JPH0830299A - Voice coder - Google Patents

Abstract

PURPOSE:To reduce the chances of making a selection error during a preliminary selection and to improve the tone quality by conducting a closed loop processing employing sound source signals to make a preliminary selection of a delayed code. CONSTITUTION:A delayed code is tried by a closed loop preliminary selection delayed code trial circuit 400 and processes are conducted by a closed loop adaptive vector generating circuit 340 and a closed loop evaluation function computing circuit 350. A closed loop adaptive code vector 'ad', which corresponds to a delayed code 'd' that is set by the circuit 400, is generated from a closed loop adaptive code book 390. Thus, a delayed code candidate for a final selection is decided by a closed loop final selection delayed code trial circuit 410 based on the evaluation function computed by the circuit 350. Against the delayed code set by the circuit 410, a closed loop adaptive code book retrieval displayed by the circuits 340 and 350 and a deciding circuit 360 is executed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coding system, and more particularly to a voice coding system for coding a voice signal with a low bit rate, preferably about 8 to 4 kb / s with high quality. In particular, the present invention relates to a speech coding apparatus that reduces the calculation amount of a long-term prediction unit and improves coding accuracy.

[0002]

2. Description of the Related Art In recent years, there has been an urgent need to digitize car telephones and cordless telephones via radio. Since the frequency band that can be used in radio is small, it is important to develop a method for encoding a voice signal at a low bit rate.

As a method for encoding a voice signal at a low bit rate of about 8 to 4 kb / s, for example, the eye cast (ICASSP) by BA Atal of the United States is used.
"Code Excited Linear Prediction: High Quality At Low Bit Rate" from Proceedings (M. Schroeder and BA Atal,
“Code-excited liear prediction: High quality spee
ch at low bit rates ”, ICASSP proc. 85, pp.937 ~ 94
CELP (Code Excited LPC Coding) described in a paper entitled "0, 1985)" (hereinafter referred to as "Reference 1") and the like is known.

In this method, the encoding process is performed on the transmitting side in the following procedure. First, every frame (for example, 20ms)
Then, a short-term predictive code representing the frequency characteristic of the voice is extracted from the voice signal (short-term prediction).

Next, the frame is further divided into sub-frames (for example, 5 ms) of small sections.

For each subframe, an adaptive codebook composed of excitation signals (adaptive code vectors) of subframe lengths obtained by delaying the past excitation signals by the delay samples corresponding to the respective delay codes is used to represent pitch correlation delays. The sign is determined by the following procedure (long-term prediction).

The delay code is changed (tried) by the size of the adaptive codebook to extract the adaptive code vector corresponding to each delay code.

A synthesized signal is generated using the extracted adaptive code vector, and the error power with respect to the voice signal is calculated.

An optimum delay code that minimizes the calculated error power, an adaptive code vector that responds to the optimum delay code, and its gain are determined.

Next, the error power between the composite signal by the excitation code vector extracted from the noise signal (excitation codebook) which is a kind of quantized code prepared in advance and the residual signal obtained by long-term prediction is calculated. Determine the minimum source code vector and gain (source codebook search).

An index indicating the types of the determined adaptive code vector and excitation code vector, and an index indicating the type of gain and spectrum parameter of each excitation signal are transmitted.

Specifically, the search method for the delay code of the adaptive code vector and the quantization code of the excitation code vector is performed in the following procedure.

First, a signal z [n] obtained by weighting the input voice signal x [n] in terms of auditory perception and subtracting past influence signals
To calculate.

Next, the synthetic signal Hej [n] is driven by driving the synthetic filter H composed of the quantized and dequantized spectrum parameters obtained by the short-term prediction with the code vector ej [n] of the quantization code j. n] is calculated.

Then, the quantized code j that minimizes E in the following equation (1) representing the error power of the signal z [n] and the combined signal Hej [n] is obtained.

[0016]

[Equation 1]

Here, N _s is the subframe length, H is the matrix that realizes the synthesis filter, and g _ej is the code vector ej.
Represents the gain of each. Actually, the above equation (1) is given by
It is deployed like.

[0018]

[Equation 2]

Here, the numerator Cj in the above equation (2) is the cross correlation and the denominator Gj is the autocorrelation, which can be calculated by the following equations (3) and (4), respectively.

[0020]

(Equation 3)

[0021]

[Equation 4]

The autocorrelation Gj and the crosscorrelation Cj are calculated after calculating Hej [n] by driving the synthetic filter, that is, filtering.

In this case, the filtering process is executed (tried) for the codebook size (for example, when the number of bits of the sound source is 7, the filtering process is performed for 128 codes). Therefore, the amount of calculation (number of sums of products) for one frame to be processed becomes very large.

Therefore, as one method for reducing the amount of calculation at the time of long-term prediction, Japanese Patent Laid-Open No. 3-98099 by Stewed et al. (Title of the invention: "Digital voice coder and parameters used in the coder A method for obtaining a delay code, hereinafter referred to as "reference 2") is known.

