JP3248668B2 - Digital filter and acoustic encoding / decoding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital filter using a linear prediction coefficient as a filter coefficient,
In particular, in the encoding of audio signals such as voice and musical sound, an auditory weighting filter that performs weighting in consideration of auditory characteristics,
The present invention relates to an all-pole or moving average digital filter for audio signal processing, such as a synthesis filter for decoding and synthesizing an encoded code of an audio signal, and a post filter for suppressing quantization noise of the decoded synthesized signal based on auditory characteristics.

[0002]

2. Description of the Related Art As a typical method of encoding an audio signal at a low bit rate by encoding linear coefficients obtained by linear prediction, CELP (Code E) is used.
xcited Linear Prediction:
A code excitation linear prediction) method is conventionally known.

FIG. 9 is a block diagram showing a conventional acoustic encoding apparatus based on the CELP method. FIG.
It is a block diagram which shows the conventional acoustic encoding device AC4.

In the audio coding apparatus AC4, the input speech signal input to the input terminal 11 is subjected to linear prediction analysis by the linear prediction analysis means 12 for each frame of about 5 to 20 ms, and a p-order linear prediction coefficient α _0i (i = 1, 2,...,
p) is required. The “linear prediction analysis” is based on the idea that a sample value of an audio signal is approximated by a linear combination of some sample values at an earlier time.

The linear prediction coefficient α _0i output from the linear prediction analysis means 12 is quantized by the quantization means 13, and the quantized linear prediction coefficient α _0i is set as a filter coefficient in the linear prediction synthesis filter 14. Is done. Synthesis filter 1
The transfer function h (z) of No. 4 is expressed by the following equation (1).

[0006]

(Equation 1) The excitation signal to be input to the synthesis filter 14 (the signal output from the adding means 18) is stored in the adaptive codebook 15, and the excitation signal (from the adaptive codebook 15 at a pitch cycle corresponding to the code output from the control means 16). Vector) is cut out, and this is repeated to obtain a frame length.
, And is supplied to the synthesis filter 14 as an excitation signal through the addition means 18.

The synthesized signal output from the synthesis filter 14 is
It is subtracted from the input signal by the subtraction means 19, and the subtracted difference signal is obtained by the auditory weighting filter 21.
The weighting corresponding to the masking characteristic (?) Of the auditory characteristic is performed, and the control unit 16 searches for the input code (trimming pitch period) of the adaptive codebook 15 so that the energy of the weighted difference signal is minimized. You.

After that, under the control of the control means 16, the excitation vectors are sequentially extracted from the noise codebook 22, and the gain is applied by the gain applying means 23. The excitation vector from the adaptive codebook 15 is added by the adding means 18,
The excitation vector is supplied to the synthesis filter 14 as an excitation signal, and in the same manner as described above, the excitation vector (the excitation vector output by the gain applying unit 17 and the excitation vector Is selected in combination with the excitation vector output by the.

Finally, gain applying means 17 and 23 apply predetermined gains to the respective excitation vectors output from the adaptive codebook 15 and the noise codebook 22 selected as described above. Weighting filter 21 in
Is searched for such that the energy of the output signal is minimized.

The unquantized linear prediction coefficients α _i and 2
Using two constants γ1 and γ2 of 1 or less, the following equation (2)
The transfer function w (z) is obtained by
(Z) is used for the auditory weighting filter 21.

[0011]

(Equation 2) A code indicating the quantized linear prediction coefficient, a code indicating the excitation vector selected in each of the adaptive codebook 15 and the noise codebook 22, and a code indicating each optimal gain given to the gain applying means 17 and 23 , The output of the audio encoding device AC4.

FIG. 9 (2) shows a conventional audio coding apparatus AC.
FIG.

The acoustic encoding device AC5 is an acoustic encoding device AC4. The acoustic encoding device AC5 is composed of a perceptual weighted synthesis filter 2 obtained by synthesizing the linear prediction filter 14 and the perceptual weighting filter 21.
4 is provided in place of the linear prediction filter 14. In this case, the input signal from the input terminal 11 is supplied to the subtraction means 19 through the auditory weighting filter 21.

FIG. 10 is a block diagram showing a conventional acoustic decoding device AD1 based on the conventional CELP method.

In the audio decoding device AD1, the linear predictive coefficient code in the input code input from the input terminal 31 is inversely quantized by the inverse quantizing means 32. The inverse quantized linear predictive coefficient is used as a filter coefficient. The linear prediction filter 33 is set. Depending on the pitch code in the input code,
An excitation vector is cut out from the adaptive codebook 34, and
The excitation vector is selected from the noise codebook 35 by the noise code, and the excitation vectors from the codebooks 34 and 35 are given gains by the gain applying means 36 and 37 according to the gain code in the input code. Thereafter, the signals are added by the adding means 38 and supplied to the synthesis filter 33 as an excitation signal. The synthesized signal from the synthesis filter 33 is processed by the post filter 39 and output so that the quantization noise is reduced in consideration of the auditory characteristics.

