JP3746067B2 - Speech decoding method and speech decoding apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reproduce a high quality speech with a small data amount in speech coding and decoding for performing compression coding of a speech signal to a digital signal. <P>SOLUTION: In a code-excited linear prediction (CELP) speech coding, a noise level of the speech in a concerning coding period is evaluated by using a code or coding result of at least one of spectrum information, power information, and pitch information, and various excitation codebooks 19 and 20 are used in response to an evaluation result. <P>COPYRIGHT: (C)2006,JPO&amp;NCIPI

Description

  The present invention relates to a speech encoding / decoding method and speech encoding / decoding device used when compression encoding / decoding an audio signal into a digital signal, and particularly for reproducing high-quality audio at a low bit rate. The present invention relates to a speech encoding method, a speech decoding method, a speech encoding device, and a speech decoding device.

  Conventionally, Code-Excited Linear Prediction (CELP) coding has been a typical high-efficiency speech coding method. For the technique, “Code-excited linear prediction (CELP): High-quality speech at very low bit rates "(MRShroeder and BSAtal, ICASSP '85, pp.937-940, 1985).

  FIG. 6 shows an example of the overall configuration of the CELP speech encoding / decoding method. In FIG. 6, 101 is an encoding unit, 102 is a decoding unit, 103 is multiplexing means, and 104 is separation means. The encoding unit 101 includes a linear prediction parameter analysis unit 105, a linear prediction parameter encoding unit 106, a synthesis filter 107, an adaptive codebook 108, a drive codebook 109, a gain encoding unit 110, a distance calculation unit 111, and a weighted addition unit 138. It is configured. The decoding unit 102 includes a linear prediction parameter decoding unit 112, a synthesis filter 113, an adaptive codebook 114, a driving codebook 115, a gain decoding unit 116, and a weighted addition unit 139.

  In CELP speech coding, about 5 to 50 ms is defined as one frame, and the speech of the frame is divided into spectrum information and sound source information and coded. First, the operation of the CELP speech encoding method will be described. In the encoding unit 101, the linear prediction parameter analysis means 105 analyzes the input speech S101 and extracts linear prediction parameters that are speech spectrum information. The linear prediction parameter encoding means 106 encodes the linear prediction parameter and sets the encoded linear prediction parameter as a coefficient of the synthesis filter 107.

Next, encoding of sound source information will be described. The adaptive codebook 108 stores past drive excitation signals, and outputs a time-series vector in which past drive excitation signals are periodically repeated corresponding to the adaptive code input from the distance calculation means 111. The drive codebook 109 stores, for example, a plurality of time series vectors configured by learning so as to reduce the distortion between the learning speech and the encoded speech, and the drive input from the distance calculation unit 111. A time series vector corresponding to the code is output.
The time series vectors from the adaptive codebook 108 and the drive codebook 109 are weighted and added by the weighting addition means 138 according to the respective gains given from the gain encoding means 110, and the addition result is synthesized as a drive excitation signal. Supplied to the filter 107 to obtain encoded speech. The distance calculation unit 111 obtains the distance between the encoded speech and the input speech S101, and searches for an adaptive code, a drive code, and a gain that minimize the distance. After the encoding is completed, the code of the linear prediction parameter, the adaptive code that minimizes the distortion between the input speech and the encoded speech, the drive code, and the gain code are output as the encoding result.

Next, the operation of the CPEL speech decoding method will be described.
On the other hand, in the decoding unit 102, the linear prediction parameter decoding unit 112 decodes the linear prediction parameter from the code of the linear prediction parameter and sets it as a coefficient of the synthesis filter 113. Next, adaptive codebook 114 outputs a time series vector in which past drive excitation signals are periodically repeated corresponding to the adaptive code, and drive codebook 115 outputs a time series vector corresponding to the drive code. To do. These time series vectors are weighted and added by weighting and adding means 139 according to the respective gains decoded from the gain code by the gain decoding means 116, and the addition result is supplied to the synthesis filter 113 as a driving sound source signal. The output voice S103 is obtained.

