JP2002366195A - Method and device for encoding voice and parameter - Google Patents

Abstract

PROBLEM TO BE SOLVED: To encode a voice pitch being an analysis/synthetic voice encoding parameter, voiced/unvoiced information and harmonic spectrum amplitude through the use of a small number of bits. SOLUTION: The sound pitch is encoded by changing-over a frame for quantization by the uniform/differential selection of a logarithm pitch and a frame for quantization by an inter-frame interpolation pitch candidate index. The voiced/unvoiced information is encoded by changing-over a frame for quantization by information of the voiced/unvoiced information which is selected from the representative voiced/unvoiced information and as frame without the transmission of voiced/unvoiced information. A harmonic spectrum amplitude string is made into a model by an autoregressive linear prediction model which is adjusted by spectrum. An LSP coefficient is encoded by changing-over a frame for vector quantization and a frame for quantization by an interpolation candidate index. An LPC gain is quantized by changing-over a frame for the uniform quantization of the logarithm gain and a frame for differential quantization or quantization by the interpolation candidate index after correcting the change of sound power which is generated by LSP coefficient quantization.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coding parameter code for digitizing a voice signal, obtaining voice coding parameters representing the characteristics of the voice signal at predetermined time intervals, and coding the obtained voice coding parameters. The present invention relates to an encoding method and apparatus, which transmits or stores the encoded audio encoding parameter, restores the audio encoding parameter when necessary from the transmission destination or storage destination, and converts the audio signal from the restored audio encoding parameter. It is suitable for use in digital cellular phones, digital voice storage devices, and the like that transmit voice through synthesis.

[0002]

2. Description of the Related Art Digitized audio signals can be subjected to various digital signal processing such as data compression, error processing, and multiplexing. Widely adopted. To digitize analog audio signals,
Generally, sampling is performed at a sampling frequency that is twice or more the input voice frequency band, and quantization is required in a quantization step that cannot be discerned by ears. Therefore, a wider transmission frequency bandwidth is required compared to analog signals. Therefore, the audio signal once digitized is subjected to data compression by various coding methods and modulation methods according to the required sound quality. As a method to obtain a high compression ratio of audio data, an analysis-synthesis-type audio encoding system that actively uses the features of audio and a method of efficiently quantizing the audio encoding parameters obtained from it are considered. Have been.

[0003] For example, the MBE (Multi-Band Excitation) method or IM which is partially used in a satellite cellular phone.
The BE (Improved Multi-Band Excitation) method is a type of this analysis-synthesis-type speech coding method. The speech is extracted at predetermined time periods (20 msec) into segments of a predetermined time length to form a frame. For each frame, a voice pitch (or a voice fundamental frequency as a reciprocal thereof), a voice harmonic spectrum amplitude sequence obtained from the frequency spectrum of the voice of the frame, voiced / unvoiced information for each frequency band obtained by dividing the frequency spectrum into an appropriate frequency region (Voic
ed / Unvoiced information or V / UV information) is used as a voice coding parameter, and for each frame, the voice pitch is 8-bit uniform quantization, and V / UV information v [k] (k is a band number) for each band. ) Is a K-bit quantization (K: variable length of up to 12 bits with the maximum number of bands) as a binary value represented by a binary number of 0/1, and a voice harmonic amplitude sequence is a two-dimensional transform of an inter-frame prediction difference value and its DCT. (Discrete cosine transform)
Quantization by K) bits gives a speech coding rate of 4.15 kbps.

FIG. 8 is a diagram showing a configuration of a general speech coded transmission apparatus. From the sampled and quantized audio digital signal input from the audio input terminal 301, the audio encoding parameter extracting unit 302 extracts a segment of a predetermined time length at a predetermined time period to form a frame, and A speech coding parameter is extracted every time. The voice coding parameters to be extracted differ depending on the voice coding method. For example, in the above-mentioned MBE method, it is a voice pitch, a voice harmonic spectrum amplitude sequence, and V / UV information of each frequency band. The parameter encoding unit 303 effectively encodes the extracted speech encoding parameters to reduce the code amount, and sends the encoded speech encoding parameters to the transmission path 305 via the transmission unit 304.
The signal received by the receiving unit 306 is restored to the speech coding parameters by the parameter decoding unit 307, and the speech
Reference numeral 08 generates a synthesized speech by the operation opposite to that of the speech encoding parameter extraction unit 302, and outputs a speech digital signal from a speech output terminal 309.

FIG. 9 is a block diagram of the speech coding parameter extraction unit 302 in the case of the MBE system. The digital input audio signal is input from an input terminal 301 to a fundamental frequency estimator 401, where the fundamental frequency of the audio is estimated. The estimated value of the fundamental frequency is calculated as the reciprocal value of the time at which the autocorrelation function of the time delay is maximized. The frequency spectrum calculation unit 402 obtains a voice frequency spectrum by frequency-analyzing a finite-length voice signal cut out from a frame using a window function such as a Hamming window. The fundamental frequency correction unit 403 performs an AbS (Analysis) under a minimum error condition between the estimated speech fundamental frequency, the spectrum synthesized by the window function, and the speech frequency spectrum.
-by-Synthesis) method to simultaneously obtain the corrected voice fundamental frequency ωo and the harmonic spectrum amplitude sequence. The voiced intensity calculation unit 404 divides the frequency band into a plurality of frequency bands k (k = 1, 2,..., K) based on the modified voice fundamental frequency ωo, and synthesizes a synthesized spectrum for each frequency band. And the difference between the voice frequency spectrum and the threshold
Outputs V / UV information v [k]. Spectrum envelope calculator 40
5 is based on V / UV information v [k].
For each harmonic spectrum amplitude and unvoiced band obtained by the method, the root mean square value of the frequency spectrum in the frequency band of each harmonic is output as a spectrum envelope | A (ω) |.

FIG. 10 is a block diagram of the parameter encoding unit 303 in the case of the IMBE system. The fundamental voice frequency ωo input to the input terminal 501 is uniformly quantized to 8 bits by a fundamental frequency quantization unit 502 in a predetermined quantization range and quantization step, and the quantized value
B0 is output to the output terminal 503. The V / UV information v [k] input to the input terminal 504 is input to a V / UV information quantization unit 505,
For example, when the number of frequency bands K is 12, it outputs to the output terminal 506 as a binary 12-bit value B1 represented by 12 pieces of information of 0 or 1. The spectrum envelope | A (ω) | input to the input terminal 507 is a discrete harmonic spectrum amplitude sequence | A (ωi) |, (i = 1, 2, 3,..., N, N: harmonic Number).