[0016]

In the conventional audio signal encoding apparatus in the time domain such as CELP, the conventional perceptual weighting filter, synthesis filter, and post-filter are of the order of 10 to 20 which can model the formant of speech. Since the filter coefficients are determined by the linear prediction, it is not possible to represent a fine spectral structure of an acoustic signal having a large number of unequally spaced peaks that are stationary in the frequency domain.

Therefore, in order to obtain a composite signal having a fine spectral structure, the excitation signal input to the synthesis filter needs to have information on the fine spectral structure. Requires a large amount of information bits, and a codebook search-type coding scheme such as CELP requires a large amount of calculation for codebook search to code the excitation signal. However, there is a problem that processing in real time is impossible.

As a means for expressing the fine spectrum structure, a method of simply increasing the order of linear prediction can be considered. According to this method, when compared with a method of quantizing the obtained prediction coefficient, the prediction There is a problem that the prediction order cannot be increased with the calculation accuracy required for the process of obtaining the coefficients.

In the above-mentioned conventional example, the auditory weighting and noise suppression are controlled only by parameters such as linear prediction coefficients of about the order of 10 to 20 and a single pitch frequency. Control is impossible.

The present invention provides, as a digital filter, a fine spectral structure of an audio signal including speech, auditory weighting and noise suppression based on the fine spectral structure.

[0021]

According to a first aspect of the present invention, a p-order linear prediction analysis is performed on an input signal, and a filter coefficient is determined from the p-th linear prediction coefficient. The obtained prediction residual signal is subjected to nth-order linear prediction, a filter coefficient is determined from the nth-order linear prediction coefficient, and the obtained all-pole or moving average digital filters are connected in cascade. It is a digital filter.

According to a second aspect of the present invention, there is provided an encoding apparatus for determining an encoding code so that a difference signal between an acoustic input signal and a synthesized signal is minimized. As an all-pole or moving average digital filter to be applied, a p-order linear prediction analysis is performed on the input signal, and a filter coefficient is determined from the p-order linear prediction coefficient.
An nth-order linear prediction analysis is performed on the prediction residual signal obtained by the pth-order linear prediction, and a filter coefficient is determined from the nth-order linear prediction coefficient. This is an acoustic encoding device in which filters are connected in cascade.

According to a third aspect of the present invention, in a coding apparatus that determines a coding code so that a difference signal between an acoustic input signal and a synthesized signal is minimized, weighting is performed on the difference signal in accordance with auditory characteristics. As an all-pole or moving average digital filter, a p-order linear prediction analysis is performed on the input signal, a filter coefficient is determined from the p-th linear prediction coefficient, and an m-th order The linear prediction analysis is performed, the n-th linear prediction analysis is performed on the prediction residual signal obtained by the m-th linear prediction analysis, the filter coefficients are determined from the n-th linear prediction coefficients, and the digital coefficients respectively obtained are determined. This is an acoustic encoding device in which filters are connected in cascade. Note that the above m may be equal to p or slightly different.

According to a fourth aspect of the present invention, there is provided an encoding apparatus for determining an encoding code such that a difference signal between an acoustic input signal and a synthesized signal is minimized. As an all-pole or moving average digital filter to be applied, a p-order linear prediction analysis is performed on the input signal, and a filter coefficient is determined from the p-order linear prediction coefficient.
An input signal to a synthesis filter in which a filter coefficient is determined from a quantized linear prediction coefficient in a past frame is stored, an n-order linear prediction analysis is performed on the stored signal, and a filter is performed based on the n-th linear prediction coefficient. Determine the coefficient,
This is an acoustic encoding device in which digital filters obtained in each case are connected in cascade.

According to a fifth aspect of the present invention, a digital filter for synthesizing a composite signal is provided in an audio encoding / decoding apparatus which determines an encoding code so as to minimize a difference signal between the audio input signal and the composite signal. Performs a p-order linear prediction analysis on the input signal, quantizes the p-order prediction coefficient to generate a quantized prediction coefficient, and determines filter coefficients from the p-order prediction coefficient and the quantized prediction coefficient, respectively. Performs an m-th linear prediction analysis on the synthesized signal in the past frame, performs an n-th linear prediction analysis on the prediction residual signal obtained by the m-th linear prediction, and obtains an n-th linear prediction coefficient , A filter coefficient is determined from, and the digital filter obtained is connected in cascade. Note that m may be equal to p or slightly different.

According to a fifth aspect of the present invention, a digital filter for synthesizing a composite signal is provided in an audio encoding / decoding apparatus which determines an encoding code so that a difference signal between the audio input signal and the composite signal is minimized. , Performs a p-order linear prediction analysis on the input signal, quantizes the p-th prediction coefficient to generate a quantized prediction coefficient, determines a filter coefficient from the p-th quantization prediction coefficient, and performs synthesis in a past frame. An m-order linear prediction analysis is performed on the signal, an n-th linear prediction analysis is performed on the prediction residual signal obtained by the m-th linear prediction, and a filter coefficient is determined from the n-th linear prediction coefficient. ,
This is an audio encoding / decoding device that connects the digital filters obtained in cascade. Note that m may be equal to p or slightly different.