  As a conventional speech coding / decoding method improved for the purpose of improving playback speech quality by the CELP speech coding / decoding method, “Phonetically-based vector excitation coding of speech at 3.6 kbps” (S. Wang and A. Gersho, ICASSP '89, pp.49-52, 1989). FIG. 7 in which the same reference numerals are assigned to the means corresponding to FIG. 6 shows an example of the overall configuration of this conventional speech coding / decoding method. In FIG. The code book switching means, 119 is a first drive code book, and 120 is a second drive code book. In the decoding means 102 in the figure, 121 is a drive codebook switching means, 122 is a first drive codebook, and 123 is a second drive codebook. The operation of the encoding / decoding method having such a configuration will be described. First, in the encoding unit 101, the voice state determination unit 117 analyzes the input voice S101 and determines whether the voice state is one of two states, voiced / unvoiced, for example. The drive codebook switching means 118 uses the first drive codebook 119 for coding, for example, if voiced, and the second drive codebook 120 for coding, if it is unvoiced, according to the voice state determination result. The book is switched, and which drive codebook is used is encoded.

  Next, in the decoding means 102, the drive codebook switching means 121 is assumed to use the same drive codebook used in the encoding means 101 according to the code of which drive codebook was used in the encoding means 101. The first drive codebook 122 and the second drive codebook 123 are switched. By configuring in this way, a drive codebook suitable for encoding is prepared for each state of speech, and the quality of reproduced speech is improved by switching the drive codebook according to the state of input speech can do.

Japanese Patent Laid-Open No. 8-185198 discloses a conventional speech coding / decoding method for switching a plurality of driving codebooks without increasing the number of transmission bits. In this method, a plurality of drive codebooks are switched and used in accordance with the pitch period selected in the adaptive codebook. As a result, it is possible to use a drive codebook adapted to the characteristics of the input speech without increasing transmission information.
"Code-excited linear prediction (CELP): High-quality speech at very low bit rates" (MRShroeder and BSAtal, ICASSP '85, pp.937-940, 1985) Phonetically-based vector excitation coding of speech at 3.6kbps "(S.Wang and A.Gersho, ICASSP '89, pp.49-52, 1989) JP-A-8-185198

  As described above, in the conventional speech coding / decoding method shown in FIG. 6, synthesized speech is generated using a single drive codebook. In order to obtain high-quality encoded speech even at a low bit rate, the time-series vectors stored in the drive codebook are non-noise including many pulses. For this reason, when noisy speech such as background noise or frictional consonants is encoded and synthesized, the encoded speech has a problem of generating unnatural sounds such as harshness and dust. This problem can be solved if the driving codebook is composed only of noisy time-series vectors, but the quality of the entire encoded speech deteriorates.

  In the improved conventional speech encoding / decoding method shown in FIG. 7, a plurality of driving codebooks are switched according to the state of input speech to generate encoded speech. Thus, for example, when the input speech is noisy unvoiced parts, a driving codebook composed of noisy time series vectors is used, and for other voiced parts, a driving codebook composed of non-noisy time series vectors is used. Therefore, even if a noisy voice is encoded and synthesized, an unnatural grim sound is not emitted. However, since the same driving codebook is used on the decoding side as that on the encoding side, it is necessary to newly encode and transmit which driving codebook is used, which hinders the reduction of the bit rate. There was a problem.

  In the conventional speech coding / decoding method for switching a plurality of driving codebooks without increasing the number of transmission bits, the driving codebook is switched according to the pitch period selected in the adaptive codebook. However, the pitch period selected in the adaptive codebook is different from the pitch period of the actual speech, and it is not possible to determine whether the state of the input speech is noisy or non-noisy from its value alone. The problem that the encoded speech is unnatural is not solved.

  The present invention has been made to solve such a problem, and provides an audio encoding / decoding method and apparatus for reproducing high-quality audio even at a low bit rate.

