JPH0258100A - Voice encoding and decoding method, voice encoder, and voice decoder - Google Patents

Abstract

PURPOSE:To obtain a synthetic speech which does not deteriorate in quality even if there is a transient part of a speech or a change part between vowels by dividing a frame of a sound source signal in a speech section into pitch cycles, and indicating a multipulse in a pitch section as one representative section and an amplitude correction coefficient and a phase correction coefficient in other sections. CONSTITUTION:A discrete speech signal is inputted to a transmission side and a spectrum parameter indicating a spectrum envelope and a pitch parameter indicating pitch are extracted by frames; and a frame section is divided into pitch sections corresponding to the pitch information and the sound source signal is outputted as a combination of the multipulse in one section and correction information regarding the multipulse. On a reception side, the signal is restored to a driving sound source signal and a speech synthetic signal is generated by using the spectrum parameter. Namely, a representative section for each frame is searched for in a voiced section to transmit the amplitude and position of the multipulse and amplitudes and phase correction coefficients of other pitch sections as sound source information and also send the spectrum and pitch parameter of a composing filter as auxiliary information.

Description

Detailed Description of the Invention [Field of Industrial Application] The present invention relates to a voice encoding/decoding method, a voice encoding device, and a voice decoding device, and in particular, the present invention relates to a voice encoding/decoding method, a voice encoding device, and a voice decoding device. The present invention relates to a speech encoding/decoding method that enables high-quality encoding and decoding with a relatively small amount of calculation in less than s, and an apparatus for encoding and decoding the same.

[Conventional technology]

As a method for encoding audio signals at a low bit rate of about 4.8 kb/s, for example, Japanese Patent Application No. 59-27243
The pitch interpolation multi-pulse method disclosed in Japanese Patent Application No. 5 and Japanese Patent Application No. 60-178911 is known. In this method, on the transmitting side, the spectral parameters and pitch parameters representing the spectral characteristics of the audio signal are extracted from the audio signal for each frame, and in the voiced section, the sound source signal of 1 frame is divided into pitch sections. One pitch section (representative section) among multiple pitch sections
is expressed as a multi-pulse, and the amplitude 1 position, spectrum, and pitch parameter of the multi-pulse in a representative section are transmitted. Furthermore, in the silent section, the sound source of frame (2) is represented by a small number of multipulses and a noise signal, and the five amplitude positions of the multipulse, the gain and index of the noise signal are transmitted.

On the other hand, on the receiving side, in voiced sections, the multi-pulses in the representative section and the multi-pulses in adjacent frames are used to interpolate the multi-pulses to restore the pulses in pitch sections other than the representative section, and the drive sound source signal of the frame is restore.

Furthermore, in the unvoiced section, the frame's sound source signal is restored using the index and gain of the multipulse and the noise signal.

Furthermore, the restored drive sound source signal is input to a synthesis filter using spectral parameters to output a synthesized speech signal.

[Problem to be solved by the invention]

However, in the conventional method described above, in voiced sections, the drive sound source signal of the frame is restored by interpolation between multi-pulses in the representative section. In frames where the signal characteristics are changing, the driving sound source signal restored by interpolation is significantly degraded, and as a result, the sound quality of the synthesized speech is degraded. It is known that areas where the characteristics of speech change significantly are extremely important for phonological perception and perception of naturalness, but conventional methods cannot sufficiently restore information in these areas, resulting in poor sound quality. There was a major problem: deterioration.

The purpose of the present invention is to solve the above-mentioned problems, and to provide an audio encoding/decoding method that achieves good sound quality even at a low bit rate with a relatively small amount of calculation, and an audio encoding device and audio decoding device suitable for the method. It is about providing.