First, a logarithmic converter 508 performs logarithmic conversion, and a subtractor 509 predicts a harmonic spectrum amplitude sequence 521 predicted from the harmonic spectrum amplitude sequence of the previous frame.
(This is referred to as a “prediction difference value sequence”), and is passed to the block conversion unit 510. The block conversion unit 510 sequentially classifies the prediction difference value sequence into six types according to the order of harmonics to obtain two-dimensional data.
A DCT coefficient is calculated and passed to a T (discrete cosine transform) unit 511, passed to a quantization unit 512, and a code of a prediction difference value sequence is obtained by a combination of a uniform quantization method and a vector quantization method selected in advance according to the order of the DCT coefficient. The converted data B2 is output to the output terminal 513. Quantization restoration unit 514, inverse DCT
The unit 515 and the block restoring unit 516 restore the quantized prediction difference value sequence, add the result to the predicted harmonic spectrum amplitude sequence 521 in the adder 517, and restore the quantized spectrum envelope value of the input spectrum envelope of the current frame. You. The quantized spectrum envelope value is transmitted to the frame delay unit 51.
8, the spectrum envelope predicting unit 519 predicts the spectrum envelope value of the next frame based on the newly input voice fundamental frequency ωo of the next frame and the fundamental frequency of the previous frame. It is led to a subtractor 509 and prepares for quantization of the spectral envelope of the next frame.

[0008] The principle of this MBE system is that DWG Riffin an
d JSLim "Multi-band ExcitationVocoder", IEEE Tra
nsactions on Acoustics, speech, and signal process
ing, vol. 36, No. 8, August 1988, pp 1223-1235. Also, the encoding method of the encoder is USP-5491722 (Methods for speec
h transmission, Feb. 13, 1996).

On the other hand, in order to further increase the efficiency of the quantization of the harmonic spectrum amplitude sequence, a method of modeling the harmonic amplitude sequence with a linear prediction model (LPC) and quantizing the LPC modeled coefficients (linear prediction coefficient and gain). Has also been devised. (AMKondoz "Digital Speech", John W
iley & Sons, Ltd, 1995, pp. 256-261) LPC modeling is shown in equation (1).

(Equation 1) The harmonic spectrum amplitude sequence is modeled as an output spectrum of a J-order LPC synthesis filter H (ω) using a pulse sequence having a pitch G with a repetition pitch of amplitude G as a sound source signal,
A variable-length harmonic amplitude sequence is represented by an amplitude value (or gain, gain) G and J (for example, 10) LPC coefficients ak. It is well known that if the LPC coefficients are converted to a linear spectrum pair (LSP), the interpolation characteristics are excellent. By using this modeling together with the frame interpolation of the LPC coefficients, a lower bit rate can be obtained. Although it has been proposed that the sound quality can be improved, it is also reported that the sound quality is deteriorated. As described above, an analysis-synthesis-type speech coding scheme that extracts and encodes speech coding parameters based on a speech synthesis model has been proposed as a method of realizing low bit rate speech coding by digitizing speech. , Some of which are in practical use.

[0010]

The above-described analysis / synthesis type speech coding method is effective for low bit rate speech coding, but tends to be accompanied by sound quality deterioration peculiar to the analysis / synthesis type depending on analysis conditions. As a method for improving the synthesis sound quality of the analysis-synthesis type speech coding method, by setting a short speech frame update period, the change of the speech parameters in the speech frame is reduced, and the sound quality is improved while being the analysis-synthesis type. A way is being considered. Regarding the effect of improving the coded voice quality when the frame update period is set to be short, see, eg, "Study of Voice Coding Schemes for Commercial Mobile Communications", IEICE National Convention, D-14-2, p171, reported in Mar.2000. If the update period of the audio frame is shortened, the number of coded bits increases. Therefore, it is necessary to reproduce the coded parameters by interpolation or the like to reduce the parameter redundancy. However, if the number of bits of the parameter is reduced by simple interpolation, significant deterioration occurs. Like this
In order to achieve both a low bit rate of the analysis-synthesis type speech coding method and an improvement in speech quality by shortening the frame update period, there is a problem of an efficient quantization method of speech coding parameters. In the case of lowering the bit rate by using the reproduction of parameters, there is a problem that the quantization method is designed while sufficiently considering the balance between the interpolation method and the sound quality deterioration.

Accordingly, the present invention provides a speech encoding parameter encoding method and apparatus in an analysis-synthesis speech encoding system which can achieve both a lower bit rate and an improved speech quality by shortening a frame update period. It is intended to be.

[0012]

In order to achieve the above object, a speech encoding parameter encoding method according to the present invention comprises a speech encoding parameter obtained from an audio signal digitized and divided into frames of a predetermined time length. A speech pitch as the speech encoding parameter, wherein a quantization pitch is obtained by one of a quantization method and a difference quantization method or a uniform quantization method. Encoding by a combination of a frame and a frame to be quantized by an index of an interpolation pitch selected from a plurality of interpolation pitch candidates calculated using the quantization pitches of the preceding and succeeding frames; Unvoiced information is represented by an index of representative voiced / unvoiced information selected from a limited number of representative voiced / unvoiced information. Encoding, separating the harmonic spectrum amplitude sequence as the speech encoding parameter into a linear prediction coefficient by a linear prediction model or a line spectrum pair and a gain derived therefrom, and for the linear prediction coefficient or the line spectrum pair, a vector Interpolated quantization using a frame to be quantized by a quantizer such as a quantizer and an index of a candidate point selected from a plurality of candidate points obtained by a linear interpolator from quantized linear prediction coefficients of a preceding and succeeding frame or a line spectrum pair. Encoding in combination with a frame to be corrected, correcting the gain in accordance with a change in harmonic spectral power of the frame generated by quantization of the linear prediction coefficient, and obtaining a correction gain, and logarithmizing the correction gain. , The first gain quantizer as it is A frame for uniform quantization, and the output value of the difference quantizer that said first reference quantization gain of the gain quantizer of the previous frame,
From the output values of the interpolation quantizer obtained by selecting a plurality of interpolation candidates of the output of the first quantizer of the previous and subsequent frames,
The method includes a step of encoding the correction gain in combination with a frame to be quantized by a second gain quantizer that selects and quantizes a smaller error. Further, the step of obtaining the correction gain is obtained by a linear prediction model using a harmonic spectral power obtained by a sum of squares of a harmonic amplitude sequence before linear prediction modeling, a quantized linear prediction coefficient, and a gain before quantization. The correction gain is calculated by multiplying the gain by a ratio of the harmonic spectrum power obtained from the sum of squares of the harmonic spectrum amplitude value.