This method is realized by the procedure shown in FIG. First, the delay code is set by the size of the adaptive codebook (300), and the code vector corresponding to the set delay code is adapted to the open loop adaptive codebook (370) in which past perceptually weighted speech signals are accumulated. Create more (310).

An evaluation function is calculated from the created code vector z _d (320), and several delay codes having a small evaluation function are extracted as code candidates (330).

The processing from 300 to 330 is called preselection by open loop processing.

Next, a code located near a code candidate selected by preliminary selection is selected (330), and for each code, a closed-loop adaptive codebook (390) in which past excitation signals are stored. ) Is used to create a code vector a _d (340).

An evaluation function is calculated using the created code vector (350), and a code that minimizes the evaluation function is extracted as an optimum code (360).

Since the open loop processing which does not use the result of the past coding does not include the filtering processing, the amount of calculation is much smaller than the method in which the closed loop processing is performed for all codes. In addition, it is possible to reduce large search errors by obtaining some candidates by open loop processing.

As described above, since the number of search codes by the closed loop can be effectively reduced by selecting the delay candidate by the preliminary selection by the open loop processing, the calculation amount of the long-term prediction can be performed without lowering the preliminary accuracy. Achieved reduction.

[0032]

However, in actuality, since the search accuracy of the search by the open loop processing is lower than that of the closed loop processing, selection error frequently occurs in the preliminary selection by the open loop processing.

An error in selecting a preliminary candidate for delay is one of the causes of sound quality deterioration.

Therefore, an object of the present invention is to solve the above problems and to provide a speech coder which realizes long-term prediction with a low calculation amount and high accuracy.

[0035]

Means for Solving the Problems (1) A speech coding apparatus according to the present invention, in a first aspect, speech analysis for determining a short-term predictive code representing a frequency characteristic of a speech signal for each constant section of the speech signal. Section, a signal accumulating section for accumulating signals in the past speech coding section, a delay code trial section for trying a delay code representing a pitch correlation of a speech signal, and a delay code by closed loop processing from the delay codes. A preliminary selection unit for long-term prediction for selecting a candidate, a main selection unit for long-term prediction for determining an optimum delay code from the candidates by closed-loop processing, and a residual signal determined by the optimum delay code An excitation codebook search unit that determines an optimum quantized code.

(2) The speech coding apparatus according to the present invention is the second
From the viewpoint of, a speech analysis unit that determines a short-term prediction code that represents the frequency characteristic of the speech signal for each constant section of the speech signal, a signal storage unit that accumulates signals in past speech coding sections, and a pitch of the speech signal. Preliminary selection delay code trial unit for making trial while thinning out the delay code indicating the correlation uniformly or nonuniformly, and a long-term prediction preliminary selection unit for selecting a delay code candidate by closed loop processing from the delay codes. , A long-term prediction main selection unit that determines an optimum delay code from the candidates by closed loop processing, and an excitation codebook search that determines an optimum quantization code for a residual signal determined by the optimum delay code Part.

(3) The speech coding apparatus according to the present invention is the third
From the viewpoint of, a voice analysis unit that determines a short-term predictive code that represents the frequency characteristic of the voice signal for each constant section of the voice signal, a signal storage unit that stores the signal in the past voice encoding section, A preselected delay code trial unit that selects only a code in a predetermined section from the delay codes representing the pitch correlation and tries the selected delay code, and a delay code candidate by closed loop processing from the delay codes. A long-term prediction pre-selection unit, a long-term prediction main selection unit that determines an optimum delay code from the candidates by closed-loop processing, and an optimum residual signal determined by the optimum delay code. An excitation codebook search unit that determines a quantization code.

(4) The speech coding apparatus according to the present invention is the fourth
From the viewpoint of, a speech analysis unit that determines a short-term prediction code that represents the frequency characteristic of the speech signal for each constant section of the speech signal, a signal storage unit that accumulates signals in past speech coding sections, and a pitch of the speech signal. A preselected delay code trial unit that selects only a code in a predetermined section from the delay codes representing the correlation and decimates the selected delay code uniformly or non-uniformly, and a closed loop from the delay codes Preliminary selection unit for long-term prediction for selecting a delay code candidate by processing, main selection unit for long-term prediction for determining an optimum delay code from the candidate by closed loop processing, and residual determined by the optimum delay code An excitation codebook search unit that determines the optimum quantization code for the signal.

(5) The speech coding apparatus according to the present invention is the fifth aspect.
From the viewpoint of, a speech analysis unit that determines a short-term prediction code that represents the frequency characteristic of the speech signal for each constant section of the speech signal, a signal storage unit that accumulates signals in past speech coding sections, and a pitch of the speech signal. A preselected delay code trial unit that selects only a code in a predetermined section of the delay codes representing the correlation and thins out the selected delay code uniformly or non-uniformly, and a closed loop from the delay codes. Preliminary selection unit for long-term prediction that selects a delay code candidate by processing, using the delay code candidate, select a delay code, delay code selection unit of the main selection to try, from among the delay code, A long-term prediction main selection unit that determines an optimum delay code by closed loop processing, a long-term prediction main selection unit that determines an optimum delay code from the delay codes, and the optimum delay Excitation codebook searching unit for determining an optimal quantization codes to residual signal determined by the sign, and a.

