JP3353852B2 - Audio encoding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech encoding method, and more particularly to a speech encoding method, in which a filter representing a spectral envelope characteristic of speech including a musical tone is driven by a sound source vector to synthesize speech. The present invention relates to a high-efficiency speech coding method for digitally coding a signal sequence with a small amount of information.

[0002]

2. Description of the Related Art In a digital mobile communication system, a high-efficiency voice encoding method is used to efficiently use a storage medium in a service of efficiently using radio waves or storing voice or music. Is used. At present, as a method for efficiently encoding speech, an original speech is divided into sections called frames, which are arranged at regular intervals of about 5 to 50 ms, and the speech of one frame is divided into a characteristic of a linear filter that represents an envelope characteristic of a frequency spectrum. A method has been proposed in which the information is separated into two pieces of information of a driving sound source signal for driving the filter, and each piece of information is encoded. In this method, as a method of encoding a drive excitation signal, a method of separating and encoding a periodic component considered to correspond to a pitch period (fundamental frequency) of a voice and other components is known. Code-Excited Linear Prediction (CEL) is an example of an encoding method of the driving excitation information.
P). This code-driven linear prediction coding method (CELP
) Are described in MRSchroeder and BSAtal, “Cod
e-Excited Linear Prediction (CELP): High Quality Spe
ech at Very Low Bit Rates ", IEEE Proc. ICASSP-85, pp.
937-940,1985.

[0003] The above encoding method will be described with reference to FIG. For the original speech input to the input terminal 1, the linear prediction analysis unit 2 calculates a linear prediction parameter representing the frequency spectrum envelope characteristic of the original speech. The obtained linear prediction parameters are encoded by the linear prediction parameter encoding unit 3 and sent to the linear prediction parameter decoding unit 4. The linear prediction parameter decoding unit 4 reproduces a filter coefficient from the received code and sends it to the synthesis filter 5. The details of the linear prediction analysis and the encoding of the linear prediction parameters are described in, for example, "Digital Speech Processing" by Sadahiro Furui (Tokai University Press). Here, the linear prediction analysis, the linear prediction parameter encoding unit, the linear prediction parameter decoding unit, and the synthesis filter can be replaced with non-linear ones.

[0004] From the adaptive codebook 7, the immediately preceding past drive excitation vector stored in the buffer and corresponding to one to several frames of the immediately preceding drive excitation vector having a length corresponding to a certain period. By clipping and repeating the clipped vector until the length of the frame is reached, a time-series vector candidate corresponding to the periodic component of the voice is output.

[0005] From the noise codebook 8, candidates for a time-series code vector having a length of one frame corresponding to a non-periodic component of speech are output. As these candidates, a predetermined number of candidate vectors according to the number of bits for encoding are stored independently of the input speech. The time series vector candidates output from the adaptive codebook 7 are multiplied by the weight created by the weight creating unit stored in the weighting codebook 11 by the multiplier 9. On the other hand, the time series vector candidates output from the noise codebook 8 are multiplied by the weight created by the weight creation unit stored in the weight codebook 11 by the multiplier 10. The result of the multiplication in the multiplication unit 9 and the result of the multiplication in the multiplication unit 10 are added in the addition unit 12 to be candidates for the drive excitation vector.

[0006] The synthesis filter 5 is a linear filter using the output of the linear prediction parameter decoding unit 4 as a filter coefficient, and outputs a candidate of a driving sound source vector which is an output of the addition unit 12 and outputs a candidate of a reproduced voice. In general, the order of the synthesis filter 5, that is, the order of the linear prediction analysis, is generally about 10 to 16 order. As described above, the synthesis filter 5 can be a non-linear filter.