Further, the voice coding parameter coding device of the present invention is a voice coding parameter coding device for coding voice coding parameters obtained from a voice signal digitized and divided into frames of a predetermined time length. So,
The voice pitch as the voice coding parameter, a frame to be quantized by any one of the quantization methods by the selection of the difference quantization method and the uniform quantization method, and a plurality of calculated using the quantization pitch of the preceding and following frames Means for encoding by a combination of frames to be quantized by an index of an interpolation pitch selected from interpolation pitch candidates, and voiced / unvoiced information as the voice coding parameter are selected from a limited number of representative voiced / unvoiced information. Means for encoding with representative voiced / unvoiced information index, and a step of separating the harmonic spectrum amplitude sequence as the speech encoding parameter into a linear prediction coefficient by a linear prediction model or a line spectrum pair and a gain derived from the linear prediction coefficient, For coefficients or line spectrum pairs, quantize using a vector quantizer And a frame to be interpolated and quantized by an index of candidate points selected from a plurality of candidate points obtained by a linear interpolator from quantized linear prediction coefficients or line spectrum pairs of the preceding and following frames. Means, means for correcting the gain in accordance with a change in the harmonic spectral power of a frame generated by quantization of the linear prediction coefficient, to obtain a correction gain, and logarithmizing the correction gain, and as it is, a first gain quantizer. , The output value of the difference quantizer based on the quantization gain of the first gain quantizer of the previous frame, and the output of the first quantizer of the preceding and succeeding frames. A second gain amount for selecting and quantizing a value having a smaller error from the output values of the interpolation quantizer obtained by selecting the plurality of interpolation candidates And has a means for encoding the correction gain by a combination of a frame to be quantized by the encoder.

As described above, in the present invention, the coding of the voice pitch is performed with respect to the logarithmically converted pitch.
Switching between the differential quantization method and the uniform quantization method, selecting the frame with the smaller error from the input voice pitch and quantizing, and the number of the interpolation point candidate closest to the plurality of interpolation points of the inter-frame pitch Is selected, and the frame to be quantized with the selected number is switched and used by frame repetition, thereby reducing the number of encoded bits of the pitch. For voiced / unvoiced information (V / UV information) coding, V / UV information and its occurrence frequency are obtained in advance for many voice frames, and a fixed number of representative V / UV information are obtained from among them. Information is selected in advance, and from the representative V / UV information,
Means is used for encoding with the number (index) of the representative V / UV information most similar to the V / UV information. In addition, a frame that does not transmit V / UV information is appropriately inserted, and decoding of a frame to which V / UV information is not transmitted is performed in order to decode a frame having higher sound energy among the preceding and following frames. V / UV
It restores using the information. The number of encoded bits of V / UV information is reduced by the above two means.
Further, regarding encoding of the harmonic spectrum amplitude sequence, the harmonic spectrum amplitude sequence is modeled by an autoregressive J-order linear prediction model (AR model), and is represented by a linear prediction coefficient (LPC) and a gain. The LPC coefficient is converted into an LSP (line spectrum pair), and then the vector is subjected to vector quantization to perform vector quantization of J LSPs per frame, and an interpolation point closest to a plurality of interpolation points between the quantized LSP frames. A candidate number is selected, and quantization is performed using a combination of frames to be quantized with the selected number. In addition, the gain corrects a change in frame power caused by an error in harmonic amplitude when the linear prediction coefficient is restored from a linear prediction coefficient quantized by LSP vector quantization and interpolation, and a logarithmic value of the corrected gain. Means for quantizing by a combination of a frame in which uniform quantization is performed, a case in which an amount of difference from the previous frame is quantized, and a frame in which an error is smaller when an interpolation value between the preceding and succeeding frames is selected and a gain having a smaller error is selected. Take.
By performing quantization of the harmonic spectrum amplitude sequence in the two-stage configuration as described above, encoding of the harmonic amplitude sequence at low bits is performed while suppressing a change in frame power caused by a quantization error of the linear prediction coefficient. be able to.

[0015]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of a speech coding parameter coding method and a speech coding parameter coding apparatus to which the coding method is applied according to the present invention are the analysis-synthesis type speech coding method. A case where the present invention is applied to the MBE or IMBE speech encoding method will be described as an example. It should be noted that this speech coding parameter coding apparatus is the same as that shown in FIG.
And the speech encoding parameters extracted by the speech encoding parameter extraction unit 302, that is, the speech pitch (or the speech fundamental frequency ωo, which is the reciprocal thereof), and the speech harmonic spectrum. Efficiently encode the amplitude sequence and voiced / unvoiced information (V / UV information) of each frequency band.

FIG. 1 is a block diagram showing an example of the configuration of a speech coding parameter coding apparatus to which a speech coding parameter coding method according to the present invention is applied. For example, FIG.
The voice pitch (or voice fundamental frequency ωo) obtained by the voice coding parameter extraction unit shown in is input to the input terminal 101, the voice pitch is logarithmically converted by the logarithmic conversion unit 102, and the logarithmic voice pitch P [n] ( n is the frame number).
Logarithmic voice pitch is described in the literature (Thomas Eriksson and Hong-Goo
Kang, "Pitch Quantization in low Bit-Rate Speech
Coding ", ICASSP '99, pp489-492, 1999), it is known that the limit of human detection to the change in logarithmic pitch is not significantly affected by the value of logarithmic pitch. Therefore, since the quantization step width can be made uniform, the conversion is convenient.

The logarithmic pitch P [n] is alternately switched for each frame (or for each sub-frame when sub-frames are formed) by the input switching unit 103, and the two output terminals 1
04 or 116 is output. When output to 104, it is guided to uniform quantization section 105 and subtraction section 112. The uniform quantizing section 105 performs uniform quantization at a fixed quantization step, and the quantized logarithmic pitch P1 ′ [n] (1
06) is input to the pitch comparison unit 108. On the other hand, the subtraction unit 112 obtains the difference logarithmic pitch from the input logarithmic pitch and the quantization logarithmic pitch of the previous frame received from the delay unit 111 and inputs the difference logarithmic pitch to the difference quantization unit 113. The delay unit 111 passes the quantized logarithmic pitch of the immediately preceding frame to the current frame. The difference quantization unit 113 performs the difference quantization in a uniform difference quantization step or in a non-uniform quantization step set such that the difference quantization step expands as the difference input amplitude increases with reference to the difference value of zero. The unit 114 adds the result to the quantization logarithmic pitch of the previous frame, which is used as the reference for the difference, and outputs the quantization logarithmic pitch P2 ′ [n] by the difference quantization to 107. The pitch comparison unit 108 compares P1 ′ [n] and P2 ′ [n], selects a quantized logarithmic pitch having a smaller error from the logarithmic pitch P [n] before quantization, and generates a uniform quantization index. The index output from the quantizer selected from N1 and the differential quantization index N2 is output as a pitch code to one input of the pitch code switching unit 110. By arranging the indexes (selection numbers) of N1 and N2 so that the numbers do not overlap, it is possible to determine which quantization method has been selected from the output index numbers. An output terminal 109 outputs the quantized logarithmic pitch P ′ [n] from the quantizer selected by the pitch comparing unit 108.