In addition, in the first to fourth aspects of the present invention, the delay code of the main selection is selected by using the delay code candidates selected by the preliminary selection unit of the long-term prediction, and the delay code of the main selection is tried. It is characterized by having a selection unit.

In the present invention, in the fifth aspect, preferably, the delay-delay code selecting section for main selection is
Among the delay code candidates, a delay code having the maximum evaluation function is determined, and a code near a code having a harmonic relationship with the determined delay code is tried as a code for main selection. . In the present invention, instead of determining one candidate, a plurality of candidates may be obtained by preliminary selection and a code near the delay of each candidate may be selected as the code for main selection.

Further, according to the present invention, in a speech coder including a long-term prediction circuit for outputting an optimum delay code of a delay code representing a pitch correlation of a speech signal, the long-term prediction circuit comprises:
A delay code section is selected and / or a delay code is thinned out to extract a delay code, and a closed loop search is performed on the extracted delay code to derive a preselected delay code. A speech coder configured to perform a closed loop search on a selected delay code to determine an optimum delay code.

Here, the closed loop search is to generate an adaptive code vector for a given delay code, generate a synthetic signal using the generated adaptive code vector, and generate a synthetic signal and a speech signal. Of the delay power is calculated, and the delay code that minimizes the calculated error power is determined. That is, the closed loop search is to perform a filtering process using a past sound source signal, and is different from an open loop search that does not require a filtering process because a past speech signal is used.

[0044]

The present invention realizes long-term prediction by two-stage selection of closed loop preliminary selection and closed loop main selection.
By performing the preselection using the closed loop processing, it is possible to realize the preselection with higher accuracy than the conventional open loop preselection.

Further, according to the present invention, at the time of closed loop preliminary selection, only the delay code of a part of the section (for example, only the intermediate value of the delay values is extracted from all the delay codes) is selected, Alternatively, by thinning out the selected codes uniformly or non-uniformly (for example, thinning every one short delay code, thinning every two long delay codes), without degrading accuracy,
A reduction in the amount of calculation can be realized.

[0046]

Embodiments of the present invention will be described below with reference to the drawings.

[0047]

First Embodiment A first embodiment of the present invention will be described below with reference to FIGS. 1 (A) is an encoding process, (B)
Represents a decoding process, and is composed of the same elements as those described in the conventional example.

FIG. 2 is a long-term prediction circuit for the encoding process of FIG.
It is a figure showing the details of 160 and showing the composition of the 1st example of the present invention.

First, each constituent module will be described with reference to FIG.

The input terminal 100 is a voice input terminal of the encoder. The buffer circuit 110 is a circuit that stores an audio signal. The LPC analysis circuit 120 is a circuit for extracting an LPC coefficient which is a spectrum parameter of voice. The parameter quantization circuit 130 is a circuit that quantizes the LPC coefficient.
The weighting circuit 140 is a circuit that performs perceptual weighting on the audio signal. Adaptive codebook 150 is a codebook that stores past sound source signals.

The long-term prediction circuit 160 is a circuit that searches for a delay code (adaptive code vector) representing pitch correlation.

The sound source codebook 170 is a codebook in which subframe-length sound source code vectors representing long-term prediction residuals are stored. The excitation codebook search circuit 180 is a circuit that determines an optimum quantized code (excitation code vector) from the excitation codebook.

The gain codebook 190 is a codebook in which parameters representing the gain terms of the adaptive code vector and the excitation code vector are accumulated. The gain codebook search circuit 200 is a circuit that determines the quantization gains of the adaptive code vector and the excitation code vector from the gain codebook.

The multiplexer 210 is a circuit that outputs a combination of code sequences.

The demultiplexer 220 is a circuit for decoding the encoded code into a code sequence.

The synthesizing filter 230 is a circuit for reproducing a sound signal from the generated sound source and sound synthesizing filter.

The output terminal 240 is an audio output terminal of the decoder.

Referring to FIG. 2, closed loop adaptive code vector generation circuit 340 generates a closed loop adaptive code for the set delay code d using the past sound source signal stored in closed loop adaptive code book 390. This is a circuit that generates a vector a _d .

Closed loop evaluation function calculation circuit 350
Is the audio signal z of the current frame stored in the audio buffer.
And enter the closed loop adaptive code vector a _d and calculates the correlation of the zero-state synthesis signal H · a _d, as the evaluation _{function, <z, H · a d} > 2 / <H · a d, H · a _This is a circuit for calculating _d >. Here, <,> represents a correlation.

The optimum delay code determination circuit 360 is a circuit that determines the optimum delay code that maximizes the evaluation function.

The audio buffer 380 is a buffer for accumulating the audio signal in the section in which the encoding process is performed, and a delay code that minimizes the error power with the audio signal is searched for.

Closed Loop Adaptive Codebook 390
Is a buffer for storing past sound source signals.