[0007] In the distortion calculator 6, the synthesis filter 5
Calculate the distortion between the input sound and the reproduced sound candidate which is the output of. The calculation of the distortion is often performed in consideration of, for example, a coefficient of a synthesis filter such as auditory weighting or an unquantized linear prediction coefficient. Codebook search control unit 13
In, a cyclic code, a noise code, and a weight code are selected such that distortion between the reproduced sound candidate supplied from the synthesis filter 5 and the input sound input to the input terminal 1 is minimized, and the driving in the frame is selected. Determine the sound source vector.

The cyclic code, the noise code, the weight code, and the linear prediction parameter code output from the linear prediction parameter encoding unit 3 determined by the codebook search control unit 13 are as follows:
It is sent to the code sending unit 14 and stored according to the form of use, or sent to the receiving side. The above method is a typical code-driven linear predictive coding method. As a method of performing higher quality coding using a smaller amount of operation and less memory, pitch-driven excitation source code-driven linear predictive coding is used. (PSI-CELP: Pitch Synchronous Innovation-
CELP) method has been proposed. For details of this encoding method, see, for example, “Literature: Mano, Moriya, Miki, Omuro,
“Study of Half Rate Speech Codec for Car Phone”,
IEICE Technical Report, SP-92-133 (1933) "," Oya, Suda, Miki,
“Pitch SynchronousInnovation CELP (PSI-CELP) -PDC
Half-rate speech CODEC- ", IEICE Technical Report, RCS93-78 (1993)"
It is described in.

A description will now be given, with reference to FIG. 2, of a pitch synchronous excitation source code driving linear prediction encoding method. For the original speech input to the input terminal 1, the linear prediction analysis unit 2 calculates a linear prediction parameter representing the frequency spectrum envelope characteristic of the original speech. The obtained linear prediction parameters are encoded by the linear prediction parameter encoding unit 3 and sent to the linear prediction parameter decoding unit 4. The linear prediction parameter decoding unit 4 reproduces a filter coefficient from the received code, and sends it to the synthesis filter 5. At this time, it is efficient to encode the linear prediction parameter by using the correlation between the reproduced value of the prediction parameter in the past frame or the transmission code. For example, the reproduction value (or vector) of the prediction parameter in a plurality of past frames
Auto-regressive predictive coding that transmits a difference (value or vector) between the linear combination of the prediction parameter and a difference (linear difference) between the transmission code (value or vector) of the prediction parameter in a plurality of past frames. (Value or vector) is preferably used with the moving average type predictive coding.

From the adaptive codebook 71, one or several frames of the immediately preceding drive excitation vector, which is the immediately preceding drive excitation vector stored in the buffer, having a length corresponding to a certain period. By clipping and repeating the clipped vector until the length of the frame is reached, a time-series vector candidate corresponding to the periodic component of the voice is output. From the fixed codebook 72, a time-series code vector candidate having a length of one frame corresponding to the non-periodic component of speech is output. The adaptive codebook 71 and the fixed codebook 72 are used by being adaptively switched so that the distortion of the input speech and the synthesized speech in the current frame is minimized.

In the above method, a plurality of random codebooks are prepared. Each of the noise codebooks 81 and 82 outputs a time-series code vector candidate having a length of one frame corresponding to the non-periodic component of the voice. It is cycled by a cycle corresponding to a pitch of 71. At this time, if the fixed codebook 72 is selected instead of the adaptive codebook 71, the periodic processing is not performed. Also, instead of periodicizing all of the candidates output from the random codebooks 81 and 82, only some of the candidates may be periodicized.
Each of the vector candidates subjected to the periodization processing is added in the adding unit 23 to become a noise code vector candidate.

The time series vector candidates output from the adaptive codebook 71 or the fixed codebook 72 are multiplied by the weight created by the weight creating unit stored in the weighting codebook 11 by the multiplier 9. On the other hand, the noise vector candidates that have been subjected to the periodic processing are added by the adding unit 23 and then multiplied by the weight created by the weight creating unit stored in the weight codebook 11 by the multiplier 10. The result of the multiplication in the multiplication unit 9 and the result of the multiplication in the multiplication unit 10 are added in the addition unit 12 to be candidates for the drive excitation vector. At this time, it is efficient to use a vector quantization technique for weight coding.