The other output 116 of the input switching unit 103
The logarithmic pitch P [n] appearing on the basis of the quantized logarithmic pitches of the current frame and the previous frame obtained from the input / output terminal of the delay unit 111 is used as the plurality of interpolation pitch candidates created by the interpolation pitch candidate creation unit 117. Are compared by the interpolation point comparison unit 119,
An interpolation point index N3 that gives a pitch closest to the logarithmic pitch 116 of the current frame is output to the other input of the pitch code switching unit 110 as a pitch interpolation code. Pitch code switching section 110 switches and outputs a pitch code in accordance with the movement of input switching section 103.

FIG. 2 shows an interpolation pitch candidate creating section 11 shown in FIG.
FIG. 7 is a diagram for explaining the operation of No. 7; The example shown in FIG. 2 is an example in which the number of interpolation point candidates is 4 and quantization is performed with 2 bits using the selection number (index). The quantized logarithmic pitch of the front frame of the current frame is P '[n + 1], and the quantized logarithmic pitch of the rear frame is P' [n-1]. Is indicated by a cross. Among the four interpolation pitch candidates, the interpolation quantization pitch P ′ [n] closest to the input logarithmic pitch P [n] is selected, and 2 is selected as an index for giving the interpolation quantization pitch in this example. . P [n] is the pitch of the frame between P '[n + 1] and P [n-1]. For example, if the frame is divided into two subframes, P [n] is The first subframe of the current frame, P ′ [n + 1], corresponds to the quantization pitch of the second subframe of the current frame, and P ′ [n−1] corresponds to the quantization pitch of the second subframe of the previous frame. ing. Although P '[n + 1] and P' [n-1] are not included in the interpolation pitch candidates in the arrangement of the interpolation pitch candidates in FIG. 2, the interpolation pitch candidates are represented by P '[n + 1] at both ends. P '[n-1]
Can also be set. In that case, P '[n +
Interpolation pitch candidates other than [1] and P '[n-1] are two points. FIG.
When the position of the interpolation pitch candidate is set excluding both ends as in the example of FIG. 2, two points excluding both ends can be selected even with one bit. Therefore, the interpolation point arrangement of FIG. It can be said to be effective when there are as few as 2 bits.

Returning to FIG. 1, voiced / unvoiced information (V / UV
Information) is input from the input terminal 131, and the V / UV information is converted into a frame (or subframe) by the frame thinning unit 133.
Thinned out. For example, V / UV information is output only once for two frames (or two subframes) and input to voiced / unvoiced comparison section 134. The representative voiced / unvoiced information codebook 132 stores V / V obtained in advance from many voice frames.
In the method described later, a limited number is selected from the most frequently generated UV information and stored. Currently entered
The voiced / unvoiced comparison unit 134 selects a representative V / UV information value b1 ′ closest to the V / UV information value b1 and outputs the index of the representative V / UV information as a voiced / unvoiced code 135. V /
The UV information value b1 (or b1 ′) is a V / UV information value v [k], k = 1, 2,... For each frequency band obtained by dividing the audio frequency spectrum by, for example, three times the interval of the audio fundamental frequency. , K (V [k] is 0 or 1) assigned to each bit of the binary number.

(Equation 2) The distance between the V / UV information values is represented by the absolute value of the difference between b1 and b1 '. The representative V / UV information is set independently for each band number determined by the audio fundamental frequency.

The representative V / UV information can be selected and created by the following method. That is, a large number of V / UV information values are obtained in advance from a large number of audio frames and are classified for each band number K. A set of V / UV information values classified for each band number {b1 _i }
, The frequency of occurrence of each b1 _i is tabulated. For example, when the number of bands is 5, the value of b1 takes an integer value of 0 to 31, and the occurrence frequency is totaled for each b1 value. For example, when quantizing V / UV information with 2 bits, it is necessary to select four types of representative V / UV information from among them. For this selection, a method of selecting four items from the highest occurrence frequency is conceivable. However, in some cases, adjacent representative V / UV information values may be selected, which is not appropriate as the representative V / UV information value. Cases occur. In particular, when the number of bands K is large, there is a high possibility that a representative V / UV information value having a high frequency of occurrence exists adjacently.

For this reason, V / UV information having the lowest frequency of occurrence
Distribute the frequency of occurrence to the frequency of occurrence of adjacent V / UV information
From the target V / UV information
The method of reducing becomes effective. Figure 3 shows V / UV information (V / UV pattern
), The number of occurrences with the highest frequency
Figure explaining how to erase low V / UV patterns
is there. Here, the least frequently occurring V / UV pattern
If the frequency is q [n], q [n] is divided into q1 and q2
V / UV pattern occurrence frequency q [n-l₁] And q [n + l_Two]
Minute. The allocation amount of q1 and q2 is the frequency of occurrence of adjacent V / UV pattern
q [n-l₁] And q [n + l _Two] Size and up to adjacent V / UV pattern
Distance l₁And l_TwoIs determined by the following equation according to the proximity of

(Equation 3)

Returning to FIG. 1, coding of the spectral envelope will be described. For the spectral envelope, a discrete spectrum is input terminal 1 as a harmonic spectrum amplitude sequence.
The signal is input to the spectrum correction unit 141. The operation of the spectrum correction unit 141 will be described later. The spectrum sequence corrected by the spectrum correction unit 141 is modeled by an autoregressive linear prediction model by a linear prediction modeling unit 142, and gain G and a higher order (for example, J order, J = 10)
(LPC coefficient a [j], j = 1,2,3, ..., J, J
Is predicted order). Further, the LPC coefficient a [j] is converted to a line spectrum pair (LSP) F [j]. LSP
Has a value from 0 to π and has little perceptual deterioration due to linear interpolation. Therefore, it is widely and generally used as a coefficient for modeling a spectral envelope. LP converted to LSP
The C coefficient is calculated by the LSP quantizer 143 in the LSP codebook 1
44, and the index of the codebook is output to the LSP code switching unit 146.
Here, the LSP quantization unit 143 outputs one frame (or one frame).
The number of frames to be quantized is reduced by performing quantization every other subframe). On the other hand, the quantized LSP enters the LSP interpolation unit 145, and calculates a plurality of LSP interpolation candidates between the quantized LSP immediately before and immediately after,
L closest to the LSP of the current frame from the LSP interpolation candidate
Select an SP interpolation candidate and enter its LSP interpolation candidate number in LS
Output to P code switching section 146. LSP quantizer 143
In the frames that have not been quantized by the above, the increase in the code amount is suppressed by performing the quantization by the LSP interpolation. L
In the SP code switching unit 146, the LSP quantization unit 143 or the LSP interpolation unit 1 is used for each frame (or for each subframe).
The quantization index is switched from 45 and output to the output terminal 154 as an LSP code.