[Means to solve the problem]

The audio 41 encoding/decoding method of the present invention includes inputting a discrete audio signal on the transmitting side, extracting a spectral parameter representing a spectral envelope and a pitch parameter representing a pitch from the audio signal for each frame. The frame section is divided into pitch sections according to the pitch information thereof, and the sound source signal of the audio signal is divided into pitch sections according to the pitch information thereof, and the sound source signal of the audio signal is divided into multipulses of one pitch section among the pitch sections and correction information regarding the multipulses, or a combination of noise and a pulse train. On the receiving side, the driving sound source signal of the frame is restored using the multi-pulse of the one pitch section, correction information regarding the multi-pulse, or a combination of the noise and pulse train, and the pisotiva parameter, and the spectral parameter is The feature is that the synthesized speech signal is obtained by using

The speech encoding device of the present invention also includes a parameter calculating means for extracting and encoding a spectral parameter representing a spectral envelope and a pitch parameter representing a pitch from an input discrete audio signal for each frame; The audio signal is divided into pitch sections according to parameters, and at least one of the amplitude and phase of the multipulse in one pitch section of the pitch sections and the multipulse in the other pitch sections are used as the sound source signal of the audio signal for each frame section. excitation signal calculation means for determining and encoding correction information for correction or a combination of noise and pulse train; and a multiplexer for outputting a combination of the output code of the parameter calculation means and the output code of the excitation signal calculation means. It is characterized by having

Furthermore, the audio decoding device of the present invention includes means for separately decoding a code representing a spectral parameter, a code representing a pitch parameter, and a code representing a sound source signal, and a means for decoding a frame in accordance with the decoded pitch parameter. The driving sound source signal of the frame is divided into pitch sections, generates a multi-pulse for one pitch section, and generates pulses in other pitch sections using correction information for correcting at least one of the amplitude and the phase of the multi-pulse. a driving signal restoring means for restoring the driving sound source signal of the frame by restoring the driving sound source signal or using a combination of noise and a pulse train; and a synthesis filter for obtaining and outputting synthesized speech using the driving sound source and the decoded spectral parameters. It is characterized by

[Effect]

According to the present invention, it is possible to divide a frame into pitch periods and represent a sound source signal in a voiced period using multipulses in one pitch period (representative period) and correction information in other pitch periods. More preferably, the correction information may be an amplitude correction coefficient or a phase correction coefficient.

Such processing of the sound source signal is effective in avoiding deterioration of the drive sound source signal in the conventional method, and makes it possible to obtain synthesized speech with good sound quality even in parts where the characteristics of the sound change greatly. In addition, since the sound source signal can be represented by a combination of noise and multipulses in areas other than voiced sections, good synthesized speech can be obtained even for various consonants.

〔Example〕

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of a speech encoding/decoding method and an encoding device and a decoding device therefor according to the present invention. Further, FIG. 2 is a diagram for explaining a representative section in a voiced frame, a multi-pulse of the representative section, an amplitude correction coefficient, and a phase correction coefficient.

As shown in FIG. 1, a transmission system for encoding and decoding audio signals is composed of an encoding device on the transmitting side and a decoding device on the receiving side.

In this embodiment, the transmitting side includes a buffer memory 110, a pitch analysis circuit 130, a pitch encoding circuit 150, and a parameter calculation circuit 140 for parameters representing spectral parameters.
and a K-parameter encoding circuit 160.

Further, an impulse response calculation circuit 170, an autocorrelation function calculation circuit 180, a subtracter 190, and a weighting circuit 20
0, a cross-correlation function calculation circuit 210, a sound source signal calculation circuit 220, an encoding circuit 230, and a multiplexer 26
0, and an amplitude/phase correction coefficient calculation circuit 27
0, a noise memory 225, a drive signal restoration circuit 283, a synthesis filter 281, and an interpolation circuit 282.

On the transmitting side, an audio signal is supplied to an input terminal indicated by reference numeral 100, and an encoded output is supplied via a multiplexer 260 to which outputs of a pitch encoding circuit 150, a K-parameter encoding circuit 160, and an encoding circuit 230 are supplied. Sent to the receiving side.

