y [m, k] = Pred {c [m-1, k], c [m-2, k], ....,} = a1 * c [m-1, k] + a2 * c [m -2, k] + a3 * c [m-3, k] + ... (Equation 5) Next, FIG. 3 shows the block configuration of (Equation 5). As shown in FIG. 3, in the in-band signal predictor 2 using the linear prediction method, the prediction value y [m, k] is obtained by multiplying the past value of the frequency component c [m, k] by the coefficient an. It is obtained by adding each term later. The residual component e [m, k] is the frequency component c from the predicted value y [m, k].
It is obtained by subtracting [m, k].

Here, when the input acoustic signal is a stationary signal, if past frequency components c [mn, k] (n = 1,2, ...) are calculated backwards, More accurately predicted value y
Can predict [m, k]. However, it is sufficiently practical to discontinue with past values up to the frequency component c [m-2, k] or c [m-3, k], for example. Next, a lattice variable coefficient digital filter applicable to the in-band signal predictor 2 will be shown.

Am (D) = Am-1 (D) -km Bm-1 (D) (Equation 6) Bm (D) = D (Bm-1 (D) -kmAm-1 (D)) (Equation 7) Incidentally, D is a delay operator and km is a PARCOR coefficient. Here, the block configurations of (Equation 6) and (Equation 7) are shown in FIG. As shown in FIG. 4, the predicted value A2 (D) corresponding to the input xi is output. Also, by setting the PARCOR coefficient km to a register value that can be set by the control unit,
For example, the PARCOR coefficient km can be adjusted as needed from a CPU (Central Processing Unit).

In the above processing, it is preferable to perform the processing with high accuracy, but since it is expressed by a real number expression, the amount of information is large even though the amplitude of the residual component itself is compressed. Therefore, it is converted into a short bit amount and the actual information compression is performed. Residual component encoder 3 shown in FIG.
Corresponds to one of the 64 blocks of the residual component encoder 13 shown in FIG. The residual component encoder 3 quantizes the residual component obtained by the in-band signal predictor 2 and converts it into a code that can be represented by a short number of bits to generate output code data.

In the residual component encoder 3, for example,
The following simple rounding integerizing function is used. Quant (x) = rint (x / Q [m, i]) (Equation 8) Requant (x) = x * Q [m, i] (Equation 9) where rint (x) is the integer closest to x Is a rounding function that takes as a value, and Q is a quantization step. For example, when the quantization step Q is 0.5, Quant (x) increases by 1 in 0.5 steps.

Therefore, when the quantization step is increased, the input can be represented by using a smaller code amount, but the accuracy is coarsely represented. The quantization step Q is controlled by the band code amount allocation output from the code amount allocator 4. Here, for simplicity, the code amount allocation is the value itself of the quantization step Q of each band. However, this method is also taken as an example for simplicity, and it goes without saying that the quantization step Q may be transmitted in another format. Further, since the value of the quantization step Q is necessary at the time of decoding, the output code also includes the quantization step Q. However, the quantization step Q
It is not necessary to directly encode the value of the same, and the same quantization step Q in the acoustic signal decoding device as in the acoustic signal encoding device is used.
The value of the quantization step Q may be transmitted by any method as long as the code generation / interpretation rule is determined so that the value of can be obtained.

Using this quantization function, the input / output relationship of the residual component encoder 3 is expressed as follows. q [m, k] = Quant (e [m, k]) (Equation 10) The code amount allocator 14 analyzes the input acoustic signal and generates band code amount allocation. Further, the code amount assigner 14 uses the IS
The input audio signal is analyzed based on the O / IEC 11172.3 international standard to generate 32 band code amount allocation values, and the code amount for each band is optimally controlled according to the input audio signal. Thus, it is possible to efficiently encode.

The formatter 15 shown in FIG. 2 has a format suitable for transmitting or storing the 32 band code amount allocation values of the code amount allocator 14 and the output code data of 64 blocks of the residual component encoder 3. Assemble to digital output. Although the code amount allocation of the residual component encoder 3 is performed based on the quantization step Q, since the quantization step Q is on average equal to the quantization noise level of the reproduction frequency component, for example, the auditory model The synthetic minimum audible level npart _n { _n = 0, ..., 31} calculated in step 1 can also be used as the quantization step. In this case, the synthetic minimum audible level npart _n
Is given by the energy, so to make it correspond to the quantization step, for example, the following calculation method may be used.