According to a sixth aspect of the present invention, there is provided an encoding / decoding device for determining an encoding code such that a difference signal between an acoustic input signal and a synthesized signal is minimized, as a digital filter for synthesizing a synthesized signal. Perform a p-order linear prediction analysis on the input signal, quantize the p-th prediction coefficient to generate a quantized prediction coefficient, determine a filter coefficient from the p-th quantized prediction coefficient, Saves the signal input to the synthesis filter in which the filter coefficient is determined from the quantized linear prediction coefficient, performs n-th linear prediction on the stored signal, and determines the filter coefficient from the n-th linear prediction coefficient And an acoustic encoding / decoding device for cascade-connecting the obtained digital filters.

According to a seventh aspect of the present invention, as a digital filter for performing a process of acoustically suppressing quantization noise on a decoded synthesized signal obtained by encoding a sound, a decoded synthesized signal including a past frame is used. An m-th linear prediction analysis is performed on the prediction residual signal, an n-th linear prediction is performed on the prediction residual signal obtained by the m-th linear prediction analysis, and a filter coefficient is determined from the n-th linear prediction coefficient. This is an audio decoding device in which the obtained digital filters are connected in cascade. Note that m may be equal to p or slightly different.

An eighth aspect of the present invention is a digital filter for performing a process of acoustically suppressing quantization noise on a decoded and synthesized signal obtained by encoding a sound by using a p-order linear prediction coefficient in a past frame. Saves the input signal to the synthesis filter for which the filter coefficients have been determined, performs n-order linear prediction analysis on the stored signal, determines filter coefficients from the n-th linear prediction coefficients, and determines the current frame. And a digital filter whose filter coefficient is determined by a p-order linear prediction coefficient obtained from the encoded code in the above, and the digital filter obtained respectively are connected in cascade.

[0030]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1A is a block diagram showing an audio encoding apparatus AC1 according to an embodiment of the present invention.

This acoustic encoding apparatus AC1 is an embodiment corresponding to claim 5, and in the conventional acoustic encoding apparatus AC4 shown in FIG. 9, a synthesis filter 44 is provided instead of the synthesis filter 14, and the m-th order Linear prediction analysis means 41, LP
It is provided with a C inverse filter 42, an nth-order linear prediction analysis unit 43, a second filter coefficient determination unit 43a, and a first filter coefficient determination unit 13a.

In the audio coding apparatus AC1, first, an input signal in the current frame is subjected to linear prediction analysis to obtain a p-order linear prediction coefficient α _0i (i = 1, 2,..., P). As the linear prediction coefficient α _0i , a coefficient obtained by the linear prediction analysis unit 12 in the conventional audio encoding device AC4 can be used. Further, the prediction coefficients are quantized to obtain quantized prediction coefficients α _i (i = 1, 2,..., P). Usually p
Is about 10 to 20.

Next, the synthesized signal over the past frame is subjected to linear prediction analysis by the linear prediction analysis means 41 to obtain an m-th linear prediction coefficient α ^′ _k (k = 1, 2,..., M). In addition,
As mentioned above, m may be equal to p or slightly different. In the case of performing linear prediction analysis, any of an asymmetric window and a symmetric window such as a Hamming window may be used as a window applied to a signal sequence to be analyzed.

Next, LPC inverse filtering is performed on the past synthesized speech by the digital filter 42 whose transfer characteristic is expressed by the following equation (3) using the linear prediction coefficient α ^′ _k as a filter coefficient, Find the difference signal.

[0035]

(Equation 3) Next, the prediction residual signal of the obtained past synthesized signal (for example, the synthesized signal one pitch before) is converted to a linear prediction analysis unit 43.
And a linear prediction coefficient β _{j of} order n (j =
1, 2,..., N). In order for the n-th linear prediction to represent a higher-order correlation that cannot be completely predicted by the m-th linear prediction, n is preferably larger than m.
For example, when the encoding target is music, prediction of the 100th or higher order may be optimal.

Next, using the obtained coefficients α _i and β _j ,
Digital filters whose transfer characteristics are expressed by the following equations (4) and (5) are connected in cascade to form a digital filter whose transfer characteristics are expressed by an equation (6).

FIG. 1B is a block diagram showing an acoustic encoding device AC2 according to another embodiment of the present invention.

Depending on the nature of the signal to be encoded, only the excitation signal from some of the excitation codebooks among the plurality of excitation codebooks is converted into the synthesis filter 4 like the acoustic encoding device AC2.
4 and the other excitation signals from the excitation codebook may be filtered by a conventional synthesis filter.