On the receiving side, as shown in FIG.
0, a sound source decoding circuit 300, a noise memory 310, a decoding circuit 315, a pitch decoding circuit 320, a K parameter decoding circuit 330, and a drive signal restoration circuit 3.
40, an interpolation circuit 335, and a synthesis filter circuit 350. The encoded output from the transmitting side is supplied to a demultiplexer 290, and synthesized speech is extracted through an output terminal 360.

Encoding of the audio signal input to the input terminal 100.

In the decoding process, on the transmitting side, a discrete audio signal is input, and from the audio signal, a spectral variable that represents the spectral envelope and a pitch parameter that represents the pitch are extracted for each frame, and the frame section is converted into the pitch information. The sound source signal of the audio signal is represented by a combination of a multipulse of one pitch section among the pitch sections, correction information regarding the multipulse, or noise and a pulse train, and on the receiving side, restoring the drive sound source signal of the frame using the multi-pulse of one pitch section and correction information regarding the multi-pulse or a combination of the noise and pulse train and the pitch parameter, and obtaining a synthesized speech signal using the spectral parameter. carried out by.

Hereinafter, the principle will first be explained with reference to the example shown in FIG. 2.

The speech encoding/decoding method and apparatus according to the present invention shown in FIG. For other pitch sections within the same frame, find the amplitude correction coefficient ck + phase correction coefficient d for the multi-pulse. Then, for each frame, as sound source information, the pitch position within the frame of the representative section, The amplitude correction coefficient cli + phase correction coefficient d of the multi-phase/L/S amplitude in the representative section and the other pitch sections in the same frame are transmitted, and the spectral parameters, pitch parameters, and voiced/unvoiced discrimination information are transmitted as auxiliary information. +Il is transmitted. The representative section may be found by searching for a section where the best synthesized speech signal is found, or may be fixed within the frame. The former has better sound quality, but requires more calculations.

Below, how to find the amplitude correction coefficient ck + phase correction coefficient dk,
We will show how to search for representative sections. Now, let T be the average pitch period found for each frame. Figure 2 (a) shows how one frame of audio waveform is divided into subframe sections of T.
, shown in (b). Here, a case will be described in which a representative section is searched.

For example, a subframe that is a candidate for the representative section is subframe ■. The amplitude 1 position of a predetermined number of multipulses is determined for subframe (■). As for how to obtain multipulses, it is known to use a cross-correlation function Φxh and an autocorrelation function Rhh.
“Mu” by zawa, Ono, Ochiai
lti-pulse Excited 5peech
Coder Ba5edon Maximum Cro
ss-correlation 5earch Alg
orithm.

(GLOBECOM 83. IEEE Global
Telecommunications Conf.
erence, lecture number 23.3.1983) (Reference 1), so detailed explanation thereof will be omitted here.

The two amplitude positions of the multi-pulse in the representative section are each g=
, mi (i possible 1 to L). This is shown in FIG. 2(c). Amplitude correction coefficient 01 in section k other than the representative section
1+ The phase correction coefficient d, is the synthesized speech x+, (n) synthesized for the section k using these and the synthesis filter,
Weighting with the audio Xk (n>) of the corresponding section can be determined to minimize the error power Ek.The weighting and the error power E are given by the following equation (1).

Ek=Σ ((xt+ (n)-1k(n))*w (
n) 1 sentence k (n)=(kΣ g=・h (n-mi
-T-dk) Here, w (n) indicates the impulse response of the perceptually weighted filter. However, this filter may not be provided. Further, h (n) represents an impulse response of a synthesis filter for synthesizing speech. ck.

dk can be obtained by minimizing the +11 formula. To do this, for example, first fix dk, partially differentiate equation (1) with respect to ck, set it to 0, and obtain the following equation.

Here, Xwk (n) and intersection w++ (n) are
Xwk (n) −)(k(n) *w (n
) ... (4a) xw, l (n) -Σ g
= Hh (n-mi-'r dh).

Therefore, find the value of equation (3) for various dkO values,
(3) Ek in equation (1) is minimized by finding a combination of dkick that minimizes C3.