The synthesis filter 5 is a linear filter that uses the output of the linear prediction parameter decoding unit 4 as a filter coefficient, and outputs a candidate of a reproduced sound source with a driving excitation vector candidate output from the addition unit 12 as an input. In general, the order of the synthesis filter 5, that is, the order of the linear prediction analysis, is generally 10 to 16 order. As described above, the synthesis filter can be a non-linear filter.

In the distortion calculator 6, the synthesis filter 5
Then, the distortion between the reproduced voice candidate output as the input voice and the input voice input to the input terminal 1 is calculated. The calculation of this distortion is
Often, this is done taking into account the coefficients of the synthesis filter, such as auditory weighting, or unquantized linear prediction coefficients. The codebook search control unit 13 selects a periodic code or fixed code that minimizes the distortion between each reproduced speech candidate and the input speech, a noise code and a weight code of each noise codebook, and selects the frame. Is determined.

The periodic code or fixed code, noise code, weight code, and linear prediction parameter code output from the linear prediction parameter coding unit 4 determined by the codebook search control unit 13 are transmitted to a code transmitting unit 14. And stored according to the form of use or transmitted to the receiving side.

[0016]

Speech is a non-stationary signal whose characteristics change every moment, and the amount of information required for encoding differs depending on each moment. If the amount of information required varies from moment to moment, encoding with one model and a constant bit rate, as in the conventional predictive encoding method described above, can reduce the signal sequence to less information. It is not preferable in terms of efficiency in digitally encoding the amount. If the total amount of information is limited, the quality is degraded.

According to the present invention, by switching both the encoding method and the bit rate in accordance with the characteristics of speech which change from moment to moment, the speech of which speech is digitally encoded with a small amount of information while maintaining high quality is maintained. The present invention provides an encoding method.

[0018]

Means for Solving the Problems Claim 1: A time series vector which is generated by repeatedly retrieving past excitation vectors extracted from the adaptive codebook for each frame or for each subframe at a cycle corresponding to the pitch cycle. A time-series vector extracted from a noise codebook, or both of these time-series vectors, a filter is driven to reproduce an audio including a musical tone, thereby encoding an input audio. Prior to encoding, the features of the frame section or subframe section are extracted,
The encoding method and bit rate corresponding to the extracted features are selectively encoded, and the past driving excitation vector extracted from the adaptive codebook is pitch-performed for the speech section determined to be voiced when extracting the features of the input speech. A time series vector repeatedly created with a cycle corresponding to
A time series vector extracted from the noise codebook is used as a time series vector periodicized by a period corresponding to the pitch period of the adaptive codebook. The pitch period of the adaptive codebook for a frame or subframe determined to be a target is determined to be a constant multiple of the past pitch period or a fraction of a constant.
The variation width is expressed by adding and subtracting a minute change to twice, and the quantization width of the minute change to be added or subtracted to a constant multiple of the past pitch period or one-fold of the constant is narrow when the change is small and large when the change is small. At times, a speech encoding method that is widely expressed is constructed.

In addition, the present invention provides a time series vector and a noise codebook which are created by repeating past driving excitation vectors extracted from the adaptive codebook for each frame or for each subframe in a cycle corresponding to a pitch cycle. In the speech encoding method for encoding the input speech by driving the filter by using the time series vector extracted from the or both time series vectors and reproducing the speech including the musical tone, the input speech is encoded. Prior to this, the feature of the frame section or sub-frame section is extracted, the encoding method and bit rate corresponding to the extracted feature are selectively encoded, and the speech determined to be voiced when the feature of the input speech is extracted. For the interval, the past driving excitation vector extracted from the adaptive codebook is repeated at a cycle corresponding to the pitch cycle. Uses an encoding method that reproduces audio by driving a filter using the time series vector generated from the noise code book and a time series vector obtained by cycling the time series vector extracted from the noise codebook with a cycle corresponding to the pitch cycle of the adaptive codebook. Then, for a frame or subframe in which the speech section is determined to be stationary, the pitch cycle of the adaptive codebook is expressed by adding or subtracting a minute change to a constant multiple or a constant fraction of the past pitch cycle. Then, a speech encoding method is described in which a quantization width of a minute change to be added or subtracted to a constant multiple of the past pitch period or a fraction of the constant is taken to be wider than the quantization width at the time of one. .