Here, the function and purpose of the spectrum correcting section 141 will be described. When modeling a harmonic spectrum amplitude sequence with a linear prediction model, it is better to have many spectral points, but if there is a spectrum that does not fit the model, it distorts the modeling, and the spectral error after modeling Increase. Generally, the 0th-order harmonic spectrum amplitude (DC component) of a voice is lower than other harmonic spectrum amplitudes, and a modeling error is likely to occur. Also, since the 0th-order harmonic spectrum amplitude is an unnecessary component at the time of audio decoding (the audio signal hardly contains a DC component), it can be said that the amplitude may be adjusted and changed to a level that is easy to model. For the above reasons, the spectrum correction unit 141 corrects the 0th-order harmonic spectrum amplitude so that it can be easily modeled by a linear prediction model. Specifically, a value obtained by correcting the second harmonic amplitude H2 by the coefficient α according to the following equation is replaced with the 0th harmonic amplitude.

(Equation 4) Here, H ₀ ′ is the replaced zero-order harmonic spectrum amplitude, and H ₁ , H ₂ , and H ₃ are the first-order, second-order, and third-order harmonic spectrum amplitudes, respectively. Further, since the number of samples of the harmonic spectrum amplitude sequence decreases depending on the pitch frequency, the input harmonic spectrum amplitude sequence is interpolated between discrete spectra. Here, the logarithmic value of the harmonic spectrum amplitude is linearly interpolated with respect to the frequency to create interpolated spectrum data.

The harmonic restoration unit 153 calculates the spectral amplitude of the harmonic of the pitch frequency from the LPC modeling formula (1) using the LPC coefficient ak restored from the quantized LSP code and the LPC gain G before quantization. Calculated as the harmonic spectrum amplitude restored
52. The error calculator 152 calculates an error of the spectrum power with respect to the original input harmonic spectrum amplitude sequence A [l] and inputs it to the gain corrector 147. The gain correction unit 147 corrects the power error of the gain G of the linear prediction modeling unit 142 based on the output of the error calculation unit 152, and the corrected gain is used as the gain quantization unit 1
48, where the corrected logarithmic gain is uniformly quantized. The gain code of the uniformly quantized gain is output to one input of the gain code switching unit 151. The gain quantization unit 148 reduces the number of frames to be quantized by performing quantization every other frame (or every other subframe).

For a frame that has not been subjected to quantization, the gain difference quantization section 149 or the gain interpolation quantization section 150 performs quantization to suppress an increase in the code amount. The gain difference quantization unit 149 performs difference quantization on the logarithmic gain of the frame that has not been quantized by the gain quantization unit 148, with reference to the quantization gain quantized immediately before by the gain quantization unit 148. Similarly, the gain interpolation quantization unit 150 calculates the logarithmic gain of the frame not quantized by the gain quantization unit 148 from the interpolation gain candidates obtained by linear interpolation from the quantization gains of the immediately preceding and subsequent frames. The interpolation gain having the smallest error is selected and interpolation quantization is performed. Then, of the outputs of the gain difference quantization unit 149 and the gain interpolation quantization unit 150, the gain code of the gain with the smallest quantization error is output to the other input of the gain code switching unit 151. Here, the gain difference quantization unit 14
9 and the gain interpolation quantization unit 150 are set such that there is no overlap, so that it is possible to know which quantization method has been selected, and it is not necessary to send a selection code. The gain code switching unit 151 switches between the gain code of the gain quantization unit 148 and the gain code of the gain difference quantization unit 149 or the gain interpolation quantization unit 150 for each frame (or for each subframe) and outputs them as LSP gain codes. Output to terminal 155.

As described above, by correcting the frame power error caused by the change in the amplitude of the harmonic spectrum due to the LSP quantization to the gain before quantization, it is generated especially in a transient state such as when a sound is generated or disappeared. It is possible to suppress noise generation due to an excessive error in the audio amplitude of a frame (or a subframe) that is likely to occur.

Next, the flow of processing of the speech coding parameter coding section shown in FIG. 1 according to the present invention will be described. FIG. 4 is a flowchart of quantization of the fundamental voice frequency ωo. The process is started from 701 in FIG. In step 702, a voice fundamental frequency ωo and a frame number m to be quantized are set. Next, the logarithmic pitch P is calculated in 703, and the evenness or oddness of m is determined in 704. If it is even (EVEN), 705 and 70
In step 6, the uniform quantization and the difference quantization are performed, and the uniform quantization pitch P_u and the uniform quantization index Index_u, and the difference quantization pitch P_d and the difference quantization index Index_d
Is calculated.

Next, at step 708, the difference quantization error (| P_d-P
|) Is determined, and if it is smaller than a certain threshold Th, the pitch P [m] of the m-th frame (this is represented as P [2n] in the meaning of an even frame) is P_d at 710, and its index Index [m] ( This is expressed as Index [2n] in the meaning of an even-numbered frame.)
And On the other hand, if it is determined in 708 that | P_d-P | is equal to or greater than Th, the uniform quantization error (| P_u-P |) is determined in 707.
And the difference quantization error (| P_d-P |). If the uniform quantization error is small, the pitch P [2n] of the m-th frame is determined at 709.
Is P_u, the index Index [2n] is Index_u, and in the opposite case, the pitch P [2n] is P_d and the index [2n] is Index at 710.
_d. The pitch P [2n] of the m-th frame is 711.
Is delayed by 2 frames to P [2n-2], and the differential quantization 7
The reference logarithmic pitch of 06 is used as the reference logarithmic pitch when the next even frame is differentially quantized.