In this way, C
k and dk are determined, and the error power E defined by the following formula for the entire frame is determined by the following formula (5).

E= Σ E, ・ ・
- (5) Here, N is the number of subframes included in the frame. However, the weighting of the representative pitch section (in the example of FIG. 2, the subframe section ■), the error power E2 is determined by the following equation.

*w(n))” ・ ・ ・
(6) To search for a representative pitch section, find the values of equations (1) to (6) for all candidates for representative pitch sections, and (5)
The section that minimizes the value of the expression can be set as the representative pitch section. Figure 2(c) shows the case where the representative pitch interval after the search is subframe ■, and the multi-pulse of the representative interval and the
) Then = 1. 2. 4-, 5) Sound source V, (n)
An example is shown in which is generated according to the following formula.

vt+(n)=chΣ gi・δ (n −mi−'
r - dh) ・ ・ ・ (7) In the voiced section, by the method explained above, a representative section is searched for each frame, and the amplitude and position of the multipulse in the representative section, and the amplitude 1 phase correction coefficient ck of other pitch sections are determined. , dk as sound source information, and further transmits the spectral parameters and pitch parameters of the synthesis filter as auxiliary information, solving the problems of the conventional method.
/s can provide good sound quality.

On the other hand, in the silent section, the sound source is multipulse and noise.

It is expressed as a combination of For this specific configuration,
Reference may be made to the specification of Japanese Patent Application No. 178911/1984.

Furthermore, the contents of the encoding process and decoding process will be specifically explained, including the operations of each element on the transmitting side and the receiving side in FIG.

In FIG. 1, on the transmitting side, an audio signal is input from an input terminal 100, and the audio signal for one frame is stored in a panzer memory 110. Pitch analysis circuit 130 calculates an average pitch period T from the audio signal of the frame. As this method, for example, a method based on an autocorrelation method is known, and for details, the pitch extraction circuit of each of the above-mentioned patent applications can be referred to. In addition to this method, other well-known methods (for example, cepstrum method, 5IFT method, phase change cabinet method, etc.) can be used. Pitch encoding circuit 150
outputs the code obtained by quantizing the average pitch period T with a predetermined number of bits to the multiplexer 260, and also outputs the code obtained by decoding the average pitch period T' to the sound source signal calculation circuit 220 and the interpolation circuit 282. , and output to the drive signal restoration circuit 283.

The K parameter calculation circuit 140 calculates K parameters of a predetermined order M by performing well-known LPG analysis from the frame audio signal as a parameter representing the spectral characteristics of the frame audio signal. Regarding this specific method, reference can be made to the parameter calculation circuit in each of the above-mentioned patent applications. In addition, the K parameter is PARC
The OR coefficients are the same. K parameter encoding circuit 1
60 is a multiplexer 260 which receives the code 1k obtained by quantizing the parameters with a predetermined number of quantization bits.
At the same time, it is decoded and further converted into linear prediction coefficients ai' (i=1 to M), and then sent to the weighting circuit 2.
00, interpolation circuit 282, impulse response calculation circuit 170
Output to. For the method of encoding K parameters and converting K parameters into linear prediction coefficients, reference can be made to the specifications of each of the above-mentioned patent applications.

The impulse response calculation circuit 170 uses the linear prediction coefficients to calculate the impulse response hw(n) of the perceptually weighted synthesis filter, and outputs it to the autocorrelation function calculation circuit 180. The autocorrelation function calculation circuit 180 is
The autocorrelation function Rhh(n) of the impulse response is calculated and outputted up to a predetermined delay time. For the operations of the impulse response calculation circuit 170 and the autocorrelation function calculation circuit 180, reference can be made to the specifications of each of the above-mentioned patent applications.

The subtracter 190 subtracts the output of the synthesis filter 281 by one frame from the frame audio signal x (n), and outputs the subtraction result to the weighting circuit 200 . Weighting is circuit 20
0 passes the above subtraction result through an auditory weighted filter whose impulse response is represented by w (n), and the weighted signal x, (
n) and output it.