In a third aspect of the present invention, in the speech encoding method according to the first aspect, a quantization width of a minute change to be added or subtracted to a constant multiple of the past pitch period or a constant multiple thereof is set as:
A speech encoding method is described which is expressed by being wider than the quantization width of 1 time.

Further, claim 4: claims 1 to 3
In the speech coding method described in the above, for a frame or a subframe in which the speech section is determined to be stationary, a spectrum parameter, a gain code vector, a noise code vector, or bits allocated to all of them. A speech encoding method is provided in which the number is assigned less than the number of bits assigned in a frame or subframe that is determined to be voiced and not stationary.

Claim 5: Claims 1 to 4
In the speech encoding method described in the above, between adjacent frames or sub-frames, extracted or quantized prior to encoding, the change in pitch and power and the spectrum is more than a predetermined threshold Therefore, a speech encoding method for judging that the speech section is stationary because of its small size is constructed.

[0023]

An embodiment of the present invention will be described with reference to the drawings.
FIG. 3 is a diagram for explaining a method of analyzing the characteristics of the input speech, classifying it into several patterns (modes), and switching the encoding method according to these patterns. FIG. 3 shows an example of the four modes. The original speech input to the input terminal 1 is first sent to the speech feature analysis unit 20, where various parameters indicating the features of the speech are calculated. Representative examples of this parameter include voice power, the maximum value of the modified correlation function, the delay time (open loop pitch) at which the modified correlation function is maximized, and the amount of spectrum change. These calculated parameters are sent to the mode determination unit 40. In the mode determination unit 40, the characteristic parameters obtained by these open loops and the quantization parameters in the past frame stored in the buffer unit 50, such as the pitch period of the adaptive codebook of the previous frame, the power of the synthesized voice, the previous frame Is determined, to which of the predetermined modes the voice of the current frame belongs is determined, and based on the determination result, the changeover switches 41 and 42 are switched and the encoding units 31 to 3
Select one of 4

FIG. 4 is a flowchart for explaining a method of classifying voice sections in the case of four modes. First, it is determined from each of the input parameters whether each frame is a voice section or a non-voice section. This is determined by paying attention to the normal audio power. A certain threshold is determined, and when the power is larger than this, it is determined that the voice section is used, and when the power is less than this, it is determined that the voice section is not used. The distinction between speech and non-speech determined by this method need not be strict in a phonetic sense. It is sufficient that the classification accuracy is within a range in which the quality does not deteriorate when encoded in the non-voice mode. Normally, the threshold value of the power is such that the amplitude of the input voice is quantized by 16 bits and the average power per sample is a value of several hundreds to about 10,000.

Next, it is determined whether the frame determined to be a voice section is a voiced sound or an unvoiced sound. Usually, this determination is made using the maximum value of the audio power and the modified correlation function, and the audio power is a certain value, for example, the average power per sample is tens of thousands to 100,000 or less and the maximum value of the modified correlation function Is less than a certain threshold, for example, 0.1 to 0.3, it is determined that the sound is unvoiced. Otherwise, it is determined that the sound is voiced. In this case as well, the judgment need not be strictly phonetically, and it is sufficient that the judgment does not lead to deterioration in quality.