The closed loop preliminary selection delay code trial circuit 400 is a circuit that changes (tries) the delay code by the codebook size.

Closed loop main selection delay code trial circuit
Reference numeral 410 is a circuit that preselects a delay code from the evaluation function calculated by the closed loop evaluation function calculation circuit 350 and changes (tries) the selected delay code.

The flow of processing of this embodiment will be described with reference to FIG.

First, in the encoder processing of (A), the audio signal from the input port 100 is stored in the buffer 110 to which the audio signal is input. The LPC analysis circuit 120 performs short-term predictive analysis using the voice signal of a fixed sample stored in the buffer 110, and calculates the LPC coefficient representing the spectral characteristic of the voice signal.

The spectral parameter obtained by the LPC analysis is quantized by the parameter quantization circuit 130,
The quantized code of the LPC coefficient is sent to the multiplexer 210 and is dequantized and used for the subsequent coding processing.

The voice signal stored in the buffer 110 is weighted by using a quantized / dequantized LPC coefficient.
It is perceptually weighted at 140 and used for subsequent codebook searches.

A codebook search is performed using the adaptive codebook, sound source codebook, and gain codebook.

First, the long-term prediction circuit 160 first performs long-term prediction, determines the optimum delay code representing the pitch correlation, transfers the code (delay code) to the multiplexer 210, and generates an adaptive code vector. .

Next, after subtracting the effect of the obtained adaptive code vector, the excitation codebook search circuit 180 performs an excitation codebook search to determine a quantization code, generate an excitation code vector, and generate the code ( Quantized code) to the multiplexer 210.

After the adaptive code vector and the sound source code vector are obtained, the gain codebook search circuit 200 outputs 2
Calculates the gain of one sound source and multiplexes its code
Transfer to 210. The multiplexer 210 combines each code, converts it into a transmission code, and outputs it.

In the decoder processing of FIG. 1B, the demultiplexer decomposes the input transmission code into each code. L
The filter coefficient is decoded from the code representing the PC coefficient and transferred to the synthesis filter 230.

An adaptive code vector is generated from the delay code by using the adaptive code book 150. A sound source code vector is generated using a sound source code book 170 from a quantized code representing a sound source code.

The gains of the adaptive code vector and the sound source code vector are calculated from the gain code, and each sound source is multiplied by the gain term to generate the input signal of the synthesis filter. Finally, the synthesis filter 230 synthesizes the audio signal using the input signal.

Next, this embodiment will be described with reference to FIG. This corresponds to the long-term prediction circuit 160 of FIG.

First, the closed loop preliminary selection delay code trial circuit 400 tries a delay code, and the closed loop adaptive vector generation circuit 340 and the closed loop evaluation function calculation circuit 350 perform the processing for each delay code.

Closed loop adaptive vector generation circuit 34
At 0, the closed loop adaptive code vector a _d corresponding to the delay code d set by the closed loop preliminary selection delay code trial circuit 400 is generated from the closed loop adaptive codebook 390.

Next, a closed loop evaluation function calculation circuit
The evaluation function is calculated at 350.

According to the evaluation function calculated by the closed loop evaluation function calculation circuit 350, delay code candidates for main selection are closed loop main selection delay code trial circuit 41.
Decide with 0.

Closed loop main selection delay code trial circuit
A general closed-loop adaptive codebook search represented by a closed-loop adaptive vector generation circuit 340, a closed-loop evaluation function calculation circuit 350, and an optimum delay code determination circuit 360 is performed on the delay code set in 410.

[0082]

Second Embodiment A second embodiment of the present invention will be described with reference to FIGS.

FIG. 3 shows details of the long-term prediction circuit 160 shown in FIG. 1, and is a block diagram of the second embodiment of the present invention.

Table 1 is a table showing a specific example of delay code selection in the closed loop preliminary selection delay code trial circuit 401.

Each constituent module will be described with reference to FIG.

The closed loop preliminary selection delay code trial circuit 401 is a circuit which changes (tries) delay codes while thinning out the delay code uniformly or nonuniformly.

The other parts are the same as those of the first embodiment shown in FIG.
Is the same as

Since the overall processing flow is the same as that of the first embodiment, only the processing of the long-term prediction circuit 160 shown in the block diagram of FIG. 3 will be described below.

First, the closed loop preliminary selection delay code trial circuit 401 makes the delay code try while thinning out the delay code uniformly or non-uniformly, and the processing of the closed loop adaptive vector generation circuit 340 and the closed loop evaluation function calculation circuit 350 for each delay code. I do.

As a method of selecting a delay code in the closed loop preliminary selection delay code trial circuit 401, for example, a method using a code indicated by "○ in the case of selection" in Table 1, that is, a code having a short delay value is A method of thinning out every two codes and a method of thinning out every third code for a long code are used.

By such thinning-out, the selection code of 128 is reduced to 40.

In the closed-loop adaptive code vector generation circuit 340, the closed-loop adaptive code vector a _d corresponding to the delay code d set by the closed-loop preliminary selection delay code trial circuit 401 is generated from the closed-loop adaptive codebook 390.