Q [m, k] = (npart _n ) ^1/2 (Equation 11) k = 2n, 2n + 1 The coded residual component thus obtained and the quantization step are: The formatter 15 shapes the connected bit string. In this way, the property of the filter decomposition capability of the auditory system can be utilized to the limit by the band divider 1 and the code amount allocator 4, and the dynamic property of the auditory sense and the minute frequency dependency of the auditory sense are , 64 in-band signal predictors 2 can be used almost to the limit, and by this combination, the auditory property is used by the natural property of each component,
It is sufficient that the auditory model and the control of the code generation amount controlled by the auditory model are simpler than those in the conventional method.

The operation of the audio signal coding apparatus shown in FIG. 1 will be described below. The acoustic signal encoding device includes a band divider 11
The input audio signal is divided into 64 frequency components by, and the code amount allocator 14 allocates 32 code amounts corresponding to each frequency band of the input audio signal. Next, after predicting the next coefficient from the previous coefficient sequence corresponding to the 64 frequency components divided by the band divider 11 by the in-band signal predictor 12, the frequency component is subtracted from the predicted value to obtain the residual error. After calculating the components, the residual component encoder 13 quantizes the residual components obtained by the in-band signal predictor 12 in correspondence with the 32 code amounts assigned by the code amount assigner 14, Since the encoded residual component is encoded, the wide area masking characteristic is used by dividing the band into 64 bands and assigning 32 code amounts between the bands, and the proximity masking effect is used as a filter characteristic of the in-band signal predictor. It can be used efficiently by control, and the time resolution and frequency resolution can be optimally adjusted by adaptively controlling the filter characteristics of the in-band signal predictor to the characteristics of the acoustic signal. Also, for acoustic inputs in the range that the in-band signal predictor can predict, the output code can be significantly reduced, and especially for waveforms containing sine waves and many harmonic components, frequency components are predicted for each band, By synthesizing them, the waveform over the entire band can be accurately predicted, and the total code amount can be reduced.

Next, FIG. 5 shows the auditory masking characteristics used in the audio signal coding apparatus according to one embodiment of the present invention. In FIG. 5, a masked sound is called a masker, and a masked sound that cannot be heard is called a masky. In this case, the audible limit (masking threshold) of another pure tone is expressed in the presence of the pure tone A (masker). Like proximity masking, the masking effect appears stronger as the frequencies of the masker and the masky become closer, so that the proximity masking effect can be efficiently used by controlling the filter characteristics of the predictor.

On the other hand, since the masking effect appears weaker as the frequencies of the masker and the masky are farther apart, the wide area masking effect is utilized by band division and code amount allocation between bands. As described above, in the present embodiment (claim 1), the band divider 11 divides the input acoustic signal into a plurality of frequency components, and the code amount allocator 14 determines the code amount corresponding to each frequency band of the input acoustic signal. assign. Next, after predicting the next coefficient from the previous coefficient sequence corresponding to the frequency component divided by the band divider 11 by the in-band signal predictor 2, the frequency component is subtracted from the predicted value to obtain the residual component. ,
After the residual component encoder 3 quantizes the residual component obtained by the in-band signal predictor 2 corresponding to the code amount assigned by the code amount assigner 4, the quantized residual component is Since coding is performed, wide-area masking characteristics can be used by band division and code amount allocation between bands, and proximity masking effect can be efficiently used by controlling the filter characteristics of the in-band signal predictor. By adaptively controlling the filter characteristics of the in-signal predictor to the characteristics of the acoustic signal, it is possible to optimally adjust the time resolution and frequency resolution capability, and furthermore, the acoustic range within which the in-band signal predictor can predict. For the input, the output code can be significantly reduced, and especially for waveforms that contain sine waves and many harmonic components, the frequency components are predicted for each band, and synthesized over the entire band. Waveform can accurately predict, it is possible to reduce the overall amount of code.

Thus, in this embodiment (claim 2),
A band divider 11 is a polyphase filter bank that separates an input sound signal into frequency components, a discrete Fourier transformer that separates the input sound signal into frequency components, and separates an input sound signal into frequency components. Of the discrete Hartley transformers, the discrete cosine transformer that performs an orthogonal transformation on the input acoustic signal to separate it into frequency components, and the modified discrete cosine transformer that performs an orthogonal transformation on the input acoustic signal to separate it into frequency components Since it is configured by any one, it is possible to select and use the function of dividing into a plurality of frequency components without losing the information amount of the input acoustic signal.