[0039]

(Equation 4)

[0040]

(Equation 5)

[0041]

(Equation 6) FIG. 2A is a diagram illustrating an example of a logarithmic power spectrum of an input signal to be encoded, and FIG. 2B is a diagram illustrating a logarithmic power spectrum characteristic of a filter representing an envelope in a synthesis filter of the acoustic encoding device AC1. FIG. 2C is a diagram illustrating an example of a logarithmic power spectrum characteristic of a filter representing a fine structure in the synthesis filter by the acoustic encoding device AC1, and FIG.
FIG. 7 is a diagram illustrating an example of a logarithmic power spectrum characteristic of a synthesis filter by the audio encoding device AC1.

In the above embodiment, a filter representing the envelope of the power spectrum characteristic of the signal to be encoded and a filter representing the fine structure are obtained, and both filters are cascaded.

When expressing the power spectrum characteristic of the signal to be encoded shown in FIG. 2A, the envelope shape is represented by a low-order filter coefficient α _i quantized as in the prior art.
Expressed as (2), the fine structure is obtained by converting a past synthesized speech to a higher-order filter coefficient β obtained by a higher-order analysis.
_The power spectrum characteristic as shown in FIG. 2 (4) can be expressed by _j as shown in FIG. 2 (3) and the cascade connection of these filters. The above embodiment, compared to the conventional example,
An extra information bit for transmitting to the decoding device is not required, and more complex power spectrum characteristics of the signal to be encoded can be expressed.

In one linear prediction analysis, it is difficult to stably obtain a prediction coefficient for performing a higher-order analysis that can express a fine structure. Since the signal to be performed is the prediction residual signal from which the low-order correlation has been removed, it is easy to stably obtain the prediction coefficient even in the high-order analysis. If a stable prediction coefficient cannot be obtained, a value obtained in a previous frame or an initialized value can be used as a stable prediction coefficient.

Even when the initialized value is used as a stable prediction coefficient, if a quantized low-order prediction coefficient is obtained, at least the gain by the low-order prediction is guaranteed.

In the above embodiment, in order to input the excitation signal to the synthesis filter and obtain the synthesized signal, in addition to the conventional filtering operation, filtering by the filter coefficient β _j is required. Once the impulse response and the quiescent response of the synthesis filter are obtained once, as is often used in the acoustic encoding device AC5, the excitation from the moving average digital filter using the impulse response as a filter coefficient and the codebook is performed. By selecting the excitation signal so that the error between the signal obtained by convolving the signal and the difference signal obtained by subtracting the quiescent response from the input signal is minimized, a filter is applied to all the excitation signals in the codebook. Filtering operation using the coefficient β _j can be avoided, and in this case, the amount of computation required only for codebook search is the same as in the past.

FIG. 3 is a block diagram showing another embodiment of the present invention, and is a block diagram showing an auditory weighting filter 211 obtained by improving the auditory weighting filter 21 in the conventional audio encoder AC4.

The auditory weighting filter 211 is an invention corresponding to the first and second aspects.
Perform p-order linear prediction analysis to find p-order prediction coefficient
First-order linear prediction analysis means 12 and first filter coefficient determination means 12a for determining filter coefficients of an all-pole or moving average digital filter based on a p-th prediction coefficient
And the input signal is passed through an LPC inverse filter (61) to obtain p
Means 62 for performing an nth-order linear prediction on the signal sequence obtained as a prediction residual of the next linear prediction, and determining filter coefficients of an all-pole or moving average digital filter based on the nth prediction coefficient A second filter coefficient determining means 62a, a first all-pole or moving average digital filter 63-1 whose filter coefficient is determined by a p-order linear prediction, and a filter coefficient determined by an n-th linear prediction This is an auditory weighting filter having a second all-pole or moving average digital filter 63-2 and connection means for cascading the first filter and the second filter. Dividing the analysis into two in this way makes the filter design free. Note that the filter 211 can be used for an audio encoding device and an audio decoding device, and can also be used for an encoding / decoding device for non-audio such as video.

When the filter 211 is used in an audio encoding device, the spectral envelope of the audio input signal is modeled by linear prediction analysis, and the difference signal between the audio input signal and the synthesized signal of the encoded code is obtained. In the acoustic encoding device that determines the encoding code so that is minimized and weights the difference signal according to the auditory characteristics, a p-order linear prediction analysis is performed on the audio input signal. , A p-order linear prediction analysis means 12 for obtaining a p-th prediction coefficient, a first filter coefficient determination means 12 a for determining a filter coefficient based on the p-th prediction coefficient, and the input to the LPC inverse filter 61. N-order linear prediction analysis means 43 for performing an n-th order linear prediction analysis on the p-th order linear prediction residual signal obtained through the Based on A second filter coefficient determining means 62a for determining the data coefficients, said plurality of digital filters 6
3-1 and 63-2.
In this case, precise auditory weighting can be performed.

FIG. 4 is a block diagram showing an auditory weighting filter 212 according to another embodiment of the present invention. The auditory weighting filter 212 is a conventional audio encoding device AC4.
This is an improved version of the auditory weighting filter 21 in the first embodiment, and is an embodiment corresponding to claim 3.