The speech decoding method according to the present invention is a code-driven linear prediction (CELP) speech decoding method that synthesizes speech based on an input speech code using a drive codebook and an adaptive codebook.
A gain decoding step of acquiring a gain by multiplying a time series vector output from the adaptive codebook by decoding the input speech code;
A time-series vector generation step of generating a second time-series vector in which the number of samples having an amplitude value of zero is different from the first time-series vector by changing the first time-series vector output from the drive codebook When,
A selection step of selecting the first or second time series vector based on the gain;
A speech synthesis step of synthesizing speech using the time-series vector selected in the selection step;
It is characterized by having.

A speech decoding apparatus according to the present invention is a code-driven linear prediction (CELP) speech decoding apparatus that synthesizes speech based on an input speech code using a drive codebook and an adaptive codebook.
Gain decoding means for acquiring a gain multiplied by a time-series vector output from the adaptive codebook by decoding the input speech code;
Time series vector generating means for changing the first time series vector output from the drive codebook and generating a second time series vector in which the number of samples having an amplitude value of zero is different from the first time series vector When,
Selecting means for selecting the first or second time series vector based on the gain;
Speech synthesis means for synthesizing speech using the time-series vector selected by the selection means;
It is characterized by having.

According to the speech decoding method according to the first aspect of the invention and the speech decoding apparatus according to the second aspect of the invention, in addition to the time-series vector output from the drive codebook, the time-series vector is changed to change the time-series vector. Since the vector is to generate another time-series vector with a different number of samples with zero amplitude value, there is no need to provide a plurality of driving codebooks, and the amount of memory required to store the driving codebook is reduced. There is.
In addition, based on the gain decoded from the input speech code in order to multiply the time series vector output from the adaptive codebook, it is decided to switch the time series vector having a different number of samples whose amplitude value is zero. Since it is not necessary to separately send information used for the transmission, it is possible to suppress an increase in the number of transmission bits on the code side.

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

Embodiment 1 FIG.
FIG. 1 shows the overall configuration of Embodiment 1 of a speech encoding method and speech decoding method according to the present invention. In the figure, 1 is an encoding unit, 2 is a decoding unit, 3 is a multiplexing unit, and 4 is a separation unit. The encoding unit 1 includes a linear prediction parameter analysis unit 5, a linear prediction parameter encoding unit 6, a synthesis filter 7, an adaptive codebook 8, a gain encoding unit 10, a distance calculation unit 11, a first drive codebook 19, 2 driving code book 20, noise level evaluation unit 24, driving code book switching unit 25, and weighting addition unit 38. The decoding unit 2 includes a linear prediction parameter decoding unit 12, a synthesis filter 13, an adaptive codebook 14, a first driving codebook 22, a second driving codebook 23, a noise level evaluation unit 26, and a driving codebook switching. Section 27, gain decoding section 16, and weighted addition section 39. In FIG. 1, 5 is a linear prediction parameter analysis unit as a spectrum information analysis unit that analyzes the input speech S1 and extracts a linear prediction parameter that is spectral information of speech, and 6 is that encoding the linear prediction parameter that is spectral information. A linear prediction parameter encoding unit as a spectrum information encoding unit that sets the encoded linear prediction parameter as a coefficient of the synthesis filter 7, and 19 and 22 are a first drive in which a plurality of non-noise time-series vectors are stored. Codebooks 20 and 23 are second drive codebooks in which a plurality of noisy time series vectors are stored, 24 and 26 are noise level evaluation units for evaluating the noise level, and 25 and 27 are driven by the noise level. It is a drive codebook switch part which switches a codebook.

  The operation will be described below. First, in the encoding unit 1, the linear prediction parameter analysis unit 5 analyzes the input speech S1 and extracts linear prediction parameters that are speech spectrum information. The linear prediction parameter encoding unit 6 encodes the linear prediction parameter, sets the encoded linear prediction parameter as a coefficient of the synthesis filter 7, and outputs it to the noise level evaluation unit 24. Next, encoding of sound source information will be described. The adaptive codebook 8 stores past drive excitation signals, and outputs a time-series vector in which past drive excitation signals are periodically repeated corresponding to the adaptive code input from the distance calculation unit 11. The noise level evaluation unit 24 uses the encoded linear prediction parameter and the adaptive code input from the linear prediction parameter encoding unit 6 to detect the spectrum inclination, short-term prediction gain, and pitch variation as shown in FIG. The degree of noise in the coding section is evaluated, and the evaluation result is output to the drive codebook switching unit 25. The driving codebook switching unit 25 performs encoding according to the evaluation result of the noise level, for example, using the first driving codebook 19 if the noise level is low and using the second driving codebook 20 if the noise level is high. Switch the driving codebook to be used.