Thus, in this embodiment (claim 3),
Since the code amount assigner 14 is configured to assign the code amount corresponding to the minimum audible level of the auditory model, the synthetic minimum audible level calculated by the auditory model can be used as a quantization step and the minimum code amount is assigned. be able to. (Embodiment 2) The acoustic signal decoding apparatus shown in FIG. 1 has a deformatter 16 that decomposes an input bit string into 64 blocks of encoded data, and a code amount that decomposes the input bit string into 32 code amount allocation values for each frequency component. Extractor 10, residual component decoder 17, 64 divided into 64 frequency blocks
In-band signal predictor 18 divided into frequency blocks
1 of full band synthesized from 64 and 64 frequency component inputs
The band synthesizer 19 outputs two acoustic signals.

FIG. 6 is a configuration diagram corresponding to each frequency block of the acoustic signal decoding apparatus shown in FIG. As shown in FIG. 6, the acoustic signal decoding apparatus includes a deformatter 16, a residual component decoder 7, an in-band signal predictor 8, a band synthesizer 9 and a code amount extractor 10.

The deformatter 16 decomposes a series of transmitted or stored bit strings, outputs the 32 band code amount control data Q to the code amount extractor 10 all at once, and the code data for 64 blocks. Is decomposed and output to each residual component decoder 7. The code amount extractor 10 decomposes a group of band code amount allocation values output from the deformatter 16 into 32 band code amount allocation values, and
2 in charge of a frequency corresponding to one band code amount allocation value
Output to the residual component decoders 7, and similarly, 32 band code amount allocation values for the same are output to the total of 64 residual component decoders 7. For example, if the value of the band code amount control data Q is simply used for code amount control as described above, Q [k] may be extracted from the encoded data.

The residual component decoder 7 corresponds to one of the 64 blocks of the residual component decoder 17 shown in FIG. The residual component decoder 7 decodes the input code data into a residual component based on the band code amount allocation value output from the code amount extractor 10. Here, the residual component decoder 7 restores the input coded residual component q [m, k] to the original real value e '[m, k]. Therefore, when the quantization is performed with finite precision using the above-mentioned quantization function (Equation 10) and then the encoding is performed, the output uses the above-described quantized inverse quantization function (Equation 8) and (Equation 9). The value is represented as follows.

E ′ [m, k] = Requant (Quant (e [m, k])) (Equation 12) In (Equation 12), e [m, k] is the above-mentioned acoustic signal encoder. Is the value of the residual component before encoding of. FIG. 7 is a block diagram of the in-band signal predictor 8 in the acoustic signal decoder that is an embodiment of the present invention.

In-band signal predictor 8 corresponds to one of the 64 blocks of in-band signal predictor 18 shown in FIG. The in-band signal predictor 8 predicts the next coefficient from the previous coefficient sequence of the same frequency component, corrects the predicted value with the decoded residual component, and outputs it as a pseudo frequency component. Figure 7
As shown in, the in-band signal predictor 8 has exactly the same configuration as the in-band signal predictor 2 in the acoustic signal encoder. Therefore, the relationship between input and output is expressed as follows.

Y '[m, k] = Pred {c' [m-1, k], c '[m-2, k], ...} (Equation 13) Next, the frequency component c' [m , k] is the predicted value y '[m, k] and residual component
Reproduce as follows using e '[m, k]. c '[m, k] = y' [m, k] -e '[m, k] (Equation 14) Here, the prediction value y' [m, k] is the prediction predicted by the acoustic signal decoder 8. It is a value, and the frequency component c ′ [m, k] is the frequency component reproduced by the acoustic signal decoder. Therefore, since (Equation 14) is a modification of (Equation 4), y '[m, k] = y [m, k], e' [m, k] = e [m, k] When the relation of (Equation 15) is established, c '[m, k] = c [m, k] (Equation 16), and the frequency component reproduced by the acoustic signal decoder is transmitted to the acoustic signal encoder. It will be exactly the same as what was entered. However, in reality, as shown in (Equation 12),
Since there is a quantization error and a prediction error, it does not look like (Equation 16).