The auditory weighting filter 212 models the spectral envelope of the audio input signal by linear prediction analysis, and minimizes the difference signal between the audio input signal and the synthesized signal of the encoded code. A sound coding apparatus that determines a coding code and weights the difference signal according to auditory characteristics, performs a p-order linear prediction analysis on the sound input signal, and calculates a p-order prediction coefficient. The p-order linear prediction analysis means 12 to be obtained, the first filter coefficient determination means 12a for determining a filter coefficient based on the p-order prediction coefficient, and the synthesized signal for the past frame in the audio input signal. On the other hand, an m-th linear prediction analysis means 51 for performing an m-th linear prediction analysis to obtain an m-th prediction residual signal, and an n-th linear prediction analysis based on the m-th prediction residual signal. An n-th linear prediction analysis means 53 for obtaining an n-th prediction coefficient; a second filter coefficient determination means 53a for determining a filter coefficient based on the n-th prediction coefficient; -1 and 54-2 in cascade.

First, the input signal in the current frame is subjected to linear prediction analysis, and a p-order linear prediction coefficient α _0i (i = 1, 2,...)
..., p). As the linear prediction coefficient α _0i , the one obtained by the linear prediction analysis unit 12 in the audio encoding device AC4 can be used. Usually, p is about 10 to 20.

Next, the input signal over the current frame and the past frame is subjected to linear prediction analysis by the linear prediction means 51,
The m-th order linear prediction coefficient α ^′ _k (k = 1, 2,..., m) is obtained. Note that, as described above, m may be equal to p or slightly different. In the case of performing linear prediction analysis, a window applied to a signal sequence to be analyzed may be an asymmetric window or a symmetric window such as a Hamming window.

Next, the input signal over the current frame and the previous frame, the LPC coefficients alpha ^'a _k as the filter coefficient, the transfer characteristic is alpha of the above formula ^(1)' a _k inv
LPC inverse filtering is performed by the digital filter 52 replaced with (αk) to obtain a prediction residual signal.

Next, the obtained prediction residual signal of the past synthesized signal is subjected to linear prediction analysis by the linear prediction means 53, and
An nth-order linear prediction coefficient β _j (j = 1, 2,..., n) is obtained. Since the n-th linear prediction represents a higher-order correlation that cannot be predicted by the m-th linear prediction, n is preferably larger than m. For example, when the encoding target is music, the prediction of the 100th or higher order may be optimal in the same manner as in the above embodiment.

Next, the obtained coefficients α _0i , inv (β _j )
Using the constants γ1, γ2, γ3, and γ4 of 1 or less, a digital filter 54 whose transfer characteristic is represented by the following equation (7) is used as the perceptual weighting filter 21 in the audio encoding device AC4.

[0057]

(Equation 7) In the above-described embodiment, the conventional filter coefficient α _0i and the auditory weighting filter based on the power spectrum envelope of the input signal represented by the constants γ1 and γ2 of 1 or less are _added to the high-order filter coefficient β _0i and the By adding an auditory weighting filter based on the fine structure of the power spectrum of the input signal represented by the constants γ1 and γ4,
Even for an input signal in which a plurality of pitches are mixed, finer control according to the auditory characteristics can be performed.

In the above embodiment, a low-order linear prediction analysis is performed on the synthesized signal or the input signal, and a high-order linear prediction analysis is performed on the prediction residual signal. The digital filter using the predicted coefficients can express more complex power spectrum characteristics than before. Therefore, in encoding such as the CELP method in which a large number of excitation vectors are passed through a filter, it is necessary to have power spectrum characteristics of various acoustic signals as variations of excitation vectors in the synthesis filter excitation codebook by using the synthesis filter. It is effective in that there is no. Further, it is effective in that weighting according to auditory characteristics and noise suppression can be more finely controlled than in the past.

With this hearing weighting filter 212, accurate hearing weighting can be performed.

FIG. 5 is a block diagram showing an auditory weighting filter 213 according to another embodiment of the present invention. The auditory weighting filter 213 is a conventional audio encoding device AC4.
This is an improved version of the auditory weighting filter 21 in the first embodiment, and is an embodiment corresponding to claim 4.

The auditory weighting filter 213 models the spectral envelope of the audio input signal by linear predictive analysis, and minimizes the difference signal between the audio input signal and the synthesized signal of the coded code. A sound coding apparatus that determines a coding code and weights the difference signal according to auditory characteristics, performs a p-order linear prediction analysis on the sound input signal, and calculates a p-order prediction coefficient. P-order linear prediction analysis means 12 to be obtained, first filter coefficient determination means 2a for determining a filter coefficient based on the p-order prediction coefficient, and quantization prediction in a past frame of the audio input signal. A signal storage unit 15 for storing a signal input to the synthesis filter whose filter coefficient is determined based on the coefficient, and an n-th order line for the signal stored in the signal storage unit N-order linear prediction analysis means 62 for performing prediction analysis to obtain an n-th prediction coefficient; second filter coefficient determination means 62a for determining a filter coefficient based on the n-th prediction coefficient; This is an auditory weighting filter having connection means for cascade-connecting the filters 63-1 and 63-2.

