JP2853824B2 - Speech parameter information coding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high-efficiency speech encoding method for digitally encoding a speech signal sequence with a small amount of information, and more particularly to a speech parameter information encoding method for spectral envelope and power parameters.

[0002]

2. Description of the Related Art In digital mobile radio communications and voice storage services, various high-efficiency voice coding methods are used in order to use radio waves and storage media efficiently. As a method of encoding the sound of 8 kHz samples at 8 kbit / s or less, CELP, VSELP, multi-pulse encoding, transform encoding by weighted vector quantization, and the like are known. In either method, spectrum envelope information and power information are encoded and transmitted for each frame.

[0003]

By the way, in the above-mentioned conventional coding, in order to increase the information compression efficiency, a method of extending a frame update period, a difference from a past frame, a prediction value, and the like. Is known. However, if the frame update cycle is 40 m
Above s, a problem arises in that it is not possible to follow changes in the spectral characteristics and power of the speech waveform, and the coding distortion increases. Further, if the parameters are destroyed due to a code error, distortion occurs in the coded speech in a long section. On the other hand, when the difference between the parameter of the past frame and the difference between the predicted value and the predicted value are encoded, the information can be compressed by utilizing the temporal continuity of the parameter even in a short update cycle, but the influence of the past code error is over a long time. It has the disadvantage of propagating.

The present invention has been made in view of the above-mentioned circumstances, and when encoding spectral information and power with a small amount of information, avoids the propagation of code errors, maintains a relatively short update cycle, and maintains the parameter update. An object of the present invention is to provide a speech parameter information coding method capable of reducing quantization distortion by using temporal continuity.

[0005]

In order to solve the above-mentioned problems, according to the first aspect of the present invention, a speech is divided into frames of a predetermined number of samples, and the shape of a spectral envelope and a parameter of a power are divided in frame units. In the parameter information encoding method of extracting a parameter set and quantizing a set of parameters, the encoder uses a past vector, that is, a part of output vectors used when quantizing a past frame, and a current vector, that is, a code. The weighted averaging operation of the two reproduced vectors in the book is performed, and the criterion that minimizes the distortion between the averaged vector and the ideal parameter vector obtained by the analysis is used. The output vector is selected, its number is transmitted, and the decoder adds the past vector and the current reproduction vector corresponding to the transmitted code. And outputs an average.

Further, according to the invention described in claim 2, according to claim 1,
In the described speech parameter information encoding method, when a code error is expected in the transmission path, the output vector of the codebook of the immediately preceding frame is transformed into a past vector. According to a third aspect of the present invention, in the speech parameter information encoding method according to the first aspect,
The ratio of the weighted average is set to a unique value for each group of the codebook vectors of the current frame, for each individual vector, or for each element of the vector.

Further, according to the invention described in claim 4, according to claim 1,
In the described speech parameter information encoding method, the ratio of a plurality of types of weighted average is set in advance, and in the encoder,
In order to reduce the quantization distortion, a ratio and a codebook vector are selected in combination, and a code specifying the ratio and a code specifying the vector are transmitted in combination. And a weighted average of the current vector and the current vector.

[0008] According to the fifth aspect of the present invention, in the first aspect,
In the described speech parameter information coding method, when the decoder detects that there is an error in the transmission information of the past frame, the decoder reduces the ratio of the past vector, takes a weighted average, and takes the weight of the current frame. When it is detected that there is an error in the transmission information, a weighted average is obtained by reducing the ratio of the current vector.

[0009]

According to the first aspect of the present invention, in the encoder, the past vector, that is, a part of the output vector used in the quantization of the past frame, and the current vector, that is, the reproduction in the codebook. Perform a weighted average operation of the two vectors of the vector and select an output vector in the codebook using a criterion that minimizes the distortion between the averaged vector and the ideal parameter vector obtained by the analysis. , And the decoder outputs a weighted average of the past vector and the current reproduction vector corresponding to the transmitted code.

According to the second aspect of the present invention, in the speech parameter information encoding method according to the first aspect, if a code error is expected in a transmission path, the codebook of the immediately preceding frame is used. Is transformed into a past vector.