The band synthesizer 19 is an inverse polyphase filter bank IPFB (Inverse Polyphase Filter B) for synthesizing an acoustic signal from each frequency component generated by the in-band signal predictor.
ank), an inverse discrete Fourier transformer IDFT that synthesizes an acoustic signal from each frequency component generated by the in-band signal predictor
(Inverse Discrete Fourior Transform), an inverse discrete Hartley transformer IDHT (Inverse Discrete Transform) that synthesizes an acoustic signal from each frequency component generated by an in-band signal predictor.
Hartley Transform), an inverse discrete cosine transformer IDCT (Inverse Discrete C) that synthesizes an acoustic signal by performing an inverse orthogonal transform from each frequency component generated by the in-band signal predictor.
Inverse transform cosine transformer IMDCT (Inverse Modifi) that synthesizes an acoustic signal by performing an inverse orthogonal transform from each frequency component generated by the in-band signal predictor.
ed Discrete Cosine Transform), and transforms it into an acoustic signal in the form of the original time-series sample by using the frequency component as an input.

The band synthesizer 19 uses a conversion reverse to the process of dividing into 64 bands by the band divider 11 of the acoustic signal coding apparatus shown in FIG. For example, when MDCT is used in the band divider 11, the band synthesizer 19
Now use IMDCT. The operation of the acoustic signal decoding device shown in FIG. 1 will be described below. The acoustic signal decoding device decodes the coded data input by the residual component decoder 17 into 64 residual components, and the in-band signal predictor 18 decodes the next coefficient from the previous coefficient sequence of each frequency component. , And the prediction value is corrected by the residual component decoded by the residual component decoder 17 to generate 64 frequency components. Next, 6 generated by the in-band signal predictor 18 by the band synthesizer 19
Since the acoustic signal is synthesized from the four frequency components, the acoustic signal can be synthesized based on the coded data whose output code is significantly reduced, and in particular, the total code amount is reduced and a large number of harmonic components are included. Waveform acoustic signals can be synthesized.

As described above, in the present embodiment (claim 4),
The coded data input by the residual component decoder 7 is decoded into a residual component, the in-band signal predictor 8 predicts the next coefficient from the previous coefficient sequence of each frequency component, and the residual component decoder The prediction value is corrected with the residual component decoded in 7 to generate the frequency component. Next, since the acoustic signal is synthesized by the band synthesizer 19 from each frequency component generated by the in-band signal predictor 8, the acoustic signal can be synthesized based on the coded data whose output code is remarkably reduced. , It is possible to synthesize an acoustic signal having a waveform including a large number of harmonic components with the overall code amount reduced.

Thus, in the present embodiment (claim 5),
The band synthesizer 19 synthesizes an acoustic signal from each frequency component generated by the in-band signal predictor 8 and an inverse polyphase filter bank that synthesizes an acoustic signal from each frequency component generated by the in-band signal predictor 8. An inverse discrete Fourier transformer, an inverse discrete Hartley converter that synthesizes an acoustic signal from each frequency component generated by the in-band signal predictor 8, and an inverse orthogonal transform from each frequency component generated by the in-band signal predictor 8. Any one of an inverse discrete cosine transformer for applying an acoustic signal and an inverse modified discrete cosine transformer for performing an inverse orthogonal transform from each frequency component generated by the in-band signal predictor 8 to synthesize an acoustic signal Composed of two.

(Embodiment 3) FIG. 8 shows the present invention (claim 6).
FIG. 3 is a configuration diagram of a local decoder 3 and an in-band signal predictor 2 that can be applied to an acoustic signal encoding device that is one embodiment of the present invention. By replacing the local decoder shown in FIG. 8 with the residual component encoder 3 in the acoustic signal encoder of FIG.
This is applicable to the audio signal encoding device of FIG. Further, by replacing the in-band signal predictor 3 in the apparatus shown in FIG. 8 with the in-band signal predictor 2 in FIG. 2 as well, the in-band signal predictor 3 can be applied to the acoustic signal coding apparatus in FIG.

The encoder with the local decoder is
Although it has the same function as the acoustic signal decoding apparatus described in the second embodiment, it is called a local decoder for the sake of distinction. It should be noted that it is also possible to directly input unformatted data to the local decoder without passing through the formatter / deformatter. The signal in the local decoder is described as c "[m, k], y" [m, k] and e "[m, k] to distinguish it from the signal in the audio signal decoding apparatus.