Further, it is determined whether the voice section determined to be voiced is a transient section or a stationary section.
In this determination, the amount of change in the feature amount between adjacent frames is checked, and if the change is small, it is determined that the frame is a stationary portion, and if the change is large, it is determined that the frame is a transient portion (unsteady). For example, the change amount of the open loop pitch is checked. A change in the pitch cycle of the adaptive codebook in the past frame and a change in the open loop pitch of the current frame are checked to determine whether or not the change exceeds a certain threshold. In general, in a coding method of the CELP system, the pitch period of the adaptive codebook often takes a value of an integral multiple or a fraction of an integer, such as a double pitch or a half pitch. If the amount of change from the multiplied value is small, it may be determined to be steady. As an example of the threshold value, it is preferable that the change is within about 10% to 20%. PSI-CELP
In the case where the adaptive codebook and the fixed codebook are switched and used as in the case of the type coding method, and when the fixed codebook is selected in the previous frame, even if it is determined that the amount of change in the pitch is large, Good. In addition to the pitch change amount, the power change amount or the ratio of the power of the synthesized voice of the previous frame to the current input voice power is a threshold value, for example, 1.2 to 2.0 times or less. It may be determined that the spectrum change amount between the frame and the frame is a threshold value, for example, 3 dB to 10 dB or less, and it is determined that the frame is stationary.

In the above, an example in which the voice section is classified into four modes has been described. However, the number of modes can be set to 2 or 8, or any other number. The non-speech and the unvoiced sound are treated as the same classification, and the classification of the transient part and the stationary part is more detailed. For example, the classification is such that the pitch is stationary but the power is transient. Further, for voiced sounds, the stationary part and the transient part can be treated as the same classification. After the speech sections are classified in this way, the encoding section is switched to perform encoding.

Next, an example of a coding model applied to each voice section will be described. An encoding method of non-voice or unvoiced sound will be described with reference to FIG. In these sections of unvoiced or unvoiced sounds corresponding to mode 1 and mode 2 in FIG. 4, there is no periodic component corresponding to the pitch of the voice, so the adaptive codebook is not used, and the noise codebook 8 and the weight code are not used. A driving excitation is generated only by the stored weight codebook 11. In these speech sections, the sensitivity of the auditory distortion to the waveform distortion is extremely low, so that the number of bits smaller than that in the voiced section is reduced by the noise codebook 8 and the weighting codebook 11, the linear prediction parameter. Can be assigned to

For voiced transients corresponding to mode 3, the most bits are required to encode the speech. Therefore, in this voice section, it is preferable to use a high-efficiency encoding method such as the PSI-CELP encoding method shown in FIG. 2 described above.
On the other hand, for the stationary part of the voiced sound corresponding to mode 4,
Although it is perceptually important, the temporal change of the audio signal is small. Therefore, in the case of a predictive coding method such as a CELP-based coding method, good quality can be maintained even if the amount of information is relatively small. it can. FIG. 6 shows an encoding method for this section.
First, since the change of the pitch frequency is gradual, the periodic code of the adaptive codebook 70 can be assigned a smaller number of bits than the transient part, and can be subjected to differential coding. The differential periodic code is converted into an absolute pitch period in a differential periodic code decoding unit 80 and sent to the adaptive codebook 70. Adaptive codebook 7
At 0, adaptive code vector candidates are created based on this pitch period. FIG. 7 shows the internal configuration of the differential periodic code decoding unit 80. In FIG. 7, the difference cycle code is converted by a difference cycle code conversion unit 83 into a multiplication factor and a minute variation multiplied by the preceding frame adaptive code pitch cycle. Magnification is 1
As mentioned above, the adaptive code pitch of the CELP-based coding method can easily become an integer multiple as described above, so that an integer multiple other than 1 or a multiple of an integer is prepared. Degradation can be reduced. What kind of magnification and minute variation should be prepared depends on the number of allocated bits. For example, when 5 bits and 32 codes can be used, 5 codes are prepared for 2 times and 1/2 times respectively. I do. At this time, the minute change may be quantized at regular intervals from the value obtained by multiplying the pitch period of the previous frame by the magnification, but even a rarely large change can be efficiently performed without increasing the number of bits. To cope with this, the period close to the value obtained by multiplying the pitch period of the previous frame by the magnification is finer in the quantization width. You should take it widely. Similarly, when there are few codes for minute fluctuations assigned to integral multiples, it is efficient to increase the quantization width.