  The first drive codebook 19 stores a plurality of non-noisy time series vectors, for example, a plurality of time series vectors constructed by learning so as to reduce the distortion between the learning speech and the encoded speech. Has been. The second drive codebook 20 stores a plurality of noisy time series vectors, for example, a plurality of time series vectors generated from random noise, and each drive code input from the distance calculation unit 11 is stored in each drive code. Output the corresponding time series vector. Each time series vector from the adaptive codebook 8, the first driving excitation codebook 19 or the second driving codebook 20 is weighted by the weighting addition unit 38 in accordance with the respective gains supplied from the gain encoding unit 10. And the result of the addition is supplied to the synthesis filter 7 as a driving sound source signal to obtain encoded speech. The distance calculation unit 11 obtains the distance between the encoded speech and the input speech S1, and searches for an adaptive code, drive code, and gain that minimize the distance. After the encoding is completed, the code of the linear prediction parameter, the adaptive code that minimizes the distortion between the input speech and the encoded speech, the drive code, and the gain code are output as the encoding result S2. The above is the characteristic operation of the speech coding method according to the first embodiment.

  Next, the decoding unit 2 will be described. In the decoding unit 2, the linear prediction parameter decoding unit 12 decodes the linear prediction parameter from the code of the linear prediction parameter, sets it as a coefficient of the synthesis filter 13, and outputs it to the noise level evaluation unit 26. Next, decoding of sound source information will be described. The adaptive codebook 14 outputs a time series vector in which past drive excitation signals are periodically repeated corresponding to the adaptive code. The noise level evaluation unit 26 evaluates the degree of noise in the same manner as the noise level evaluation unit 24 of the encoding unit 1 from the decoded linear prediction parameter and the adaptive code input from the linear prediction parameter decoding unit 12. Then, the evaluation result is output to the drive codebook switching unit 27. The drive codebook switching unit 27 switches between the first drive codebook 22 and the second drive codebook 23 in the same manner as the drive codebook switching unit 25 of the encoding unit 1 according to the evaluation result of the noise level.

  The first drive codebook 22 includes a plurality of non-noisy time series vectors, for example, a plurality of time series vectors configured by learning so as to reduce the distortion between the learning speech and the encoded speech. The second driving codebook 23 stores a plurality of noisy time series vectors, for example, a plurality of time series vectors generated from random noise, and outputs a time series vector corresponding to each driving code. The time series vectors from the adaptive codebook 14 and the first drive codebook 22 or the second drive codebook 23 are weighted and added 39 according to the respective gains decoded from the gain code by the gain decoder 16. And the result of the addition is supplied as a driving sound source signal to the synthesis filter 13 to obtain an output sound S3. The above is the characteristic operation of the speech decoding method according to the first embodiment.

  According to the first embodiment, the degree of noise of the input speech is evaluated from the code and the encoding result, and a high quality speech is reproduced with a small amount of information by using a different driving codebook according to the evaluation result. can do.

  In the above embodiment, the case where a plurality of time series vectors are stored in the drive codebooks 19, 20, 22, and 23 has been described. However, if at least one time series vector is stored, It can be implemented.

Embodiment 2. FIG.
In the first embodiment described above, two drive codebooks are switched and used, but instead of this, three or more drive codebooks may be provided and switched according to the degree of noise. According to the second embodiment, it is possible to use a driving codebook suitable not only for noise / non-noise but also for intermediate sounds such as slightly noisy. High quality sound can be played.

Embodiment 3 FIG.
FIG. 3 in which the same reference numerals are assigned to the parts corresponding to those in FIG. 1 shows the overall configuration of Embodiment 3 of the speech encoding method and speech decoding method of the present invention. The drive codebooks 29 and 31 storing the vectors are sample thinning sections that make the amplitude value of the low-amplitude samples of the time series vector zero.