For the weighting method, reference can be made to each of the above-mentioned patent applications.

The cross-correlation function calculation circuit 210 calculates the weighted signals x, (
n) and impulse response. (n) is input, the cross-correlation function Φxh is calculated and outputted up to a predetermined delay time. For this calculation method, reference can be made to the specifications of each of the above-mentioned patent applications.

The sound source signal calculation circuit 220 compares the pitch gain Pg with a predetermined threshold T to determine voiced or unvoiced. That is, when P, > Tゎ, it is voiced, p, <'r
, it is determined that there is no voice. Next, in the voiced section, as explained in the principle section above, the decoded average pitch period T
' is used to divide the frame into subframes for each pitch period in advance, and as a sound source signal, the position and amplitude of multipulses are determined for a pitch section that is a candidate for a typical one pitch section (representative section).

Next, the amplitude/phase correction coefficient calculation circuit 270 performs the above (3).
, (4a), (4b), calculate the multi-pulse amplitude correction coefficient Ck and phase correction coefficient dk for generating the sound source signal in another pitch section k. moreover,
These values are output to the sound source signal calculation circuit 220, and the sound source signal calculation circuit 220 calculates the error power E of the entire frame for several candidate sections based on the above-mentioned formulas (IL (5), (61), and The pitch section to be made the smallest is selected as the representative section, and the information P4 indicating the subframe number of the representative section, the amplitude g of the multipulse in the representative section positive 1 position mu (i = 1-1j, and the amplitude correction of other pitch sections Coefficient CI++ Phase correction coefficient dk is output.

On the other hand, in the silent section, the sound source signal is represented by a predetermined number of multipulses and a noise signal. A plurality of types of noise signals are stored in the noise memory 225 in advance, and the index and gain representing the type of noise are determined. These calculations are performed for each subframe obtained by dividing the frame into predetermined section lengths. The specific method is described in the above-mentioned Japanese Patent Application No. 17891/1989.
Reference may be made to Specification No. 1. In this case, what is transmitted as the sound source signal is the amplitude 1 position of the multipulse and the index and gain of the noise signal.

The encoding circuit 230 encodes the amplitude g of the multipulse in the representative section.
i, position m, is encoded with a predetermined number of bits and output. In addition, information Pl indicating the subframe of the representative section
, amplitude correction coefficient ck, and phase correction coefficient dk are encoded with a predetermined number of bits and output to the multiplexer 260. Furthermore, the driving signal restoration circuit 28 decodes these signals.
Output to 3.

In the voice section, the drive signal restoration circuit 283 converts the frame into the sound source signal calculation circuit 22 using the average pitch period T'.
0, and using the information P indicating the subframe of the representative section, the decoded amplitude 1 position of the multipulse in the representative section, a multipulse is generated in the representative section, and the subframes other than the representative section are generated. In the pitch section, the sound source signal ■ is generated according to the equation (7) using the multipulse of the representative section, the decoded amplitude correction coefficient, and the decoded phase correction coefficient.

(n) is restored.

On the other hand, in the silent section, multi-pulses are generated, and the noise signal is accessed from the noise memory 225 using the index of the noise signal and multiplied by a gain to restore the sound source signal. For details of the method for restoring the sound source signal in the silent section, reference may be made to the specification of Japanese Patent Application No. 178911/1983.

In the voiced section, the interpolation circuit 282 converts the linear prediction coefficients into parameters at once and calculates the pitch period T on the parameters.
' is interpolated for each subframe section, and the linear prediction coefficients are inversely exchanged and output. Interpolation is not performed in silent sections.

The synthesis filter 281 inputs the restored sound source signal, inputs the linear prediction coefficient, and obtains a synthesized speech signal for one frame, and calculates an influence signal for the next frame for one frame. is output to the subtracter 190. For the calculation method of the influence signal, reference may be made to the specification of Japanese Patent Application No. 57-231605.