With reference to FIG. 8, another example of the coding method of the voiced stationary part will be described. In the voiced sound stationary section, not only the pitch but also the change of the entire voice is small, so that the bit allocation to the noise codebook 81, the weighted codebook 11, and the linear prediction parameter is also reduced in addition to the differential coding of the adaptive code period. be able to. In particular, in the case of the PSI-CELP type encoding method, the number of bits can be reduced by using only the noise codebook 81 which is one of the noise codebooks having a plurality of channels. As a result of the experiment, even if the difference coding of the adaptive code pitch and the reduction of the number of bits allocated to the linear prediction parameters are performed, almost no deterioration in quality is observed perceptually, and the number of bits of the weight codebook is reduced. Also, the deterioration was slight.

Table 1 shows an example of the number of bits allocated to each codebook in each mode when the above method is based on PSI-CELP. Note that the subframe length is 10 milliseconds, and the fraction notation column indicates that subframes of the denominator are collectively encoded with numerator bits.
Regarding the encoding of the power of mode 1, only when all four sub-frames are in mode 1, 3
It shall be encoded by bits. 7/4 otherwise
Bit.

[0032]

[Table 1] The above method can be applied to all predictive coding methods other than the CELP and PSI-CELP coding methods.

[0033]

As described above, according to the present invention, the characteristics of the input speech are analyzed and classified into several patterns (modes). By encoding with the minimum amount of information required to maintain quality, the average bit rate or the total amount of memory for storage can be reduced compared to conventional encoding methods while maintaining high quality. Can be.

As a result of the simulation by the computer and the subjective evaluation experiment by the general public, the average bit rate 2.1
PD with 42 kbit / s and bit rate of 3.45 kbit / s
Compared with the C half-rate standard PSI-CELP system, the equivalent Q value is reduced by about 1 dB, and the average bit rate is 2.
At 218 kbit / s, almost the same quality as the PDC standard was obtained, and the effectiveness of the present invention was confirmed. And
The adaptive code pitch of the CELP coding method can easily be an integral multiple. Generally, since the pitch period does not change much in the stationary portion of speech, basically, the coding quality of the pitch period is not significantly degraded by the difference quantization from the previous frame. However, the “pitch period of the adaptive codebook” used in CELP coding does not always match the basic period as an acoustic phenomenon. For example, a frequency component that is an integral multiple of the fundamental frequency is rarely determined as a pitch component.
Due to this, when simple differential quantization is performed, quality may be deteriorated. Here, for a frame or a subframe in which the voice section is determined to be stationary,
A value obtained by adding or subtracting a minute change from a value obtained by multiplying the pitch period of the adaptive codebook by a constant multiple or a constant multiple of the past pitch period is prepared. By expressing the quantization width of the minute change to be added or subtracted as narrow when the change is small and wide when the change is large, it is possible to reduce the deterioration of quality. That is,
The minute change may be quantized at equal intervals from the change from the value obtained by multiplying the pitch cycle of the previous frame by the magnification.However, a cycle close to the value obtained by multiplying the pitch cycle of the previous frame by the magnification has the quantization width. Finely, for the part where the change from the value obtained by multiplying the pitch cycle of the previous frame by the magnification is large, by increasing the quantization width, even a rarely large change in the pitch cycle can be efficiently performed without increasing the number of bits. Can respond.

[Brief description of the drawings]

FIG. 1 is a view for explaining a CELP encoding method.

FIG. 2 is a view for explaining a PSI-CELP encoding method.

FIG. 3 is a diagram illustrating a mode switching encoding method according to the present invention.

FIG. 4 is a flowchart for explaining how to classify modes in a voice section;

FIG. 5 is a diagram illustrating an encoding method of non-voice and unvoiced sounds.

