JPH0683393A - Speech encoding device - Google Patents

Abstract

PURPOSE:To obtain the speech encoding device which can make a linear prediction error smaller. CONSTITUTION:This device is equipped with a linear predictive analytic part 2 which has a synthesized speech code book consisting of synthesized speech resynthesized on the basis of past input speech compares an input speech with respective elements in the synthesized speech code book, and finds a linear prediction coefficient for a difference speech obtained by subtracting the element having the highest correlation from the input speech and an AbS(Analysis by Synthesis) processing part 10 which minimizes the linear prediction error based upon the difference speech by using a code book for excitation consisting of exciting vectors.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coding apparatus using an analysis-synthesis coding method, among speech coding methods for efficiently coding human speech.

[0002]

2. Description of the Related Art As a typical example of an analysis-synthesis coding method, CELP (Cod (Cod) using linear prediction analysis and a codebook for excitation is used.
e-Excited Linear Prediction: Residual-driven linear prediction)
There is encoding. FIG. 6 shows a block diagram of a speech coding apparatus using this coding method. In the figure, the input voice input to the voice input unit 1 is supplied to the linear prediction analysis unit 2 and the linear prediction coefficient α is obtained. Then, the scalar-quantized α in the linear prediction coefficient quantization unit 3 is supplied to the linear predictor 4. The index i _e of the excitation vector from the excitation codebook 5 is simultaneously input to the linear predictor 4,
As a result, the linear predictor 4 outputs the linear predictive speech x _v . In the differentiator 8, the difference between the input speech x and its linear prediction speech x _v is calculated, and the prediction error e is obtained. The prediction error e is supplied to the error minimization unit 7 through the auditory weighting filter 6 in order to reduce the perceptual noise. The error minimization unit 7 obtains the mean square error of the prediction error e, and holds the minimum mean square error and the index i _e of the excitation vector at that time. Then, after performing the above processing for all the excitation vectors in the excitation codebook 5, the quantized linear prediction coefficient α and the excitation vector index i _e are transmitted to the speech decoding apparatus 30.

[0003]

However, in the conventional speech coding apparatus described above, the linear prediction error can be sufficiently reduced even if the adaptive codebook using the correlation between adjacent frames of the linear prediction residual is used. There wasn't.

The speech coding apparatus of the present invention has been made in view of the above problems, and its purpose is to determine the re-synthesized synthesized speech itself and the inter-adjacent frame correlation of its linear prediction coefficient. It is an object of the present invention to provide a speech coding apparatus that can reduce the linear prediction error and can reduce the bit rate of the code amount by utilizing the above.

[0005]

In order to achieve the above object, the speech coding apparatus of the present invention has a synthesized speech codebook composed of synthesized speech resynthesized based on past input speech. Excitation code consisting of excitation vector, which compares each element of the synthetic speech codebook with the input speech and finds the linear prediction coefficient for the difference speech obtained by subtracting the element with the highest correlation from the input speech Ab for minimizing linear prediction error with differential speech using a book
And an S processing unit.

[0006]

In other words, in the present invention, the synthesized voice re-synthesized based on the past input voice and the input voice are compared,
A linear prediction coefficient is obtained for the differential speech by subtracting the element with the highest correlation from the input speech, and the linear prediction error with the differential speech is minimized.

[0007]

FIG. 1 shows a conceptual diagram of an embodiment of the present invention. Also in this embodiment, similar to the conventional speech coding apparatus, the linear prediction analysis unit 2, the excitation codebook 5, and the AbS (Anal) are used.
ysis By Synthesis: Analysis by synthesis) Although the processing unit 10 is included, the present embodiment further includes the synthetic voice codebook 9 created based on the past synthetic voice.

The input voice supplied to the voice input unit 1 is compared with each element of the synthetic voice codebook 9 and the difference with the element having the highest correlation is obtained in the difference calculator 8. The obtained differential speech is analyzed by the linear prediction analysis unit 2 to obtain the linear prediction coefficient α.