According to a third aspect of the present invention, in the speech parameter information encoding method according to the first aspect, the ratio of the weighted average is determined for each group of the codebook vectors of the current frame or for each vector. A unique value is set for each element or for each element of the vector.

According to a fourth aspect of the present invention, in the speech parameter information encoding method according to the first aspect, a ratio of a plurality of types of weighted averages is set in advance, and the encoder performs quantization distortion. In order to reduce the ratio, select a combination of the ratio and the codebook vector, and transmit by combining the code specifying the ratio and the code specifying the vector,
The decoder performs a weighted average of the past vector and the current vector at a specified ratio.

According to a fifth aspect of the present invention, in the speech parameter information encoding method according to the first aspect, when the decoder detects that there is an error in the transmission information of the past frame, The weighted average is calculated by reducing the ratio of the past vectors, and when it is detected that there is an error in the transmission information of the current frame, the weighted average is calculated by reducing the ratio of the current vector.

[0014]

Next, an embodiment of the present invention will be described with reference to the drawings. In the present invention, a value or a vector of a spectral parameter (hereinafter, simply referred to as a vector) is represented by a weighted average vector of a plurality of vectors in the past and present frames. In normal differential coding or predictive coding, the output vector of a past frame is used as it is, but in the present invention, only a part of the output vector is updated in the previous frame among vectors created by weighted averaging, for example. It is characterized in that only a vector that has been affected by a code error is used, or a vector that is deformed assuming a code error is used. Another feature is that the ratio of the weighted average is selected and controlled.

FIG. 1 is a block diagram showing the configuration of a first embodiment in which the present invention is applied to vector quantization of spectral parameters. In the figure, a multiplier 1 multiplies a past code vector V _j one frame before by a coefficient (1-g), and supplies this to one input terminal of an adder 2. Note that the symbol g is a constant of the ratio of the weighted average. Further, the multiplier 3 multiplies the current reproduction vector W _k supplied via the changeover switch SW 1 by a constant g of the ratio of the weighted average, and supplies this to the other input terminal of the adder 2. ing. The adder 2 adds the respective supplied vector to the input, and supplies to the distortion calculator 4 as the quantization value Y _n. The distortion calculator 4 has n as an input parameter.
The spectral parameter vector X _n analyzed in the number of frames is supplied, and a reproduction vector W _k that minimizes the distortion between the previous quantized value Y _n and the spectral parameter vector X _n is selected by the changeover switch SW1. , And the number is transmitted as a symbol S1.

Therefore, in this embodiment, the spectrum parameter vector X analyzed in a certain n-th frame
quantized value Y _n of _n is represented as _{Y n = (1-g)} V j + gW k. Note that the current reproduction vector W _k is used as a past code vector V _j in the next frame.

Next, FIG. 2 shows a vector U two frames past.
This is an example in which three types of codebooks U, V, and W are used depending on the frame, although they are repeated in a cycle of three frames. Y _n = (U _i + V _j + W _k ) / 3 where U _i is an output vector determined in two previous frames, and V _j is a code vector determined in one previous frame. As the reproduction vector W _k in the n-th frame, a vector that minimizes the distortion between the quantized value Y _n and the spectrum parameter vector X _n is selected from the codebook W and transmitted. The output vector V _j is used in the next frame, and the reproduction vector W _k is used in the next frame and the next next frame. The reproduction vector W _k can be determined in the n-th frame. However, if delay is allowed, the quantization distortion can be reduced by considering the next frame or the parameter X two frames ahead. it can.

Next, FIG. 3 shows a state of encoding in the third embodiment. The quantized value Y _n is represented by Y _n = (Ui + Vj + W _k + Z _n ) / 4. In this case, in the n-th frame, W _n and Z _n are determined and the code is transmitted. In the above example, the codebooks U, V, and W are respectively different codebooks.
It is also possible to share.

In the first, second, and third embodiments described above,
First, a method of determining an ideal X _n and determining the quantization value Y _n by averaging in the parameter dimension was used. In the case of a linear prediction parameter, an inverse filter using a parameter determined in a past frame was used. There is also a method of determining the parameters of the current frame by analyzing the residual signal. In the averaging method in the parameter dimension, the average of the coefficients of the polynomials of the individual synthesis filters is the final coefficient of the synthesis filter. On the other hand, in the latter method of performing the analysis a plurality of times, the product of the individual polynomials becomes the final polynomial of the synthesis filter.