FIG. 6 is a view for explaining an encoding method of a voiced sound stationary unit.

FIG. 7 is a diagram illustrating a differential periodic code decoding unit.

FIG. 8 is a view for explaining another example of the coding method of the voiced sound stationary unit.

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-5-19795 (JP, A) JP-A-5-19796 (JP, A) JP-A-5-165500 (JP, A) JP-A-5-19795 265496 (JP, A) JP-A-5-289696 (JP, A) JP-A-6-12098 (JP, A) JP-A-7-36495 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 19/00-19/14 H03M 7/30 H04B 14/04

Claims (5)

(57) [Claims] 1. A time-series vector generated by repeating a past excitation vector extracted from an adaptive codebook for each frame or for each subframe at a cycle corresponding to a pitch cycle, and extracted from a noise codebook. In a speech encoding method for encoding an input speech by driving a filter with a sequence vector or both of these time-series vectors to reproduce a speech including a musical tone, prior to encoding the input speech, The feature of the frame section or the sub-frame section is extracted, the encoding method and bit rate corresponding to the extracted feature are selectively encoded, and the speech section determined to be voiced when the feature of the input speech is extracted is adapted. A time-series vector created by repeating past driving excitation vectors extracted from the codebook with a cycle corresponding to the pitch cycle And a time series vector obtained by periodicizing the time series vector extracted from the noise codebook with a period corresponding to the pitch period of the adaptive codebook, by using a coding method of driving a filter to reproduce the sound. For frames or subframes whose section is determined to be stationary, the pitch cycle of the adaptive codebook is expressed by adding or subtracting a minute change to a constant multiple or a constant fraction of the past pitch cycle, The audio is characterized in that the quantization width of a minute change to be added or subtracted to a constant multiple of the pitch period or one-fold of the constant is narrow when the change is small and wide when the change is large. Encoding method. 2. A time-series vector generated by repeating a past excitation vector extracted from an adaptive codebook for each frame or for each subframe at a cycle corresponding to a pitch cycle, and extracted from a noise codebook. In a speech encoding method for encoding an input speech by driving a filter with a sequence vector or both of these time-series vectors to reproduce a speech including a musical tone, prior to encoding the input speech, The features of the frame section or sub-frame section are extracted, the encoding method and bit rate corresponding to the extracted features are selectively encoded, and the speech section determined to be voiced during the feature extraction of the input speech is adapted. A time-series vector created by repeating past driving excitation vectors extracted from the codebook with a cycle corresponding to the pitch cycle And a time series vector obtained by periodicizing the time series vector extracted from the noise codebook with a period corresponding to the pitch period of the adaptive codebook, by using a coding method of driving a filter to reproduce the sound. For frames or subframes whose section is determined to be stationary, the pitch cycle of the adaptive codebook is expressed by adding or subtracting a minute change to a constant multiple or a constant fraction of the past pitch cycle, A speech encoding method characterized in that a quantization width of a minute change to be added or subtracted to a constant multiple of the pitch period or a constant multiple thereof is expressed as being wider than the quantization width when the pitch is one. 3. The speech encoding method according to claim 1, wherein a quantization width of a minute change to be added or subtracted to a constant multiple of the past pitch period or a constant multiple thereof is set to one. A speech encoding method characterized by being represented by being wider than a quantization width. 4. The speech encoding method according to claim 1, wherein a spectrum parameter or a gain code vector or a gain code vector is determined for a frame or a subframe in which the speech section is determined to be stationary. A noise code vector or a code for speech characterized in that the number of bits allocated to all of them is less than the number of bits allocated in a frame or a subframe that is voiced and determined to be non-stationary. Method. 5. A speech encoding method according to claim 1, wherein a pitch and a power between adjacent frames or sub-frames are extracted or quantized prior to encoding. And determining that the voice section is stationary based on a change in the spectrum being smaller than a preset threshold value.