The multiplexer 260 converts the code representing the sound source signal, voiced and
A code representing unvoiced, a code representing a subframe of the representative period in a voiced section, a code representing the average pitch period, and a code representing the K parameter are combined and output.

The above is a description of the operation on the transmitting side of this embodiment.

In this way, a parameter calculation circuit extracts and encodes a spectral parameter representing a spectral envelope and a pitch parameter representing a pitch from an input discrete audio signal for each frame, and a calculation circuit for extracting and encoding the spectral parameter representing the spectral envelope and the pitch parameter representing the pitch from the input discrete audio signal, and for correcting at least one of amplitude or phase with respect to the multipulse in one pitch section of the pitch sections and the multipulse in another pitch section as a sound source signal of the audio signal for each frame. a calculation circuit for an excitation signal that obtains and encodes a combination of correction information or noise and a pulse train; an output code of the calculation circuit for the parameter; and an output code of the calculation circuit for the excitation signal; The audio encoding process on the transmitting side according to the present invention can be realized by the audio encoding device configured to include an output multiplexer circuit.

On the other hand, audio decoding processing on the receiving side requires a circuit to separate and decode the code representing the spectrum parameter, the code representing the pitch parameter, and the code representing the sound source signal using a demultiplexer, and Divide into pitch sections according to the decoded pitch parameter, generate multi-pulses for one pitch section, and generate pulses in other pitch sections using correction information for correcting at least one of the amplitude or phase of the multi-pulses. a driving signal restoration circuit that restores the driving sound source signal of the frame or using a combination of noise and a pulse train, and synthesizes speech using the driving sound source and the decoded spectral parameters. This can be realized by a speech decoding device having a configuration including a synthesis filter that determines and outputs the following.

In other words, in the case of FIG. 1, on the receiving side, the demultiplexer 290 first inputs the combined codes, and demultiplexes the code representing the sound source signal, the code representing voiced/unvoiced, and the subframe of the representative period in the voiced period. The code, the code of the average pitch period, and the code representing the K parameter are separated and output.

The sound source decoding circuit 300 decodes the code representing the sound source signal and outputs it to the drive signal restoration circuit 340. Pitch decoding circuit 32
0 is the drive signal restoration circuit 340 that decodes the average pitch period.
is output to the interpolation circuit 355. The decoding circuit 315 inputs codes representing an amplitude correction coefficient and a phase correction coefficient, decodes them, and outputs them. It also decodes and outputs the code representing the subframe of the representative section.

The K parameter decoding circuit 330 decodes the code representing the parameter and outputs it to the interpolation circuit 335.

In addition to the decoded sound source information, the drive signal restoration circuit 340 also collects voiced/unvoiced information, in the case of voiced information, the decoded average pitch period, the decoded amplitude correction coefficient, the decoded phase correction coefficient, and the decoded representative. Enter the subframe position of the section,
Performs the same operation as the drive signal restoration circuit 283 on the transmission side,
The drive sound source signal of one frame is restored and output. Also,
The noise memory 310 has the same configuration as the noise memory 225 on the transmitting side.

The interpolation circuit 355 performs the same operation as the interpolation circuit 282 on the transmitting side, and performs linear interpolation for each decoded average pitch period of the parameter in the voiced section, and further converts this into a linear prediction coefficient and outputs it.

The synthesis filter circuit 350 inputs the restored excitation signal of the frame and the linear prediction coefficient, calculates one frame's worth of synthesized speech x (n), and outputs the synthesized speech x (n) through the terminal 360 .

Here, the operation of the synthesis filter is described in the above-mentioned patent application No. 57-231.
Reference may be made to the synthesis filter disclosed in the '605 specification.

This completes the explanation of the receiving side of this embodiment.

The embodiment described above is merely one configuration of the present invention, and various modifications thereof are possible.