In the encoder with local decoder shown in FIG. 8, as input to the in-band signal predictor 2 of the acoustic signal encoder, c [m-1, k], c [m-2, k], ... In place of, input the frequency component value c "[m-1, k], c" [m-2, k], ... Of the local decoder. The operation of the encoder with local decoder shown in FIG. 8 will be described below. First, the frequency component c "[m, k] input from the local decoder prediction value y" [m, k] predicted by the linear predictor 25 is subtracted by the adder 21 to obtain the residual component e [m, k]. ]] Next, the residual component e [m, k] is quantized by the quantizer 22 in accordance with the code amount allocation value Q to obtain an output value, and
Further, this value is inversely quantized by the inverse quantizer 23 corresponding to the code amount allocation value Q to obtain the residual component e "[m, k].
Local decoder prediction value y "predicted by the linear predictor 25
This residual component e "[m, k] is subtracted from [m, k] by the adder 24 to obtain the frequency component c" [m, k]. Here, the frequency components c ″ [m, k] obtained by the adder 24 are sequentially delayed elements 26 to 2
9 and becomes the input data of the linear predictor 25.

As described above, the in-band signal predictor 2 or 8
Input uses c "[m, k] in the local decoder, so y [m, k] = Pred {c" [m-1, k], c "[m-2, k], ..} (Equation 17) Note that c [m, k] and c "[m, k] are not the same value,
Since the values are close to each other, a sufficiently good prediction value y [m, k] can be obtained even by the method of (Equation 17).

Next, an encoder / decoder with a local decoder which can be applied to the acoustic signal processing apparatus will be described. The encoder / decoder with a local decoder is one in which acoustic signal decoding devices are further connected in series to the encoder with a local decoder. In this case, the residual component encoder 3 and the residual component decoder 7 are installed at remote locations, during which only encoded data is transmitted.

Next, the effect of the local decoder will be described in comparison with the case where there is no local decoder, that is, the case where the residual component encoder 3 shown in FIG. 2 and the residual component decoder 7 shown in FIG. . In the case of the residual component encoder 3 shown in FIG. 2 and the residual component decoder 7 shown in FIG. 7, for example, y [m, k] = Pred {c [m-1, k], c [m -2, k], ...} (Equation 18) y '[m, k] = Pred {c' [m-1, k], c '[m-2, k], ...} (Equation 19) e [m, k] = y [m, k] -c [m, k] (Equation 20) c '[m, k] = y' [m, k] -e '[m, k] ( Formula 21) y [m + 1, k] = Pred {c [m, k], c [m-1, k], ...} (Formula 22) y '[m + 1, k] = Pred { c '[m, k], c' [m-1, k], ...} (Equation 23) e [m + 1, k] = y [m + 1, k] -c [m + 1, Consider the case where the calculation is repeated as k] (Equation 24) c '[m + 1, k] = y' [m + 1, k] -e '[m + 1, k] (Equation 25). Note that (Equation 1
8) Up to (Equation 19), it is assumed that the predicted value is y [m, k] = y '[m, k].

Here, according to the relation of (Equation 12), the residual component e '[m, k] causes a slight error with e [m, k], so the frequency component c' [determined by (Equation 21) is obtained. For m, k], some error is added to c [m, k]. As a result, the predicted values y '[m + 1, k] and y [m + obtained by (Equation 22) and (Equation 23) are obtained.
1, k] also have different prediction values, so the frequency component c '
[m, k] and c [m, k] have very different values.

Therefore, the error between c '[m + 1, k] and c [m + 1, k] obtained by (Equation 24) and (Equation 25) is the predicted value y' [m + 1, k].
The error of the residual component e '[m + 1, k] is added to the error of to make it even larger. As described above, in the residual component encoder 3 having no local decoder, the residual component encoder 3
Since the error gradually accumulates in the prediction value output from the in-band signal predictor 8 in the acoustic signal decoding device, the error also accumulates in the difference between the input of the residual component encoder 3 and the decoded value. There was the problem of unlimited expansion.