If m is an odd number in step 704, pitch selection candidates are selected in step 713. A pitch interpolation candidate is a set of interpolation pitch candidates {Pinpol [2n-1] _i } (i = 0,1) obtained by equally dividing the interval between the quantized logarithmic pitches of the preceding and succeeding frames into a plurality of pieces by the method shown in FIG. , 2, 3,..., N−1, N are calculated at 712 as interpolation candidate points). To select a pitch interpolation candidate, an interpolation candidate point having the smallest error absolute value from the logarithmic pitch P [2n-1] of m = 2n-1 frames is selected, and its index number Index [2n-1] is set to the odd frame number. The quantized pitch interpolation code is set at 714. Note that 71
The argmin_i (x) function (function in which i is written below argmin) in 3 is a function that evaluates i as a parameter and returns i that minimizes x. In the case of an even-numbered frame, the pitch code Index_d or Index_u of the quantization method selected in 709 or 710 is similarly set in 714 as Index [2n]. The pitch code of the even frame and the pitch interpolation code of the odd frame set in this way are 7
15 is output. If the speech fundamental frequency encoding method according to the present invention shown in FIG. 4 is used, for example, 4
1 bit for bit and pitch interpolation code, 2
The audio fundamental frequency for the frame can be encoded.

FIG. 5 is a flowchart of encoding V / UV information.
The processing starts at 801 and the fundamental frequency of the sound is changed to ω at 802.
o, V / UV information value of frame to v, frame number to m
Set to. In step 803, the number of bands K for ωo is obtained.

(Equation 5) Here, floor (x) is a function indicating the maximum integer value not exceeding x, B is the number of harmonics included in each band, which is determined in advance before encoding, and about 3 is used. . At step 804, the even / odd of m is determined. At step 805, when m is an even number, the data is closest to the most inputted V / UV value v from the representative V / UV information value data 806 for each band number k selected in advance. The index i of the representative V / UV information value VCB [K] [i] having a value is selected and set as IndexV / UV [2n]. At 807, a voiced / unvoiced code is set. If the processing frame is an odd number, the voiced / unvoiced code is thinned out and not output. Using the V / UV information encoding method according to the present invention shown in FIG.
Using 2 bits for unvoiced code, V /
UV information can be encoded.

FIG. 6 is a flowchart of encoding a harmonic spectrum amplitude sequence. Processing starts from 901 and 902
To set the harmonic spectrum amplitude to A [l] and the frame number to m. In step 903, the 0th-order harmonic spectrum amplitude A [0] is corrected by multiplying the second-order harmonic spectrum amplitude A [2] by α. The method of determining the correction coefficient α is determined by the method shown in the above equation (4). 904
Linearly interpolates the logarithmic value of the harmonic spectrum amplitude sequence corrected by (1) to increase the number of spectrum sequences. In step 905, modeling is performed using the spectrum sequence increased by interpolation as a value on a modeling curve in the linear prediction model expressed by the equation (1), and an LPC coefficient a [j] and an LPC gain G are calculated.
Further, the LPC coefficient a [j] is converted into a valid line spectrum pair (LSP: Line Spectrum Pair) F [j] in a later process.

In step 906, the even frame number and the odd frame number of the current frame number are determined. If the frame number is even, F [j] is vector-quantized in step 907 using the vector quantization table 908, and the quantized LSP vector F '[j ] And the index LSPindex0 of the quantized LSP vector, and outputs LSPindex0 to 914. In addition, the quantized LSP vector F ′ [j] is calculated together with the quantized LSP vector whose frame has been delayed in 910 and the LSP interpolation candidate is calculated in 911 by linear interpolation. On the other hand, 90
If the current frame number is odd in 6, the LSP interpolation candidate F ′ [j] having the closest value to the LSP vector F [j] of the current frame before quantization is selected from among the LSP interpolation candidates calculated in 911. Is selected, and the index of the selected LSP interpolation candidate is output to 914 as LSPindex1. In 914, if the frame number is even, LSPi is used as the LSP code.
Output ndex0, LSPindex1 for odd numbers.

After outputting the LSP code at 914, the quantized LSPs of the even and odd frames are changed from the quantized LSP coefficient F ′ [j] at 909 to the quantized LPC coefficient a ′ [j], and quantized at 912. LPC coefficient a '[j] and L before quantization
From the PC gain G, the harmonic amplitude sequence A ′ [l] is restored from the quantized spectrum amplitude value of the harmonic of the fundamental sound frequency by the modeling equation (1). At 913, the gain change rate dg by LSP quantization is calculated from the ratio of the sum of squares of the harmonic amplitude sequence A [l] before quantization to the sum of squares of the restored harmonic amplitude sequence A '[l].

On the other hand, at 915, a corrected logarithmic gain G ′ = log (dg × G) is calculated using the gain change rate dg calculated at 913 with respect to the LPC gain G modeled at 905. Even number (EVEN) / odd (ODD) of frame number at 916
The LPC gain of the even-numbered frame is determined by uniformly quantizing the logarithmic gain G ′ at 917 to obtain a uniform quantization gain Gu ′ and outputting the index Gindex0 to 922.
The uniform quantization gain Gu ′ is output to the interpolation quantization 920 together with the one obtained by delaying the frame by 918.

At 919, the difference quantization is performed between the logarithmic LPC gain of the current frame and the quantization logarithmic gain of the immediately preceding frame (even frame) that has been uniformly quantized, and the quantization gain Gd ′ and its index Gindex_d are calculated. I do. 9
In step 20, an interpolation gain candidate is selected by linear interpolation or the like from the uniform quantization gains of the preceding and succeeding frames, an interpolation gain candidate closest to G ′ of the current frame (odd frame) is selected, and the quantization gain Gi ′ and its index Gindex_i are determined. calculate.
At 921, an index closer to the LPC gain G ′ before quantization is selected from the quantized LPC gains Gd ′ and Gi ′ of the odd-numbered frames, and is output to 922 as Gindex1.
At 922, Gindex0 and Gindex1 are switched according to the gain index according to the even number and odd number of the frame, and output as the LPC gain code Gindex. At 923, this processing ends.

Using the encoding method of the harmonic spectrum amplitude sequence according to the present invention shown in FIG.
A total of 25 bits of 17 bits for the code and 8 bits for the LSP gain code (5 bits for the uniform quantization + 3 bits for the difference quantization and the interpolation quantization) are used to generate a harmonic spectrum amplitude sequence for 2 frames (or 2 subframes). Can be encoded. According to the speech encoding parameter encoding procedure shown in FIGS. 4, 5, and 6, speech encoding parameters for two frames (or two subframes) are obtained by combining the direct method, the difference method, and the interpolation method. It is possible to provide a low-bit quantization method for speech coding parameters that efficiently quantizes and suppresses a change in frame power due to quantization. For example, 20 msec of one voice frame (corresponding to two subframes of 10 msec subframe) can be coded with 32 bits, and as a result, a 1.6 kbps low bit rate voice coding method is realized. can do.

When decoding the coded voice coded by using the voice coding parameter coding method of the present invention, the parameter decoding unit 307 shown in FIG.
Then, the processing reverse to that described above may be performed to restore the voice fundamental frequency (or voice pitch), voiced / unvoiced information, and the harmonic spectrum amplitude sequence, and the synthesized voice may be generated by the voice synthesis unit 308 using the restored sequences. An example of such a speech decoding unit and a speech synthesis unit will be described.