The auditory weighting filter 213 enables precise auditory weighting.

FIG. 6 is a block diagram showing an acoustic encoder AC3 according to still another embodiment of the present invention.

The acoustic coding apparatus AC3 is an embodiment corresponding to claim 6, wherein the modeling of the spectral envelope of the audio input signal is performed by linear prediction analysis, and the synthesized signal of the audio input signal and the coded code is synthesized. In the acoustic encoding / decoding device that determines the encoding code so as to minimize the difference signal from the audio signal and synthesizes the synthesized signal, a p-order linear prediction analysis is performed on the audio input signal. A p-order linear prediction analysis means 12 for obtaining a p-th prediction coefficient, a quantized prediction coefficient creation means 13 for quantizing the p-th prediction coefficient to generate a p-th quantization prediction coefficient, A first filter coefficient determining means 13a for determining a filter coefficient based on the quantized prediction coefficient of the above, and a filter coefficient is determined based on a p-th quantized prediction coefficient in a past frame of the acoustic input signal. A signal storage unit 15 for storing a signal input to the synthesis filter, and an n-th linear prediction analysis for performing an n-th linear prediction analysis on the signal stored in the signal storage unit to obtain an n-th prediction coefficient Means 43;
A second determining a filter coefficient based on a next prediction coefficient;
This is an acoustic coding apparatus having a filter coefficient determining means 43a and connection means for cascading the plurality of digital filters 44-1 and 44-2. Also, the audio encoding device AC3 can be applied to an audio decoding device.

According to the acoustic coding device AC3, there is an advantage that the spectrum can be synthesized precisely.

FIG. 7 is a block diagram showing a post filter 391 according to still another embodiment of the present invention. The post filter 391 is an improvement of the post filter 39 in the conventional audio decoding device AD1.
This is an embodiment corresponding to FIG.

The post-filter 391 is a sound decoding device that aurally suppresses quantization noise of a decoded synthesized signal obtained (synthesized) from an encoded code of an audio signal, and includes a past frame. The m-order linear prediction analysis means 71 for performing an m-order linear prediction analysis on the decoded synthesized signal to obtain an m-th prediction residual signal, and the m-order prediction obtained through the LPC inverse filter 72 An n-order linear prediction means 73 for performing an n-th linear prediction analysis on the residual signal to obtain an n-th prediction coefficient, and a first filter for determining a filter coefficient based on the n-th prediction coefficient A coefficient deciding means 73a, a second filter coefficient deciding means 32a for deciding a filter coefficient based on an nth-order linear prediction coefficient obtained from the encoded code, an envelope auditory weighting filter 74-1, A post filter in the acoustic decoding apparatus and a connection means for connecting the concrete perceptual weighting filter 74-2 to cascade.

According to the post filter 391, a post filter suitable for the auditory characteristics can be produced.

FIG. 8 is a block diagram showing a post filter 392 according to another embodiment of the present invention. The post filter 392 is an improvement of the post filter 39 in the conventional audio decoding device AD1, and is an embodiment corresponding to claim 8.

The post-filter 392 is an acoustic decoding device that aurally suppresses the quantization noise of the decoded synthesized signal synthesized from the encoded code of the audio signal. A signal storage unit 82 for storing a signal input to the synthesis filter whose filter coefficient is determined from the obtained p-th linear prediction coefficient, and performing an n-th linear prediction analysis on the stored signal; n-order linear prediction means 83 for obtaining an n-th prediction coefficient
First filter coefficient determining means 83a for determining a filter coefficient based on the n-th prediction coefficient, and determining a filter coefficient based on the n-th linear prediction coefficient obtained from the coded code in the current frame. A second filter coefficient determining means 32a for performing
4-1 and a fine structure auditory weighting filter 84-2;
This is a post filter having connection means for connecting filters 84-1 and 84-2 in cascade.

According to the post filter 392, a post filter suitable for the auditory characteristics can be produced.

[0072]

According to the present invention, a low-order linear prediction analysis is performed on a synthesized signal or an input signal, and a high-order linear prediction analysis is performed on the prediction residual signal. With the digital filter using the predicted coefficients, it is possible to obtain an effect that a more complex power spectrum characteristic can be represented as compared with the related art.

[Brief description of the drawings]

FIG. 1 (1) is a block diagram showing an acoustic encoding device AC1 according to an embodiment of the present invention, and FIG. 1 (2).
FIG. 14 is a block diagram showing an acoustic encoding device AC2 according to another embodiment of the present invention.

FIG. 2 is a diagram showing logarithmic power spectrum characteristics in the embodiment.

FIG. 3 is a block diagram showing another embodiment of the present invention;
It is a block diagram which shows the auditory weighting filter 211 which improved the auditory weighting filter 21 in the conventional audio encoding device AC4.