  The operation will be described below. First, in the encoding unit 1, the linear prediction parameter analysis unit 5 analyzes the input speech S1 and extracts linear prediction parameters that are speech spectrum information. The linear prediction parameter encoding unit 6 encodes the linear prediction parameter, sets the encoded linear prediction parameter as a coefficient of the synthesis filter 7, and outputs it to the noise level evaluation unit 24. Next, encoding of sound source information will be described. The adaptive codebook 8 stores past drive excitation signals, and outputs a time-series vector in which past drive excitation signals are periodically repeated corresponding to the adaptive code input from the distance calculation unit 11. The noise level evaluation unit 24 uses the encoded linear prediction parameter and the adaptive code input from the linear prediction parameter encoding unit 6 to calculate, for example, the noise of the coding section from the spectrum slope, short-term prediction gain, and pitch variation. The degree is evaluated, and the evaluation result is output to the sample thinning unit 29.

  The drive codebook 28 stores a plurality of time series vectors generated from, for example, random noise, and outputs a time series vector corresponding to the drive code input from the distance calculation unit 11. In accordance with the evaluation result of the noise level, the sample thinning unit 29 calculates, for example, the amplitude value of the sample that does not satisfy a predetermined amplitude value with respect to the time series vector input from the drive codebook 28 if the noise level is low. A time series vector set to zero is output, and if the noise level is high, the time series vector input from the drive codebook 28 is output as it is. Each time series vector from the adaptive codebook 8 and the sample thinning unit 29 is weighted and added by the weighting addition unit 38 in accordance with each gain given from the gain encoding unit 10, and the addition result is used as a driving excitation signal. The encoded speech is supplied to the synthesis filter 7. The distance calculation unit 11 obtains the distance between the encoded speech and the input speech S1, and searches for an adaptive code, drive code, and gain that minimize the distance. After the encoding is completed, the code of the linear prediction parameter, the adaptive code that minimizes the distortion between the input speech and the encoded speech, the drive code, and the gain code are output as the encoding result S2. The above is the characteristic operation of the speech coding method according to the third embodiment.

  Next, the decoding unit 2 will be described. In the decoding unit 2, the linear prediction parameter decoding unit 12 decodes the linear prediction parameter from the code of the linear prediction parameter, sets it as a coefficient of the synthesis filter 13, and outputs it to the noise level evaluation unit 26. Next, decoding of sound source information will be described. The adaptive codebook 14 outputs a time series vector in which past drive excitation signals are periodically repeated corresponding to the adaptive code. The noise level evaluation unit 26 evaluates the degree of noise in the same manner as the noise level evaluation unit 24 of the encoding unit 1 from the decoded linear prediction parameter and the adaptive code input from the linear prediction parameter decoding unit 12. Then, the evaluation result is output to the sample thinning unit 31.

  The drive codebook 30 outputs a time series vector corresponding to the drive code. The sample decimation unit 31 outputs a time series vector by the same processing as the sample decimation unit 29 of the encoding unit 1 according to the noise level evaluation result. The time series vectors from the adaptive codebook 14 and the sample thinning unit 31 are weighted and added by the weighting addition unit 39 according to the respective gains supplied from the gain decoding unit 16, and the addition result is used as a driving excitation signal. The output sound S3 is obtained by being supplied to the synthesis filter 13.

  According to the third embodiment, a drive codebook storing noisy time series vectors is provided, and noise characteristics are obtained by thinning out signal samples of drive sound sources in accordance with the evaluation result of the degree of speech noise. By generating a driving sound source with a low degree of sound, it is possible to reproduce high-quality sound with a small amount of information. In addition, since it is not necessary to provide a plurality of driving codebooks, there is an effect of reducing the amount of memory required for storing the driving codebook.

Embodiment 4 FIG.
In the above-described third embodiment, the time series vector samples are thinned out / not thinned out, but instead of this, the amplitude threshold at the time of thinning out the samples may be changed according to the degree of noise. According to the fourth embodiment, it is possible to generate and use a time series vector suitable not only for noise / non-noise but also for intermediate sounds such as slightly noisy. Therefore, it is possible to reproduce high quality sound.