FIG. 7 is a block diagram showing an example of the configuration of a voice decoding unit and a voice synthesis unit for decoding the coded voice coded by the voice coding parameter coding apparatus shown in FIG. Via a receiving unit (not shown), voiced / unvoiced information, pitch information, LS
The P code and the LSP gain code are terminals 1001, 1 respectively.
004, 1006, and 1009. The voiced / unvoiced code 1001 is input to the voiced / unvoiced code decoding unit 1003, where the representative voiced / unvoiced information codebook 132 has the same contents as the representative voiced / unvoiced information codebook 132 in the voice coding parameter coding unit of FIG. 1002 is used to restore voiced / unvoiced information. The voiced / unvoiced information of the frame that has not been transmitted is replaced with voiced / unvoiced information of a frame having a large frame power from the preceding and following frames.
The pitch code 1004 is input to a pitch decoding unit 1005, and a uniform / differential quantization method is determined based on the code value, and the pitch is restored by a corresponding inverse quantization method. With respect to the pitch of a frame whose pitch has been encoded by interpolation, the pitch is restored by the inverse operation of the interpolation performed at the time of encoding from the pitch of the preceding and succeeding frames using the pitch code as the pitch interpolation code.

The LSP code 1006 is used by the LSP decoding unit 10
08, the LSP code book 10 having the same contents as the LSP code book 144 of the speech coding parameter coding unit in FIG.
07 is used to restore the LSP. With respect to the LSP of the frame in which the LSP has been encoded by interpolation, the LSP of the preceding and succeeding frames is used to restore the LSP by the inverse operation at the time of encoding. The LSP gain code 1009 decodes the LSP gain by the inverse operation of the gain quantization unit 148 in FIG. Further, for a frame in which the LSP gain has been subjected to difference or interpolation quantization, a corresponding inverse quantization method is determined from the LSP code value to decode the LSP gain. From the LSP output from the LSP decoding unit 1008 and the LSP gain output from the gain decoding unit 1010, the harmonic amplitude calculation unit 1011 calculates the LP shown in the above equation (1).
The harmonic amplitude sequence is restored using the C synthesis method.

The unvoiced gain setting section 1012 regards the output of the noise generation section 1013 as a random noise spectrum, and calculates the noise spectrum power of the frequency band in which the voiced / unvoiced signal is unvoiced to the power of the harmonic amplitude sequence of the corresponding frequency band. The level of the noise spectrum is adjusted for each frequency band so that In the inverse FFT unit 1014, the noise spectrum whose level has been adjusted for each frequency band is
The real part expands to the negative frequency side symmetrically with respect to the zero frequency, and the imaginary part inverts the polarity with respect to the zero frequency and expands to the negative frequency side to perform inverse FFT. To obtain the audio signal. The frame interpolation unit 1015 interpolates and synthesizes the time axis audio signal obtained by the inverse FFT between frames to obtain an unvoiced audio synthesis signal.

On the other hand, in the voiced gain setting section 1016, for a frequency band in which the voiced / unvoiced signal is voiced, the harmonic amplitude value is calculated as the value of the harmonic amplitude sequence in the corresponding frequency band obtained from the harmonic amplitude sequence calculation section. And set the harmonic amplitude values in the other frequency bands to zero. In the harmonic synthesizing unit 1017, the initial phase is each harmonic initial phase from the phase reproducing unit 1018, and the amplitude is generated by a sine wave which is the harmonic amplitude value set by the voiced gain setting unit 1016, and the sum is voiced. To obtain a speech synthesis signal of the frequency band of Here, the phase reproducing unit 1018 sets the initial phase of each harmonic sine wave so as to maintain the phase continuity of each harmonic between frames, and at the same time, disturbs the continuity of the initial phase, thereby forming a simple sine wave. Prevents buzzing sound due to synthesis. The frame interpolator 1019 smoothes the amplitude change between frames and prevents a sharp level change between frames. The voice-synthesized signal of the voiced band and the voice-synthesized signal of the unvoiced band are added by the adder 1020, and after the post-filter unit 1021 performs sound quality improvement filter processing on the audibility, the final synthesized voice signal is supplied to the terminal 1
022.

In the above description, the harmonic spectrum amplitude sequence is quantized by the line spectrum pair (LSP) and the gain, but may be quantized by the linear prediction coefficient (LPC coefficient) and the gain. In the above description, the input switching unit 103, the pitch code switching unit 110, the frame thinning unit 133, the LSP
The code switching unit 146, the gain code switching unit 151, and the like have been described as switching on a frame-by-frame basis, but are not limited to switching on a frame-by-frame basis. Can be easily realized by changing necessary parts.

[0044]

As described above, according to the speech encoding parameter encoding method and apparatus of the present invention, the speech pitch, the V / UV information of each spectrum band, and the harmonic spectrum amplitude sequence are obtained for each speech frame. In an analysis-synthesis-type speech coding method represented by speech coding parameters consisting of, a speech pitch is encoded by switching between a frame to be differentially or uniformly quantized as a logarithmic pitch and a frame to be quantized by an inter-frame interpolation index. Thus, the number of coded bits can be greatly reduced.
In addition, V / UV information is coded with an index number of a representative V / UV value prepared in advance for each band number determined by the range of the audio fundamental frequency, so that the number of V / UV coded bits can be reduced rationally. I can do it. Further, the V / UV information is thinned out for each frame, and the V / UV information of the frame that does not transmit the V / UV information can be further reduced in the number of V / UV encoded bits by a method of inferring from the preceding and succeeding frames. .

The voice harmonic spectrum amplitude sequence is 0
After correcting the next harmonic amplitude value to a value that is easy to model, it is modeled by an autoregressive linear prediction model, and is expressed by the linear prediction coefficient and gain, which are the modeled coefficients.
The linear prediction coefficient can be encoded with a small number of bits by quantizing with a frame number to be transmitted after vector quantization and an already quantized linear prediction coefficient with an index number of an interpolation candidate having the least error by inter-frame interpolation. .
On the other hand, the gain is obtained by restoring the harmonic spectrum amplitude sequence from the quantized linear prediction coefficients, correcting it by the rate of change of the frame power by modeling and quantization of the linear prediction coefficients, and then uniformly quantizing the logarithmic gain. Quantization is performed using a combination of a frame to be quantized and a frame quantized with a smaller error from differential quantization from the previous frame or interpolation quantization from the previous and next frames. By these,
The audio harmonic spectrum amplitude sequence can be encoded with low bits while suppressing a change in the audio level of the frame due to the quantization of the harmonic spectrum amplitude sequence. As described above, according to the present invention, it is possible to provide a method and an apparatus for greatly reducing an encoding bit rate in an analysis and synthesis type speech encoding method and apparatus. Further, in the analysis-synthesis-type speech encoding method and apparatus, the analysis-synthesis-type speech encoding method of improving the encoded speech quality by shortening the frame update period of the speech encoding and preventing an increase in the number of encoded bits is improved. An audio encoding method and apparatus can be provided.