FIG. 4 is a block diagram showing an auditory weighting filter 212 according to another embodiment of the present invention. The auditory weighting filter 212 is a filter obtained by improving the auditory weighting filter 21 in the conventional audio encoding device AC4. This is an embodiment corresponding to item 3.

FIG. 5 is a block diagram showing an auditory weighting filter 213 according to another embodiment of the present invention. The auditory weighting filter 213 is a filter obtained by improving the auditory weighting filter 21 of the conventional audio encoding device AC4. This is an example corresponding to item 4.

FIG. 6 is a block diagram showing an acoustic encoding device AC3 according to still another embodiment of the present invention.

FIG. 7 is a block diagram showing a post filter 391 according to still another embodiment of the present invention. Post filter 39
Reference numeral 1 denotes an improved post filter 39 in the conventional audio decoding device AD1, which is an embodiment corresponding to claim 7.

FIG. 8 shows a post filter 39 according to another embodiment of the present invention.
FIG. 2 is a block diagram showing a second example. The post filter 392 is
Post filter 3 in conventional audio decoding device AD1
9 is an embodiment corresponding to claim 8.

FIG. 9 is a block diagram showing a conventional acoustic encoding device using the CELP method, FIG. 9 (1) is a block diagram showing a conventional acoustic encoding device AC4, and FIG. 9 (2) is a conventional acoustic encoding device. It is a block diagram showing acoustic coding device AC5.

FIG. 10 shows a conventional audio decoding apparatus A using the CELP method.
It is a block diagram showing D1.

[Explanation of symbols]

AC1, AC2, AC3: Acoustic coding device, 12: p-order linear prediction analysis means, 13: quantization means, 14: linear prediction synthesis filter, 15: adaptive codebook, 16: control means, 17, 23: gain application Means 18 Addition means 19 Subtraction means 21 Perceptual weighting filter 22 Noise codebook 41 51 m-order linear prediction analysis means 42 52 LPC inverse filter 43 53 n-th linear Prediction analysis means, 44: Linear prediction synthesis filter (n-order linear prediction synthesis filter and p-order linear prediction synthesis filter), AD1: acoustic decoding device, 54: Digital filter (envelope auditory weighting filter and fine structure auditory weighting filter).

────────────────────────────────────────────────── ─── Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI H03H 17/02 681 G10L 9/14 J H03M 7/30 9/18 E (72) Inventor Shinji Hayashi Shinjuku-ku, Tokyo No. 19-2, Nippon Telegraph and Telephone Corporation (56) References JP-A-5-257498 (JP, A) JP-A-6-250697 (JP, A) JP-A-62-1111299 (JP, A) JP-A-9-297597 (JP, A) JP-A-4-77799 (JP, A) JP-A-2-157800 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 19/06 G10L 19/04

Claims (8)