Embodiment 5. FIG.
FIG. 4 in which the same reference numerals are assigned to corresponding parts as in FIG. 1 shows the overall configuration of the speech encoding method and speech decoding method according to Embodiment 5 of the present invention, and 32 and 35 in FIG. First drive codebooks 33 and 36 storing vectors, second drive codebooks 33 and 37 storing non-noise time-series vectors, and weight determination units.

  The operation will be described below. First, in the encoding unit 1, the linear prediction parameter analysis unit 5 analyzes the input speech S1 and extracts linear prediction parameters that are speech spectrum information. The linear prediction parameter encoding unit 6 encodes the linear prediction parameter, sets the encoded linear prediction parameter as a coefficient of the synthesis filter 7, and outputs it to the noise level evaluation unit 24. Next, encoding of sound source information will be described. The adaptive codebook 8 stores past drive excitation signals, and outputs a time-series vector in which past drive excitation signals are periodically repeated corresponding to the adaptive code input from the distance calculation unit 11. The noise level evaluation unit 24 uses the encoded linear prediction parameter and the adaptive code input from the linear prediction parameter encoding unit 6 to calculate, for example, the noise of the coding section from the spectrum slope, short-term prediction gain, and pitch variation. The degree is evaluated, and the evaluation result is output to the weight determination unit 34.

  The first drive codebook 32 stores, for example, a plurality of noisy time series vectors generated from random noise, and outputs time series vectors corresponding to the drive codes. The second drive codebook 33 stores, for example, a plurality of time series vectors configured by learning so as to reduce the distortion between the learning speech and the encoded speech, and is input from the distance calculation unit 11. A time series vector corresponding to the drive code is output. In accordance with the noise level evaluation result input from the noise level evaluation unit 24, the weight determination unit 34 calculates the time series vector from the first drive codebook 32 and the second drive codebook 33, for example, according to FIG. The weight to be given to the time series vector of is determined. The time series vectors from the first drive codebook 32 and the second drive codebook 33 are weighted according to the weight given from the weight determining unit 34 and added. The time series vector output from the adaptive codebook 8 and the time series vector generated by the weighted addition are weighted and added by the weighting addition unit 38 according to each gain given from the gain encoding unit 10, The addition result is supplied as a driving sound source signal to the synthesis filter 7 to obtain encoded speech. The distance calculation unit 11 obtains the distance between the encoded speech and the input speech S1, and searches for an adaptive code, drive code, and gain that minimize the distance. After this encoding is completed, the code of the linear prediction parameter, the adaptive code that minimizes the distortion between the input speech and the encoded speech, the drive code, and the gain code are output as the encoding result.

  Next, the decoding unit 2 will be described. In the decoding unit 2, the linear prediction parameter decoding unit 12 decodes the linear prediction parameter from the code of the linear prediction parameter, sets it as a coefficient of the synthesis filter 13, and outputs it to the noise level evaluation unit 26. Next, decoding of sound source information will be described. The adaptive codebook 14 outputs a time series vector in which past drive excitation signals are periodically repeated corresponding to the adaptive code. The noise level evaluation unit 26 evaluates the degree of noise in the same manner as the noise level evaluation unit 24 of the encoding unit 1 from the decoded linear prediction parameter and the adaptive code input from the linear prediction parameter decoding unit 12. Then, the evaluation result is output to the weight determination unit 37.

  The first drive codebook 35 and the second drive codebook 36 output a time series vector corresponding to the drive code. It is assumed that the weight determination unit 37 gives weights in the same manner as the weight determination unit 34 of the encoding unit 1 according to the noise level evaluation result input from the noise level evaluation unit 26. The time series vectors from the first drive codebook 35 and the second drive codebook 36 are weighted according to the respective weights given from the weight determination unit 37 and added. The time series vector output from the adaptive codebook 14 and the time series vector generated by the weighted addition are weighted by the weighting addition section 39 according to the respective gains decoded from the gain code by the gain decoding section 16. The weighted addition is performed, and the addition result is supplied as a driving sound source signal to the synthesis filter 13 to obtain the output sound S3.