Next, a closed loop evaluation function calculation circuit
The evaluation function is calculated at 350. According to the evaluation function calculated by the closed loop evaluation function calculation circuit 350, the delay code candidate for main selection is determined by the closed loop main selection delay code trial circuit 410.

Closed loop main selection delay code trial circuit
A general closed-loop adaptive codebook search represented by a closed-loop adaptive code vector generation circuit 340, a closed-loop evaluation function calculation circuit 350, and an optimum delay code determination circuit 360 is performed on the delay code set in 410.

[0095]

Third Embodiment A third embodiment of the present invention will be described with reference to FIGS.

FIG. 4 shows the details of the long-term prediction circuit 160 shown in FIG. 1, and is a block diagram of the third embodiment of the present invention.

Table 2 is a table showing a specific example of delay code selection in the closed loop preliminary selection delay code trial circuit 402.

First, each constituent module will be described with reference to FIG.

The closed loop preliminary selection delay code trial circuit 402 selects only the code of a part of the delay code,
It is a circuit to change (trial). Since the other parts are the same as those in the first embodiment, the description thereof will be omitted.

Since the overall processing flow is the same as that of the first embodiment, the long-term prediction circuit 160 shown in the block diagram of FIG.
Only the processing of will be described.

First, in the closed loop preliminary selection delay code trial circuit 402, only a part of the delay code is selected, the selected code is tried, and the closed loop adaptive code vector generation circuit 340 The processing of the loop evaluation function calculation circuit 350 is performed.

As a method of selecting a delay code in the closed loop preliminary selection delay code trial circuit 402, for example, Table 2
The method of using the code indicated by "○ in the case of selection" is used, that is, the method of selecting only the code having the delay value between 40 and 59.5.

This selection reduces the selection code of 128 to 40.

In the closed-loop adaptive code vector generation circuit 340, the closed-loop adaptive code vector a _d corresponding to the delay code d set by the closed-loop preliminary selection delay code trial circuit 402 is generated from the closed-loop adaptive codebook 390.

Next, a closed loop evaluation function calculation circuit
The evaluation function is calculated at 350. According to the evaluation function calculated by the closed loop evaluation function calculation circuit 350, the delay code candidate for main selection is determined by the closed loop main selection delay code trial circuit 410.

Closed loop main selection delay code trial circuit
A general closed-loop adaptive codebook search represented by a closed-loop adaptive code vector generation circuit 340, a closed-loop evaluation function calculation circuit 350, and an optimum delay code determination circuit 360 is performed on the delay code set in 410.

[0107]

Fourth Embodiment With reference to FIGS. 1 and 5, a fourth embodiment of the present invention will be described below.

FIG. 5 shows the details of the long-term prediction circuit 160 of FIG. 1 and is a block diagram of the fourth embodiment of the present invention.

Table 3 is a table showing a specific example of delay code selection in the closed loop preliminary selection delay code trial circuit 403.

First, each constituent module will be described with reference to FIG.

The closed loop preliminary selection delay code trial circuit 403 selects only the code of a part of the delay code,
This is a circuit that thins out or changes (tries) the signs of the selected section uniformly or non-uniformly. The other parts are the same as those in the first embodiment.

Since the entire processing flow is the same as that of the first embodiment, the long-term prediction circuit 160 shown in the block diagram of FIG.
Only the processing of will be described.

First, the closed-loop preliminary selection delay code trial circuit 403 selects only a part of the delay code, thins out the selected codes uniformly or non-uniformly for trial, and closes the closed-loop adaptive code for each delay code. The vector generation circuit 340 and the closed loop evaluation function calculation circuit 350 are processed.

As a method of selecting a delay code in the closed loop preliminary selection delay code trial circuit 403, for example, a method using a code indicated by "○ in case of selection" in Table 3, that is, a delay value of 40 to 59.5 is used. Only the codes that take the value of are selected and the method of thinning out each code is used.

This selection reduces the selection code of 128 to 20.

In the closed-loop adaptive code vector generation circuit 340, the closed-loop adaptive code vector a _d corresponding to the delay code d set by the closed-loop preliminary selection delay code trial circuit 403 is generated from the closed-loop adaptive codebook 390.

Next, a closed loop evaluation function calculation circuit
The evaluation function is calculated at 350. According to the evaluation function calculated by the closed loop evaluation function calculation circuit 350, the delay code candidate for main selection is determined by the closed loop main selection delay code trial circuit 410.

Closed loop main selection delay code trial circuit
A general closed-loop adaptive codebook search represented by a closed-loop adaptive code vector generation circuit 340, a closed-loop evaluation function calculation circuit 350, and an optimum delay code determination circuit 360 is performed on the delay code set in 410.

[0119]

Fifth Embodiment A fifth embodiment of the present invention will be described below with reference to FIGS.

FIG. 6 shows the details of the long-term prediction circuit 160 of FIG. 1 and is a block diagram of the fifth embodiment of the present invention.

Table 4 is a table showing a specific example of delay code selection in the closed loop preliminary selection delay code trial circuit 404.