[Brief description of the drawings]

FIG. 1 is a block diagram of a speech coding parameter coding apparatus to which a speech coding parameter coding method of the present invention is applied.

FIG. 2 is a diagram for describing creation of an interpolation pitch candidate.

FIG. 3 is a diagram for explaining a method of deleting voiced / unvoiced information having a low frequency of occurrence.

FIG. 4 is a flowchart of a speech pitch encoding process.

FIG. 5 is a flowchart of encoding processing of voiced / unvoiced information.

FIG. 6 is a flowchart of an encoding process of a harmonic spectrum amplitude sequence.

FIG. 7 is a block diagram illustrating a configuration example of a speech decoding unit and a speech synthesis unit that decode coded speech encoded according to the present invention.

FIG. 8 is a configuration diagram of a general speech coded transmission device.

FIG. 9 is a block diagram of a speech coding parameter extraction unit.

FIG. 10 is a block diagram of a parameter encoding unit in a conventional speech encoding device.

[Explanation of symbols]

102 logarithmic conversion unit, 103 input switching unit, 105 uniform quantization unit, 108 pitch comparison unit, 110 pitch code switching unit, 111 delay unit, 112 subtraction unit, 113 difference quantization unit, 114 addition unit, 117 interpolation pitch candidate creation Section, 119 interpolation point comparison section, 132 representative voiced / unvoiced information codebook, 133 frame thinning section, 13
4 voiced / unvoiced comparison unit, 141 spectrum correction unit, 1
42 linear prediction modeler, 143 LSP quantizer, 1
44 LSP codebook, 145 LSP interpolation unit, 1
46 LSP code switching unit, 147 gain correction unit, 14
8 gain quantization section, 149 gain difference quantization section, 1
50 gain interpolation quantization section, 151 gain code switching section, 152 error calculation section, 153 harmonic restoration section

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Seiji Sasaki 3-2 Hikarinooka, Yokosuka City, Kanagawa Prefecture F.R.P. 5D045 CC07 DA06 DA11 5J064 AA01 BA13 BA16 BB01 BB03 BB04 BB12 BC11 BC16 BD02

Claims (3)

[Claims] 1. A speech coding parameter coding method for coding a voice coding parameter obtained from a voice signal digitized and divided into frames of a predetermined time length, wherein a voice pitch as the voice coding parameter is provided. A frame whose quantization pitch is obtained by one of the quantization methods by selecting the difference quantization method or the uniform quantization method, and an interpolation selected from a plurality of interpolation pitch candidates calculated using the quantization pitches of the preceding and succeeding frames. Coding by a combination with a frame to be quantized by a pitch index; voiced / unvoiced information as the voice coding parameter;
Encoding by an index of representative voiced / unvoiced information selected from a limited number of representative voiced / unvoiced information; and extracting a harmonic spectrum amplitude sequence as the voice coding parameter from a linear prediction coefficient by a linear prediction model or derived therefrom. And the line spectrum pair and the gain, and for the linear prediction coefficient or line spectrum pair,
A frame to be quantized by a quantizer such as a vector quantizer and an interpolated quantum by an index of a candidate point selected from a plurality of candidate points obtained by a linear interpolator from quantized linear prediction coefficients of a preceding and succeeding frame or a line spectrum pair. Coding in combination with a frame to be converted, correcting the gain according to a change in harmonic spectral power of the frame generated by quantization of the linear prediction coefficient, and obtaining a correction gain, and logarithmically calculating the correction gain. A frame to be uniformly quantized by the first gain quantizer as it is, an output value of the difference quantizer based on the quantization gain of the first gain quantizer of the previous frame, and a preceding and succeeding frame. From the output value of the interpolation quantizer obtained by selecting a plurality of interpolation candidates of the output of the first quantizer, Speech coding parameter coding method characterized by comprising the step of encoding the correction gain by the combination of the frame to be quantized by in favor less second gain quantizer for quantizing. 2. The method according to claim 1, wherein the step of obtaining the correction gain comprises using a linear prediction model using a harmonic spectral power obtained by a sum of squares of a harmonic amplitude sequence before the linear prediction model, a quantized linear prediction coefficient, and a gain before the quantization. 2. A speech coding parameter code according to claim 1, wherein a correction gain is calculated by multiplying the gain by a ratio of a harmonic spectrum power obtained from a sum of squares of a harmonic spectrum amplitude value obtained by the following. Method. 3. A voice coding parameter coding device for coding voice coding parameters obtained from a voice signal digitized and divided into frames of a predetermined time length, wherein a voice pitch as the voice coding parameter is provided. Of a frame to be quantized by one of the quantization methods by selecting the difference quantization method and the uniform quantization method, and an interpolation pitch selected from a plurality of interpolation pitch candidates calculated using the quantization pitches of the preceding and succeeding frames. Means for encoding by a combination of frames to be quantized by an index; and voiced / unvoiced information as the speech encoding parameter,
Means for encoding with an index of representative voiced / unvoiced information selected from a limited number of representative voiced / unvoiced information; and a step of converting a harmonic spectrum amplitude sequence as the voice coding parameter into a linear prediction coefficient by a linear prediction model or Separated into the line spectrum pair and the gain to be derived, and for the linear prediction coefficient or line spectrum pair,
A frame to be quantized by a quantizer such as a vector quantizer and an interpolated quantum by an index of a candidate point selected from a plurality of candidate points obtained by a linear interpolator from quantized linear prediction coefficients of a preceding and succeeding frame or a line spectrum pair. Means for encoding with a combination of frames to be converted, means for correcting the gain in accordance with a change in the harmonic spectral power of the frame generated by quantization of the linear prediction coefficient, and means for obtaining a correction gain, and logarithmizing the correction gain , The frame to be uniformly quantized by the first gain quantizer as it is, the output value of the difference quantizer based on the quantization gain of the first gain quantizer of the previous frame, From the output values of the interpolation quantizer obtained by selecting a plurality of interpolation candidates of the output of the first quantizer, Nde speech coding parameter encoding device characterized by having a means for encoding the correction gain by a combination of a frame to be quantized by the second gain quantizer for quantizing.