On the other hand, in the encoder / decoder with the local decoder, the operation of the residual component encoder 3 and the decoder is as follows. y "[m, k] = Pred {c" [m-1, k], c "[m-2, k], ...} (Equation 26) y '[m, k] = Pred {c' [m-1, k], c '[m-2, k], ...} (Equation 27) e [m, k] = y "[m, k] -c [m, k] (Equation 28 ) C "[m, k] = y" [m, k] -e "[m, k] (Equation 29) c '[m, k] = y' [m, k] -e '[m, k ] (Equation 30) y "[m + 1, k] = Pred {c" [m, k], c "[m-1, k], ...} (Equation 31) y '[m + 1, k] = Pred {c '[m, k], c' [m-1, k], ...} (Equation 32) e [m + 1, k] = y [m + 1, k] -c [m + 1, k] (Equation 33) c "[m + 1, k] = y" [m + 1, k] -e "[m + 1, k] (Equation 34) c '[m + 1 , k] = y '[m + 1, k] -e' [m + 1, k] (Equation 35) Here, it is assumed that the local decoder and the decoder are in the same state in the initial state. The values y "[m, k] and y '[m, k] produce the same predictive value because the same device remains the same. As a result, the residual component e [m, k] is calculated and quantized, and then the dequantized residual component e "[m, k]
And e '[m, k] have the same value. Predicted value y '[m, k], y "
Since [m, k] are the same value, as a result, frequency components c '[m, k] and c "
[m, k] also have the same value. As described above, the processing of one prediction value is completed, and a new value is input to the in-band signal predictor 2. At that time, the frequency component c "[m, k] is the same value as c '[m, k] in the decoder, so the next predicted values y' [m + 1, k] and y "obtained in (Equation 31) and (Equation 32) are [m + 1] is also a perfect match.

Similarly, in the processing of (Equation 30) to (Equation 34), the prediction value y "[m, k] of the local decoder and the prediction value y '[m, k] in the decoder have the same value. Since the processing continues, the prediction value y '[m
There is no difference between + n, k] and y "[m + n, k]. Therefore, from (Equation 28) and (Equation 29), for example, the frequency components c [m, k] and c" The error between [m, k] is the residual component e [m, k] and e "
It is equal to the error of [m, k], which is equal to the error due to the quantization and dequantization. Since this error is always smaller than the quantization step, the encoder with local decoder has the effect of reducing the error between the value reproduced by the audio signal decoding device and the input within a fixed value (quantization step width). can get.

As described above, in this embodiment (claim 6),
In the acoustic signal encoding device, when the residual component encoder 3 quantizes the residual component in correspondence with the code amount assigned by the code amount assigner 14 by the local decoder, it pseudo corresponds to the frequency component. After predicting the next coefficient from the previous coefficient sequence of the pseudo frequency component to be obtained, and subtracting the dequantized residual component from this predicted value to obtain the pseudo frequency component, the in-band signal predictor 2 The residual component is obtained by subtracting the frequency component divided by the band divider 11 from the prediction value obtained by the local decoder, and the residual component is quantized by the residual component encoder 3, and then converted into encoded data. Encode. On the other hand, in the acoustic signal decoding device, since this encoded data is input and decoded into an acoustic signal, the value of the frequency component reproduced by the acoustic signal decoding device and the frequency input by the acoustic signal encoding device The components can be matched with each other within a constant error, and the error of the residual component can be reduced within the quantization step width.

[0070]

According to the audio signal coding apparatus of the present invention,
Wide area masking characteristics can be utilized by band division and code amount allocation between bands, and proximity masking effect can be efficiently utilized by controlling the filter characteristics of the in-band signal predictor. By adaptively controlling the filter characteristics to the characteristics of the acoustic signal, it is possible to optimally adjust the time resolution and frequency resolution, and for the acoustic input within the range that the in-band signal predictor can predict, output Signs can be significantly reduced, and especially for waveforms that contain sine waves and many harmonic components, frequency components can be predicted for each band, and by combining them, the waveform over the entire band can be predicted with high accuracy. The code amount can be reduced.

Further, according to the acoustic signal coding apparatus of the present invention, the acoustic signal can be synthesized based on the coded data whose output code is remarkably reduced, and in particular, a large number of harmonics in which the overall code amount is reduced. It is possible to synthesize an acoustic signal having a waveform including wave components.

[Brief description of drawings]

FIG. 1 is a system configuration diagram of an audio signal encoding device and an audio signal decoding device according to an embodiment of the present invention.

FIG. 2 is a configuration diagram corresponding to each frequency block of the audio signal encoding device shown in FIG.

FIG. 3 is a configuration diagram of an in-band signal predictor using a linear prediction method that is an embodiment of the present invention.

FIG. 4 is a configuration diagram of an in-band signal predictor using a lattice-type variable coefficient digital filter that is an embodiment of the present invention.

FIG. 5 shows masking characteristics of the acoustic signal coding apparatus according to the embodiment of the present invention.