The invention according to claim 2 is a method for suppressing an increase in distortion particularly due to a transmission line code error. In this case, in FIG. 1, instead of the quantized value Y _n as the output vector, the distortion between the expected value Y _n ^* and the spectrum parameter vector X _{n as} the input parameter in the decoder considering the bit error rate is minimized. (Not shown). The expected value is the bit error rate in the transmission path (bit error rate) epsilon, it can be approximated when the number of transmission bits for one vector and m, and ^{_{Y n * = (1-mε}} ) Y n + ΣεY e. However, the second term on the right side indicates the sum of m types of vectors Y _e output when only one bit out of m transmission path codes corresponding to the vector is incorrect.

FIG. 4 shows a fourth embodiment corresponding to the third aspect of the present invention.
FIG. 3 is a block diagram showing a configuration of an embodiment of the present invention. In this embodiment, the ratio g of the weighted average is not constant, and the ratio g _k set according to the reproduction vector W _K in the codebook is used. ing. In the figure, a reproduction vector W _K is a multiplier M ₁ , in which individual ratios g _k (k = 0 to an integer of n) are set.
The changeover switch SW2 through M ₂ ,..., M _n−1 , M _n
Supplied to Further, the distortion calculator 4 selects the reproduction vector W _k by the changeover switch SW2 and selects the ratio g _k according to it so that the distortion between the quantization value Y _n and the spectrum parameter vector X _n is minimized. It is supposed to. Further, the ratio supplied to the multiplier 1 changes in conjunction with the switching of the switch SW2. Quantized value Y _n is expressed by _{Y n = (1-g k} ) V j + g k W k. In this equation, the ratio g _k is a scalar value corresponding to the reproduction vector. However, a plurality of reproduction vectors may be grouped and the scalar value may correspond to each group. Conversely, a ratio value may be set for each element of the reproduction vector. In each case, the past code vector V _j is assumed, and an optimal combination of the ratio with the past vector and the reproduction vector is selected in the sense of minimizing the distortion between the quantized value Y _n and the spectrum parameter vector X _n. I do.

FIG. 5 shows a fifth embodiment corresponding to the fourth aspect of the present invention.
FIG. 3 is a block diagram showing a configuration of an embodiment of the present invention, which is characterized in that values of ratios of a plurality of types of weighted averages are set independently of a reproduction vector. In this figure, the code vector V _j one frame before is determined by the multipliers MA ₁ , M
It is supplied to the input of A _2. The multipliers MA ₁ and MA ₂ include:
(1-g ₁ ) and (1-g ₂ ) are set as coefficients, respectively, and the outputs thereof are changeover switches SW3, SW
4. Changeover switch SW3 is adapted to be switched in response to the distortion calculation result by the distortion calculator 4, selects one of the output of the multiplier M _1, M _2, fed to one input terminal of the adder 2 I do. Further, the reproduction vector W _K is supplied to the changeover switch SW4. The changeover switch SW4 is, like the switch SW3,
Switching is performed in accordance with the result of the distortion calculation by the distortion calculator 4, and the selected reproduction vector W _K is multiplied by the multiplier M
A ₃ and supply to the input terminal of MA4. The multipliers MA ₃ and MA ₄
Have g ₁ and g 2 set as coefficients, respectively.
The output is supplied to a changeover switch SW5. The changeover switch SW5 is connected to the switches SW3 and S
As in the case of W4, switching is performed according to the result of distortion calculation by the distortion calculator 4, and the selected multipliers MA3, MA3,
The output of MA ₄ is supplied to the other input terminal of the adder 2. In such a configuration, the quantization value Y _n
Is expressed by _{Y n = (1-g i} ) V i + g i W k. Also in this case, the distortion is minimized by a combination of the reproduction vector and the ratio, but it is necessary to separately transmit selection information S2 indicating which ratio is to be used.