For example, in the above embodiment, the sound source signal is represented by a small number of multipulses and a noise signal outside the voiced section, but this is done using well-known stochastic coding.
It can also be expressed using the following method. For details on this method, see for example “5chroeder, heta1”
Code-excitedlinear predic
tion (CELP): Lightquality
5peech at very low bit r
ates, (IC^SSP, 937-940.
1985) (Reference 2). moreover,
The noise signal stored in the noise memory 225.310 can be obtained by storing a white noise signal having a predetermined probability density characteristic (for example, Gaussian distribution, etc.), or by storing a large amount of audio in advance. It may be calculated by learning from a prediction residual signal obtained by predicting a signal. Regarding the latter method, see, for example, "Vector Quantization in 5" by Makhou et al.
peach Coding, (Proc, IE
EE, vol, 73.11.1551-1588.1
985) (Reference 3) etc.

Further, in the embodiment, the audio signal of a frame is classified into two types, voiced section and unvoiced section, and different sound source signals are used, but the number of classifications may be increased. For example, using phonetic knowledge, different sound source signals may be used for classifying sounds into vowels, nasals, fricatives, plosives, etc.

In addition, in the examples, the parameters were encoded as spectral parameters and LPG analysis was used as the analysis method, but other well-known parameters such as LPS, cepstrum, improved cepstrum, -C cepstrum, mel Cepstrum etc. can also be used. Moreover, the optimal analysis method for each parameter can be used. Also, the interpolation circuit 282. j3s
As for the parameters to be interpolated and the interpolation method thereof, other well-known methods can be used. A specific interpolation method is, for example, “5peech A” by Heta1 et al.
analysis and 5 synthesis by
Linear Prediction of 5pee
A paper entitled ch Wave (J, Acoust, S
oc, 8m,, pp, 637-6551971)
(Reference 4) etc. can be referred to.

Furthermore, in the voiced section, in pitch sections other than the representative section, the amplitude correction coefficient ck and the phase correction coefficient dk are determined and transmitted, but the phase correction coefficient ck is calculated by interpolating the decoded average pitch period T' for each pitch section. It is also possible to have a configuration in which the information is not transmitted. In addition, the amplitude correction coefficient is not transmitted for each pitch section, but by approximating the value of the amplitude correction coefficient determined for each pitch section by a least squares curve or a least squares straight line, and then transmitting the coefficient of the curve or straight line. It is also possible to configure such a configuration. These can reduce the amount of information for transmission of correction information.

In addition, as a subframe division method, as shown in FIG.
It is also possible to use a division method as disclosed in Japanese Patent Application No. 178911/1983.

In addition, in order to significantly reduce the amount of calculation, in voiced sections,
The representative section may be fixed to a predetermined section within the frame (for example, the pitch section approximately in the center of the frame, the pitch section with the largest power within the frame, etc.), or the representative section may not be searched. In this case, (5) for each candidate section.

The calculation of equation (6) becomes unnecessary, and the amount of calculation can be significantly reduced, but the sound quality deteriorates.

Furthermore, in order to further reduce the amount of calculation, calculation of the influence signal can be omitted on the transmitting side. As a result, the drive signal restoration circuit 283 and the interpolation circuit 282 on the transmitting side
, the synthesis filter 281, and the subtracter 190 are no longer necessary, making it possible to reduce the amount of calculations, but the sound quality still deteriorates.

Furthermore, an adaptive post filter that operates on at least one of the pitch and the spectral envelope may be added to the receiving side after the synthesis filter circuit 350 in order to make the quantization noise more audible by shaping it. good. Regarding the configuration of the adaptive postfilter, see, for example, Kroo
A C1assof Analysis by Mr. n et al.
by-synthesis Predictive C
orders for High Quality at
Rates between 4.8 and 16k
b/s, (IEEEJSAC, vol, 6.2.
353-363.1988) (Reference 5).