First, each constituent module will be described.

The closed loop preliminary selection delay code trial circuit 404 selects only the code of a part of the delay code,
This is a circuit that thins out or changes (tries) the signs of the selected section uniformly or non-uniformly.

Closed loop main selection delay code trial circuit
A delay code 414 determines a delay code that maximizes the evaluation function calculated by the closed-loop evaluation function calculation circuit 350. The delay code has a harmonic relationship with the determined delay code. This is a circuit that causes a code in the vicinity to be tried as a code for main selection. Other parts are the same as those in the first embodiment.
Is the same as

Since the entire processing flow is the same as that of the first embodiment, the long-term prediction circuit 160 shown in the block diagram of FIG.
Only the processing of will be described.

First, the closed-loop preliminary selection delay code trial circuit 404 selects only a part of the delay code, and all the selected codes are thinned out uniformly or nonuniformly for trial.
The processing of the closed loop adaptive code vector generation circuit 340 and the closed loop evaluation function calculation circuit 350 is performed for each delay code.

As a method of selecting a delay code in the closed loop preliminary selection delay code trial circuit 404, for example, a method using a code indicated by "○ in the case of selection" in Table 4, that is, a delay value of 39 to 63 is used. Only the code that takes the value of is selected, the code having a short delay value is thinned out by one code, and the code having a long delay value is thinned out by two codes.

This selection reduces the code of 128 to 20.

In the closed-loop adaptive code vector generation circuit 340, the closed-loop adaptive code vector a _d corresponding to the delay code d set by the closed-loop preliminary selection delay code trial circuit 404 is generated from the closed-loop adaptive codebook 390.

Next, a closed loop evaluation function calculation circuit
The evaluation function is calculated at 350. According to the evaluation function calculated by the closed loop evaluation function calculation circuit 350, the delay code candidate for the main selection is determined by the closed loop main selection delay code trial circuit 414.

As a selection method, one candidate is selected in the preliminary selection, and a delay neighborhood having a harmonic relationship of the candidate delay (integral multiple and integral fraction) is selected.

Closed loop main selection delay code trial circuit
A general closed-loop adaptive codebook search represented by a closed-loop adaptive code vector generation circuit 340, a closed-loop evaluation function calculation circuit 350, and an optimum delay code determination circuit 360 is performed on the delay code set in 414.

[0133]

Sixth Embodiment A sixth embodiment of the present invention will be described below with reference to FIGS.

FIG. 7 shows details of the long-term prediction circuit 160 shown in FIG. 1, and is a configuration diagram of the sixth embodiment of the present invention. First, each constituent module will be described.

The closed loop preliminary selection delay code trial circuit 405 is the closed loop preliminary selection delay code trial circuit 400, 401, 40 described in the first, second, third, and fourth embodiments.
It is a circuit showing either 2, 403. The other parts are the same as in the fifth embodiment.

Since the entire processing flow is the same as that of the fifth embodiment, the long-term prediction circuit 160 shown in the block diagram of FIG.
Only the processing of will be described.

First, in the closed loop preliminary selection delay code trial circuit 405, the delay code is selected by any of the methods of the modules 400, 401, 402 and 403 shown in the first, second, third and fourth embodiments. , And the closed loop adaptive code vector generation circuit 340 and the closed loop evaluation function calculation circuit 350 are processed for each delay code.

In the closed-loop adaptive code vector generation circuit 340, the closed-loop adaptive code vector a _d corresponding to the delay code d set by the closed-loop preliminary selection delay code trial circuit 405 is generated from the closed-loop adaptive codebook 390.

Next, a closed loop evaluation function calculation circuit
The evaluation function is calculated at 350.

According to the evaluation function calculated by the closed loop evaluation function calculation circuit 350, delay code candidates for main selection are closed loop main selection delay code trial circuit 41.
Determined by 4.

As a selection method, a method of selecting one candidate in the preliminary selection and selecting a neighborhood of a delay having a harmonic relationship with the candidate delay (integral multiple and integral fraction) is selected.

Closed loop main selection delay code trial circuit
A general closed-loop adaptive codebook search represented by a closed-loop adaptive code vector generation circuit 340, a closed-loop evaluation function calculation circuit 350, and an optimum delay code determination circuit 360 is performed on the delay code set in 414.

As described above, according to each of the above-described embodiments, the closed loop preliminary selection selects only a part of the delay code, or thins out uniformly or non-uniformly to reduce the amount of calculation without lowering the accuracy. Can be reduced.

As an example, when 30 delay codes are selected and closed loop preselection is performed, the SNR (signal noise ratio) which is 0.1 dB higher than that when 60 delay codes are used in the conventional open loop is obtained. Has been obtained.

Although (cross-correlation) ² / (auto-correlation) is used as the evaluation function in each of the above embodiments, the same effect can be obtained with (cross-correlation) ² or the energy of the error signal.

Further, in each of the above embodiments, the signal stored in the audio buffer is the zero-state subtraction signal z, but the same effect can be obtained by using the input audio signal, the residual signal, and the perceptual weighting signal.