[0023] In the present invention of claim 5 is a method to reduce the influence when a code error is detected in the decoder, the quantized value _{Y n, Y n = (1} -g v) V i + g v It is represented by W _k . The encoder obtains W _k that minimizes the distortion as in the case of FIG. 1, but a code error is detected in the vector V _j of the past frame or the vector W _k of the current frame in the decoder, or When it is estimated that the probability of a code error is high, only the decoder side reduces the weighted average ratio of the vector containing the error to obtain an output vector (not shown). In this modification, for example, when a transmission path error is detected in the immediately preceding frame, the quantization value is set to Y _n = W _k without using information of the previous frame at all. Alternatively, in the middle, the quantized values, so that to reduce the effect of the previous frame as _{Y n = (1-√g v} ) V i + √g v W k.

[0024]

As described above, according to the present invention, since only the index corresponding to one codebook is transmitted for each frame, the amount of transmission information can be reduced even if the frame length is shortened. When the continuity with the past frame is high, the quantization distortion can be reduced. On the other hand, even if a code error occurs, the influence is small because it is averaged with past frames. A code error in the current frame affects only the next two frames. Further, if a code error can be detected using a redundant code, an advantage can be obtained that the influence of the error can be reduced by averaging while excluding erroneous parameters.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a first exemplary embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a second exemplary embodiment of the present invention.

FIG. 3 is a block diagram illustrating a configuration of a third exemplary embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a fourth embodiment (Claim 3) of the present invention.

FIG. 5 is a block diagram showing a configuration of a fifth embodiment (Claim 4) of the present invention.

[Explanation of symbols]

1,3 multiplier 2 adder 4 distortion calculator V _j code vector (output vector) g ratio of weighted average SW1 switch W _k reproduction vector Y _n quantized value X _n spectrum parameter vector U, V, W codebook Y _n ^* expected value ε bit error rate (bit error rate) M ₁ , M ₂ ,..., M _n−1 , M _n multipliers MA ₁ , MA ₂ , MA ₃ , MA ₄ multipliers

──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Satoshi Miki 1-6-1, Uchisaiwaicho, Chiyoda-ku, Tokyo Nippon Telegraph and Telephone Corporation (72) Inventor Naka Naka, 1-16-1 Uchisaiwaicho, Chiyoda-ku, Tokyo Nippon Telegraph and Telephone Corporation (56) References JP-A-1-68024 (JP, A) JP-A-5-165499 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) G10L 3/00-9/18 H03M 7/30 JICST file (JOIS)

Claims (5)

(57) [Claims] In a parameter information encoding method for dividing a speech into frames of a fixed number of samples, extracting a parameter of a shape and a power of a spectral envelope in frame units, and quantizing a set of parameters, an encoder includes: A weighted averaging operation is performed on the past vector, that is, a part of the output vectors used in the quantization of the past frame and the current vector, that is, the reproduced vector in the codebook. The output vector in the codebook is selected using the criterion that minimizes the distortion from the ideal parameter vector obtained in step 2, and its number is transmitted. And outputting a weighted average of a current reproduction vector corresponding to the encoded code and output. 2. A speech signal according to claim 1, wherein when a code error is expected in the transmission path, the output vector of the codebook of the immediately preceding frame is transformed into a past vector. Parameter information encoding method. 3. The ratio of the weighted average is determined for each group of codebook vectors in the current frame, or for each individual vector,
2. A method according to claim 1, wherein a unique value is set for each vector element. 4. A ratio of a plurality of types of weighted averages is set in advance, and an encoder selects a combination of a ratio and a codebook vector so as to reduce quantization distortion, and selects a code and a vector for specifying the ratio. 2. A speech parameter information code according to claim 1, wherein the code is transmitted in combination with a code designating the parameter, and the decoder performs a weighted average of the past vector and the current vector at a specified ratio. Chemical method. 5. When the decoder detects that there is an error in the transmission information of the past frame, the decoder lowers the ratio of the past vector and takes a weighted average, and there is an error in the transmission information of the current frame. 2. A speech parameter information encoding method according to claim 1, wherein when the fact is detected, the ratio of the current vector is reduced and a weighted average is calculated.