As is well known in the field of digital signal processing, the autocorrelation function is expressed as a power spectrum on the frequency axis.
Since the cross-correlation function corresponds to the cross-power spectrum, it can also be calculated from these. Regarding these calculation methods, please refer to Digi by Oppenheim et al.
tal Signal Processing (Pre
ntice-Hall, 1975).
Reference 6) can be referred to.

〔Effect of the invention〕

As explained above, according to the present invention, the voice source signal of the voiced section is divided into frames into pitch periods, and the multi-pulse of one pitch section (representative section) is combined with correction information, especially amplitude correction, for the other pitch sections. Since it can be expressed using coefficients and phase correction coefficients, it is possible to express not only vowel stationary intervals, but also
The great effect is that it is possible to obtain synthesized speech with almost no deterioration in sound quality even in parts where the characteristics of speech that are important for phonological perception and naturalness perception change (voiced transition parts and changes between vowels). There is. Furthermore, since the sound source signal can be represented by a combination of noise and multipulses outside the voiced section, there is a great effect that good synthesized speech can be obtained for various consonants.

Furthermore, it is possible to provide a speech encoding device and a speech decoding device suitable for such an encoding/decoding method with good sound quality.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the speech encoding/decoding method, speech encoding device, and speech decoding device of the present invention. FIG. 2 is a representative section and a representative section in a voiced frame used for explaining the present invention. FIG. 3 is an explanatory diagram showing multi-pulses, amplitude correction coefficients, and phase correction coefficients in a section. 110 ...Buffer memory 130 ...Pitch analysis circuit 140 ...K parameter calculation circuit 150
...Pitch encoding circuit 160 ...K parameter coding circuit 170 ...Impulse response calculation circuit 180 ...Autocorrelation function calculation circuit 190 ...Subtractor 200 ... Weighting circuit 210 ... Cross correlation function calculation circuit 220
... Sound source signal calculation circuit 225, 310 ... Noise memory 230 ... Encoding circuit 260 ... Multiplexer 270 ... Amplitude/phase correction coefficient calculation circuit 28
1, 350... Synthesis filter 282, 335... Interpolation circuit 283, 340... Drive signal restoration circuit Demultiplexer Sound source decoding circuit Decoding circuit Pitch decoding circuit and parameter decoding circuit

Claims (3)

[Claims] (1) On the transmitting side, a discrete audio signal is input, a spectral parameter representing the spectral envelope and a pitch parameter representing the pitch are extracted from the audio signal for each frame, and the frame section is determined according to the pitch information. dividing the audio signal into pitch sections, and representing the sound source signal of the audio signal as a combination of a multipulse of one pitch section among the pitch sections and correction information regarding the multipulse or noise and a pulse train; restoring the driving sound source signal of the frame using the multi-pulse in the pitch section and correction information regarding the multi-pulse or a combination of the noise and pulse train and the pitch parameter, and obtaining a synthesized speech signal using the spectral parameter. A speech encoding/decoding method characterized by: (2) a parameter calculation means for extracting and encoding a spectral parameter representing a spectral envelope and a pitch parameter representing a pitch from the input discrete audio signal for each frame; and dividing the frame interval into pitch intervals according to the pitch parameter. and correction information or noise for correcting at least one of the amplitude or phase of the multipulse in one of the pitch sections and the multipulse in another pitch section as the sound source signal of the audio signal for each frame section. and a multiplexer for outputting a combination of the output code of the parameter calculation means and the output code of the sound source signal calculation means. conversion device. (3) means for separately decoding a code representing a spectral parameter, a code representing a pitch parameter, and a code representing a sound source signal, and dividing a frame into pitch sections according to the decoded pitch parameter to obtain one pitch. A multi-pulse is generated for one pitch interval, and a pulse is generated using correction information for correcting at least one of the amplitude and the phase of the multi-pulse in another pitch interval to restore the driving sound source signal of the frame, or the noise and pulse train are generated. a driving signal restoring means for restoring the driving sound source signal of the frame using a combination of the above, and a synthesis filter for obtaining and outputting synthesized speech using the driving sound source and the decoded spectral parameters. conversion device.