(57) [Claims] A p-order linear prediction analysis means for performing a p-order linear prediction analysis on an input signal sequence to obtain a p-th prediction coefficient; based on the p-order prediction coefficient, an all-pole or First filter coefficient determining means for determining a filter coefficient of a moving average type digital filter; and an n-th order signal sequence obtained by passing an input signal through an LPC inverse filter and obtaining a prediction residual of the p-th order linear prediction. Means for determining the filter coefficient of an all-pole or moving average digital filter based on the n-order prediction coefficient; and a filter for performing the p-order linear prediction. A first all-pole or moving average digital filter whose coefficients are determined; and a second all-pole or moving average digital filter whose filter coefficients are determined by the n-th linear prediction. A digital filter, comprising: a filter; and connection means for cascading the first filter and the second filter. 2. A method for modeling a spectral envelope of an audio input signal by linear predictive analysis, and determining the encoding code such that a difference signal between the audio input signal and a composite signal of the encoding code is minimized. In the acoustic encoding device for weighting the difference signal according to the auditory characteristics, a p-order linear prediction analysis is performed on the audio input signal to obtain a p-order prediction coefficient. Linear prediction analysis means; first filter coefficient determination means for determining a filter coefficient based on the p-order prediction coefficient;
N-order linear prediction analysis means for performing an n-th linear prediction analysis on a p-th linear prediction residual signal obtained through the C inverse filter to obtain an n-th prediction coefficient; A perceptual weighting filter for an acoustic coding apparatus, comprising: second filter coefficient determining means for determining a filter coefficient based on the prediction coefficient of (1); and connecting means for cascading the plurality of digital filters. 3. A model of a spectral envelope of an audio input signal is performed by linear prediction analysis, and the encoding code is determined so as to minimize a difference signal between the audio input signal and a composite signal of the encoding code. In the acoustic encoding device for weighting the difference signal according to the auditory characteristics, a p-order linear prediction analysis is performed on the audio input signal to obtain a p-order prediction coefficient. Linear prediction analysis means; first filter coefficient determination means for determining a filter coefficient based on the p-order prediction coefficient; m-th order synthesis signal for a past frame in the audio input signal; An m-th linear prediction analysis means for performing a linear prediction analysis to obtain an m-th prediction residual signal; and performing an n-th linear prediction analysis based on the m-th prediction residual signal to obtain an n-th prediction coefficient N-order linear prediction An acoustic code comprising: analysis means; second filter coefficient determination means for determining a filter coefficient based on the n-th prediction coefficient; and connection means for cascading the plurality of digital filters. Auditory weighting filter in the digitizer. 4. A model of a spectral envelope of an audio input signal is performed by linear prediction analysis, and the encoding code is determined such that a difference signal between the audio input signal and a composite signal of the encoding code is minimized. In the acoustic encoding device for weighting the difference signal according to the auditory characteristics, a p-order linear prediction analysis is performed on the audio input signal to obtain a p-order prediction coefficient. Linear prediction analysis means; first filter coefficient determination means for determining a filter coefficient based on the p-order prediction coefficient; filter coefficients based on the quantized prediction coefficient in a past frame of the audio input signal Signal storage means for storing the signal input to the synthesis filter for which n is determined; n-order linear prediction analysis is performed on the signal stored in the signal storage means to obtain an n-th prediction coefficient Linear prediction analysis means; second filter coefficient determination means for determining a filter coefficient based on the n-order prediction coefficient; and connection means for cascading the plurality of digital filters. Perceptual weighting filter in an acoustic encoding device. 5. A model of a spectral envelope of an audio input signal is performed by linear prediction analysis, and the encoding code is determined such that a difference signal between the audio input signal and a composite signal of the encoding code is minimized. In the audio encoding / decoding device for synthesizing the synthesized signal, a p-order linear prediction analysis unit for performing a p-order linear prediction analysis on the audio input signal to obtain a p-th prediction coefficient A quantized prediction coefficient creating means for quantizing the p-th order prediction coefficient to generate a p-th order quantization prediction coefficient; and a first filter for determining a filter coefficient based on the p-th order quantization prediction coefficient Coefficient determining means; m-order linear prediction analysis means for performing an m-th linear prediction analysis on a synthesized signal in a past frame to obtain an m-th prediction residual signal; and the m-th prediction residual signal For n-order linear prediction analysis There,
n-th linear prediction analysis means for obtaining an n-th prediction coefficient; second filter coefficient determination means for determining a filter coefficient based on the n-th prediction coefficient; connection for cascading the plurality of digital filters Means; and an audio encoding / decoding apparatus. 6. A model of a spectral envelope of an audio input signal is performed by linear prediction analysis, and the encoding code is determined so that a difference signal between the audio input signal and a composite signal of the encoding code is minimized. In the audio encoding / decoding device for synthesizing the synthesized signal, a p-order linear prediction analysis unit for performing a p-order linear prediction analysis on the audio input signal to obtain a p-th prediction coefficient A quantized prediction coefficient creating means for quantizing the p-th order prediction coefficient to generate a p-th order quantization prediction coefficient; and a first filter for determining a filter coefficient based on the p-th order quantization prediction coefficient Coefficient determination means; signal storage means for storing a signal input to a synthesis filter whose filter coefficient has been determined based on a p-th quantized prediction coefficient in a past frame of the audio input signal; Against the stored signal to the stage, n next performs linear prediction analysis, n
N-order linear prediction analysis means for obtaining the next prediction coefficient; second filter coefficient determination means for determining a filter coefficient based on the n-th prediction coefficient; connection means for cascading the plurality of digital filters And an audio encoding / decoding device. 7. An acoustic decoding device for acoustically suppressing quantization noise of a decoded synthesized signal obtained from an encoded code of an acoustic signal, comprising: , M
M-order linear prediction analysis means for performing the following linear prediction analysis to obtain an m-th prediction residual signal; and n-order linear prediction analysis for the m-th prediction residual signal obtained through the LPC inverse filter. And n-th linear prediction means for obtaining an n-th prediction coefficient; first filter coefficient determination means for determining a filter coefficient based on the n-th prediction coefficient; n obtained from the coding code Second filter coefficient determining means for determining a filter coefficient by the following linear prediction coefficient; an envelope auditory weighting filter; a fine structure auditory weighting filter;
Connection means for cascade-connecting the envelope auditory weighting filter and the fine structure auditory weighting filter. 8. An acoustic decoding apparatus for acoustically suppressing quantization noise of a decoded synthesized signal synthesized from an encoded code of an acoustic signal, wherein the decoded signal is obtained from the input code in a past frame. p
Signal storage means for storing a signal input to a synthesis filter for which a filter coefficient has been determined from the next linear prediction coefficient; performing n-order linear prediction analysis on the stored signal to obtain an n-th prediction coefficient N first-order linear prediction means for determining a filter coefficient based on the n-th prediction coefficient; and n-th linearity obtained from the coding code in the current frame. Second filter coefficient determining means for determining a filter coefficient based on a prediction coefficient; an envelope auditory weighting filter; a fine structure auditory weighting filter; And a post filter in the audio decoding device.