  According to the fifth embodiment, the degree of speech noise is evaluated from the code and the encoding result, and a noisy time series vector and a non-noisy time series vector are weighted and added according to the evaluation result. As a result, high-quality sound can be reproduced with a small amount of information.

Embodiment 6 FIG.
In the first to fifth embodiments described above, the gain codebook may be changed according to the evaluation result of the degree of noise. According to the sixth embodiment, since a code book having an optimum gain can be used according to the driving code book, high-quality sound can be reproduced.

Embodiment 7 FIG.
In the first to sixth embodiments described above, the degree of noise in the speech is evaluated, and the drive codebook is switched according to the evaluation result, but the voiced rise or bursting consonant is determined and evaluated, The drive codebook may be switched according to the evaluation result. According to the seventh embodiment, not only the noise state of speech but also the voiced rising and bursting consonants can be further classified and the driving codebook suitable for each can be used. Can be played.

Embodiment 8 FIG.
In the first to sixth embodiments described above, the degree of noise in the coding section is evaluated from the spectral tilt, the short-term prediction gain, and the pitch variation shown in FIG. 2, but the magnitude of the gain value for the adaptive codebook output is used. May be evaluated.

It is a block diagram which shows the whole structure of Embodiment 1 of the audio | voice encoding and audio | voice decoding apparatus by this invention. It is a table | surface with which it uses for description of the noise degree evaluation in Embodiment 1 of FIG. It is a block diagram which shows the whole structure of Embodiment 3 of the audio | voice encoding and audio | voice decoding apparatus by this invention. It is a block diagram which shows the whole structure of Embodiment 5 of the audio | voice encoding and audio | voice decoding apparatus by this invention. It is a basic diagram with which it uses for description of the weight determination process in Embodiment 5 of FIG. It is a block diagram which shows the whole structure of the conventional CELP audio | voice encoding decoding apparatus. It is a block diagram which shows the whole structure of the conventional improved CELP audio | voice encoding / decoding apparatus.

Claims (2)

In a code-excited linear prediction (CELP) speech decoding method that synthesizes speech based on an input speech code using a driven codebook and an adaptive codebook,
A gain decoding step of obtaining the gain value multiplied with time-series vector adaptive codebook and decodes the input speech code is output,
Whether the gain value is large or small is evaluated, and based on the evaluation result, it is determined whether or not it is necessary to change the first time series vector output from the driving codebook, and the first time series is determined. When it is determined that the vector needs to be changed, the first time-series vector is changed to generate a second time-series vector in which the number of samples having zero amplitude value is different from the first time-series vector. A time-series vector generation step,
The first or second time series vector when the gain value is evaluated to be greater than the evaluation result.
When the speech is synthesized using the time-series vector having the larger number of samples having the amplitude value of zero and the gain value is evaluated to be small in the evaluation result, the first time or the second time A speech synthesis step of synthesizing speech using a time-series vector having a smaller number of samples having an amplitude value of zero among sequence vectors ;
A speech decoding method characterized by comprising: In: (CELP Code-Excited Linear Prediction ) speech decoding apparatus using the driving codebook and the adaptive codebook, the code excited linear prediction synthesizing speech based on the input speech code
A gain decoding means for obtaining a gain value multiplied with time-series vector adaptive codebook and decodes the input speech code is output,
Whether the gain value is large or small is evaluated, and based on the evaluation result, it is determined whether or not the first time series vector output from the drive codebook needs to be changed, and the first time series is determined. When it is determined that the vector needs to be changed, the first time-series vector is changed to generate a second time-series vector in which the number of samples having an amplitude value of zero is different from the first time-series vector. Time-series vector generating means for performing,
The first or second time series vector when the gain value is evaluated to be greater than the evaluation result.
When the speech is synthesized using the time-series vector having the larger number of samples having the amplitude value of zero and the gain value is evaluated to be small in the evaluation result, the first time or the second time Speech synthesis means for synthesizing speech using a time-series vector having a smaller number of samples with zero amplitude value among sequence vectors ;
A speech decoding apparatus comprising:

Images