Further, in each of the above embodiments, the method of performing the filtering with the synthesis filter H for the calculation of the cross-correlation and the auto-correlation is used, but the similar effect can be obtained by the method using the approximate expression. As an approximation method, for example, “Efficient Proceeders For Finding The Optimum Innovation In” by I.C.SP (ICASSP) Proceeding by Transco (IM. Transco) and BS Atal (US) can be used.
Stochastic Coders "(EFFICIENTPROCEDURE
S FOR FINDING THE OPTIMUM INNOVATION IN STOCASTIC
CORDERS, IM.Transco, BS Atal ICASSP86, pp.2375
~ 2378, 1986) and the like.

In each of the above embodiments, the optimum delay code is limited to one. However, even if a plurality of candidates are obtained and main selection is performed in the next step (excitation codebook search or gain codebook search), The same effect can be obtained by performing the simultaneous optimum search in the subsequent codebook search.

Further, in each of the above-mentioned embodiments, the delay code is described as the decimal point delay, but the same effect can be obtained even if the delay code is limited to the integer delay.

Further, although the sound source codebook is not specifically defined in each of the above embodiments, the same effect can be obtained with a noise codebook or a learning codebook learned by a vector quantization (VQ) algorithm. To be Reference 1 can be referred to for the former. In addition, the latter is
3-227087 (Japanese Patent Application No. 2-22955) and Japanese Patent Laid-Open No. 3-226704
Reference can be made to, for example, Japanese Patent Publication (Japanese Patent Application No. 2-22956).

Furthermore, in the sixth embodiment, the method of selecting the delay for the main selection has been explained by selecting one delay candidate and selecting the code near the delay code having the harmonic relationship. However, a similar effect can be obtained by a method of obtaining a plurality of candidates by preliminary selection, which is conventionally performed, and selecting a code near each candidate delay.

Although the above embodiments have been described using the LPC analysis circuit, the same effect can be obtained in other analysis methods such as the BURG method for extracting the spectrum parameter.

Although the above embodiments have been described by using the LPC coefficient, it is obvious that the same effect can be obtained by using other spectral parameters such as PARCOR coefficient and LSP coefficient.

Further, in each of the above embodiments, the excitation codebook search circuit has a one-stage configuration, but it is also clear that a similar effect can be obtained even with a multi-stage configuration.

[0155]

[Table 1]

[0156]

[Table 2]

[0157]

[Table 3]

[0158]

[Table 4]

[0159]

As described above, according to the present invention, long-term prediction is realized by two-step selection of preselection by closed loop processing and main selection by closed loop processing. By performing the closed loop processing using, the effect of improving the sound quality by reducing the selection error at the time of the preliminary selection as compared with the conventional preliminary selection by the open loop processing using the audio signal is provided.

According to the present invention, as compared with the conventional example in which the preselection is performed by the open loop processing, the amount of calculation is substantially the same, and more accurate pitch estimation can be performed. Deterioration can be suppressed.

Further, according to the present invention, in the closed loop preselection, delay codes are thinned out uniformly or nonuniformly, so that the calculation amount can be reduced without lowering the estimation accuracy.

Further, according to the present invention, in the closed loop preliminary selection, only a part of the delay code section is selected, so that the calculation amount can be reduced without lowering the accuracy.

Furthermore, according to the present invention, by selecting only a part of the interval of the delay code representing the pitch correlation of the voice signal and making the selected delay code try while decimating it uniformly or non-uniformly, The amount of calculation can be reduced without lowering the accuracy.

Further, according to the present invention, one delay candidate is selected as a delay selection method for main selection of long-term prediction,
In the case of a method of selecting a code near a delay code having a harmonic relationship with this, or by obtaining a plurality of candidates by preliminary selection,
The same effect as above can be obtained by a method of selecting a code near each candidate delay as a code for main selection.

As an example of the quantitative effect of the present invention, when 30 delay codes are selected and closed loop preliminary selection is performed, about 0.1 dB is obtained as compared with the case where 60 delay codes are used in the conventional open loop. A high SNR (signal noise ratio) is obtained.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of a speech encoding process of the present invention.
(A) represents encoder processing. (B) represents a decoder process.

FIG. 2 is a configuration diagram of a first embodiment of the present invention.

FIG. 3 is a configuration diagram of a second embodiment of the present invention.

FIG. 4 is a configuration diagram of a third embodiment of the present invention.

FIG. 5 is a configuration diagram of a fourth embodiment of the present invention.

FIG. 6 is a configuration diagram of a fifth embodiment of the present invention.

FIG. 7 is a configuration diagram of a sixth embodiment of the present invention.

[Figure 8] Conventional long-term prediction (open-closed search)
FIG.