6 is a configuration diagram corresponding to each block of the acoustic signal decoding device shown in FIG. 1. FIG.

FIG. 7 is a configuration diagram of an in-band signal predictor 8 in an acoustic signal decoder that is an embodiment of the present invention.

FIG. 8 is a configuration diagram of a local decoder 3 and an in-band signal predictor 2 which can be applied to an acoustic signal encoding device which is an embodiment of the present invention (claim 6).

[Explanation of symbols]

 1 band splitter 2 in-band signal predictor 3 residual component encoder 4 code amount assigner 7 residual component decoder 8 in-band signal predictor 9 band combiner 10 code amount extractor 11 band splitter 12 in-band Signal predictor 13 Residual component encoder 14 Code amount allocator 15 Formatter 16 Deformatter 17 Residual component decoder 18 In-band signal predictor 19 Band synthesizer

Claims (7)

[Claims] 1. An audio signal coding apparatus for inputting an audio signal and encoding while reducing the redundancy of the audio signal, and a band divider for dividing the input audio signal into a plurality of frequency components,
A code amount allocator that allocates a code amount corresponding to each frequency band of the input acoustic signal, and the predicted value after predicting the next coefficient from the previous coefficient sequence corresponding to the frequency component divided by the band divider To the in-band signal predictor for subtracting the frequency component from the in-band signal predictor to obtain the residual component, and the residual component obtained by the in-band signal predictor corresponding to the code amount assigned by the code amount assigner. An acoustic signal coding apparatus, comprising: a residual component coder that performs quantization and then codes. 2. A polyphase filter bank for separating the input acoustic signal into frequency components, a discrete Fourier transformer for separating the input acoustic signal into frequency components, and the input acoustic signal. A discrete Hartley converter for separating each frequency component into a component, a discrete cosine converter for performing an orthogonal transform on the input acoustic signal to separate into a component for each frequency, and an orthogonal transform of the input acoustic signal into a component for each frequency. The acoustic signal encoding device according to claim 1, wherein the acoustic signal encoding device is configured by any one of the modified discrete cosine converters to be separated. 3. The acoustic signal coding apparatus according to claim 1, wherein the code amount assigner is configured to assign the code amount corresponding to the minimum audible level of the auditory model. 4. An acoustic signal decoding apparatus for inputting coded data with reduced redundancy to decode into an acoustic signal, a residual component decoder for decoding the input coded data into a residual component, respectively. An in-band signal predictor that predicts the next coefficient from the previous coefficient sequence of the frequency component of, and corrects the prediction value with the residual component decoded by the residual component decoder to generate a frequency component; An audio signal decoding device, comprising: a band synthesizer that synthesizes an acoustic signal from each frequency component generated by the in-band signal predictor. 5. An inverse polyphase filter bank in which the band synthesizer synthesizes an acoustic signal from each frequency component generated by the in-band signal predictor, and each frequency component generated by the in-band signal predictor. Inverse discrete Fourier transformer for synthesizing an acoustic signal from, an inverse discrete Hartley transformer for synthesizing an acoustic signal from each frequency component generated by the in-band signal predictor, each frequency component generated by the in-band signal predictor An inverse discrete cosine transformer that synthesizes an acoustic signal by performing an inverse orthogonal transform from the inverse deformed discrete cosine transformer that synthesizes an acoustic signal by performing an inverse orthogonal transform from each frequency component generated by the in-band signal predictor, The audio signal decoding apparatus according to claim 5, wherein the audio signal decoding apparatus is configured by any one of the above. 6. The acoustic signal coding apparatus according to claim 1, wherein when the residual component encoder quantizes the residual component in correspondence with the code amount assigned by the code amount assigner, After predicting the next coefficient from the previous coefficient sequence of the pseudo frequency component corresponding to the frequency component in a pseudo manner, the local decoder for obtaining the pseudo frequency component by subtracting the dequantized residual component from the predicted value is used. The in-band signal predictor subtracts the frequency component divided by the band divider from the prediction value obtained by the local decoder to obtain a residual component, and the residual component encoder encodes the residual component. 5. The acoustic signal decoding device according to claim 4, wherein the residual component is quantized and then encoded into encoded data, and the acoustic signal decoding device according to claim 4 is configured so that the encoded data is input and decoded into an acoustic signal. Audio signal processing device. 7. An audio signal processing device, which is configured by combining the inventions of claims 1 to 6.