[Explanation of symbols]

100 voice input port 110 buffer circuit 120 LPC analysis circuit 130 parameter quantization circuit 140 weighting circuit 150 adaptive codebook 160 long-term prediction circuit 170 sound source codebook 180 sound source codebook search circuit 190 gain codebook 200 gain codebook search circuit 210 multiplexer 220 Demultiplexer 230 Synthesis filter 240 Voice output terminal 300 Open loop delay code trial circuit 310 Open loop adaptive code vector generation circuit 320 Open loop evaluation function calculation circuit 330 Closed loop delay code trial circuit 340 Closed loop adaptive code vector generation circuit 350 Closed loop evaluation Function calculation circuit 360 Optimal delay code determination circuit 370 Open loop adaptive codebook 380 Voice buffer 390 Closed loop adaptive codebook 400, 401, 402, 403, 404, 405 Closed loop preliminary selection Delay code trial circuit 410 closed-loop main selection delay code trial circuit 414 closed-loop main selection delay code trial circuit

Claims (9)

[Claims] 1. A speech analysis unit that determines a short-term prediction code that represents the frequency characteristic of the speech signal for each fixed section of the speech signal, a signal storage unit that accumulates signals in past speech coding sections, and A delay code trial unit that tries a delay code that represents a pitch correlation, a long-term prediction preliminary selection unit that selects a delay code candidate by a closed loop process from the delay codes, and an optimum by the closed loop process from the candidates. Speech having a long-term prediction main selection unit that determines a delay code, and an excitation codebook search unit that determines an optimum quantization code for a residual signal determined by the optimum delay code Encoding device. 2. A speech analysis unit for determining a short-term predictive code representing a frequency characteristic of the speech signal for each constant section of the speech signal, a signal storage unit for accumulating signals in past speech coding sections, and Preliminary selection delay code trial unit for making a trial while thinning out the delay code representing the pitch correlation uniformly or non-uniformly, and a long-term prediction preliminary selection unit for selecting a delay code candidate from the delay codes by closed loop processing. A long-term prediction main selection unit that determines an optimum delay code from the candidates by closed-loop processing; and an excitation codebook that determines an optimum quantization code for the residual signal determined by the optimum delay code. A speech coding apparatus comprising: a search unit. 3. A voice analysis unit for determining a short-term predictive code representing a frequency characteristic of the voice signal for each constant section of the voice signal, a signal storage unit for storing a signal in a past voice encoding section, A preselected delay code trial unit that selects only a code in a predetermined section from the delay codes representing the pitch correlation and tries the selected delay code, and a delay code candidate by closed loop processing from the delay codes. A long-term prediction pre-selection unit for selecting, a long-term prediction main selection unit that determines an optimum delay code from the candidates by closed-loop processing, and an optimum for a residual signal determined by the optimum delay code. A speech coding apparatus, comprising: an excitation codebook searching unit that determines a quantized code. 4. A voice analysis unit for determining a short-term predictive code representing a frequency characteristic of the voice signal for each constant section of the voice signal, a signal storage unit for storing a signal in a past voice coding section, and A preselected delay code trial unit that selects only a code in a predetermined section from the delay codes representing the pitch correlation and tries while thinning out the selected delay code uniformly or non-uniformly; Preliminary selection unit for long-term prediction that selects a delay code candidate by loop processing, main selection unit for long-term prediction that determines an optimum delay code from the candidates by closed loop processing, and residue determined by the optimum delay code An excitation codebook search unit that determines an optimum quantized code for a difference signal. 5. A speech analysis unit that determines a short-term prediction code that represents the frequency characteristic of the speech signal for each fixed section of the speech signal, a signal storage unit that accumulates signals in past speech coding sections, and A preselected delay code trial unit that selects only a code in a predetermined section from the delay codes representing the pitch correlation and tries while thinning out the selected delay code uniformly or non-uniformly; A long-term prediction preliminary selection unit that selects a delay code candidate by loop processing, a delay code selection unit that uses the delay code candidate to select and try a delay code, and a delay code selection unit from among the delay codes A main selection unit for long-term prediction that determines an optimum delay code by closed loop processing, and an optimum quantization code for a residual signal determined by the optimum delay code Source codebook searching unit, the speech coding apparatus characterized by having a. 6. The delay code selection unit of main selection for selecting a delay code using a candidate of the delay code selected by the preliminary selection unit for long-term prediction and trying the delay code, according to claim 1. The audio encoding device according to. 7. The delay code selection unit of the main selection determines a delay code having a maximum evaluation function among the delay code candidates, and selects a delay code near a code having a harmonic relationship with the determined delay code. 7. The speech coding apparatus according to claim 5, wherein the code is tried as a code for main selection. 8. The main selection delay code selection unit obtains a plurality of candidates of the delay code candidates by preliminary selection, and causes a code near each candidate delay to be tried as a main selection code. The audio encoding device according to claim 5 or 6. 9. A speech coding apparatus including a long-term prediction circuit for outputting an optimum delay code of a delay code representing pitch correlation of a speech signal, wherein the long-term prediction circuit selects a delay code section and / or selects a delay code. The delay code is extracted by thinning out, a closed loop search is performed on the extracted delay code to derive a preselected delay code, and a closed loop search is performed on the preselected delay code to optimize the delay code. A speech coding apparatus configured to determine a delay code.