The AbS processing unit 10 is a processing unit including a linear predictor 4, a difference unit 8, an auditory weighting filter 6, and an error minimizing unit 7 in FIG. The excitation vector from Codebook 5 is used to search for the optimal excitation vector. A synthesized speech obtained by adding the synthesized speech synthesized by the optimum excitation vector and the element of the synthesized speech codebook 9 having the highest correlation with the calculated input speech is added as a new element of the synthesized speech codebook 9. , Remove the oldest element in time. From the encoding device to the speech decoding device 30, in addition to the conventional linear prediction coefficient α and the index i _e of the excitation codebook, the index i of the synthetic speech codebook is used.
You need to send _{x '} . The above process is performed for each voice section obtained by dividing the input voice at regular intervals.

FIG. 2 shows the configuration of the first embodiment of the present invention.
1 to 9 and 30 in FIG. 2 correspond to the same numbers in FIGS. 1 and 6, and 11 is a synthetic speech codebook that receives an error between an input speech and an element of the synthetic speech codebook 9 and minimizes the error. An error minimization unit A having a function of controlling so as to search for an index of 9, and 12 is a difference voice x ″.
And the prediction speech x _v ″ of x ″ which is the output from the linear predictor 4.
Is an error minimization unit B having a function of receiving an error between the two signals and controlling to search an index of the excitation codebook 5 that minimizes the error, and 13 is an adder for adding voice signals to each other. The operation of the speech encoding apparatus (shown by 20 in FIG. 3) having the above configuration will be specifically described.

First, the input voice sampled at a constant time interval and supplied to the voice input unit 1 is compared and subtracted with one element of a synthesized voice codebook 9 consisting of past synthesized voices by a differentiator 8, and its The result is sent as an error to the error minimizing unit A11. The error minimization unit A11 operates the index of the synthetic speech codebook 9 to find the difference speech x ″ having the smallest error and the index ix _′ of the synthetic speech codebook 9 at that time. The obtained difference speech x ″ is It is supplied to the linear prediction analysis unit 2 and the linear prediction coefficient α is calculated for the analysis order (usually, the analysis order is about 10) using the well-known technique such as the covariance method or the autocorrelation method. The linear predictor 4 uses the linear prediction coefficient α and the excitation vector which is an element of the excitation codebook 5 to predict the predicted speech x _v ″.
Is calculated and output.

Next, in the error minimization section B12, the difference calculator 21
The error between the differential speech x ″ obtained in step S1 and the predicted speech x _v ″ is received, and the predicted speech x _v ″ with the smallest error is manipulated while operating the index of the excitation codebook 5.
And index i _e of the excitation codebook 5 at that time
Ask for. Then, the linear prediction coefficient α, the index i _e of the excitation codebook 5 and the index i _{x ′} of the synthetic speech codebook 5 are sent to the speech decoding device 30. Further, the predicted speech x _v ″ with the smallest error is added in the adder 13 to the element indicated by the index i _{x ′} of the synthesized speech codebook 9 obtained by the error minimization unit A 11, and the resultant synthesized speech x ″ is Registered as some elements of the synthetic voice codebook 9.

FIG. 3 shows the configuration of the decoding apparatus according to the first embodiment. From the speech coding apparatus 20, the linear prediction coefficient α, the index i _e of the excitation codebook 5 and the index i _{x ′} of the synthetic speech codebook 9 are sent. First, the element of the excitation codebook 5 at the position indicated by the index i _e is supplied to the linear predictor 4, and the linear predictor 4 synthesizes the predicted speech x _v ″ with the differential speech x ″ by the linear predictor 4. It The index i _{x 'element} x of synthesized speech codebook 9 position indicated by' is taken out at the adder 13, the sum of the x 'and x _v "taken,
It is output as a synthesized voice x _v . This synthesized speech x _v is registered as some elements of the synthesized speech codebook 9 similarly to the speech coder.

Next, FIG. 4 shows an example of the encoding apparatus of the second embodiment.
Shown in. This example is different as the first embodiment was using synthesized speech x _v directly synthesized speech x _v again, and linear prediction analysis, obtains a linear predictive coefficient alpha, use this as the code book. In FIG. 4, reference numeral 15 denotes the codebook, which is a linear prediction coefficient (LP for the synthesized speech x _v of the past input speech).
It is a synthetic speech LPC codebook that stores C coefficients). Again, the specific operation will be explained step by step.

First, the input speech sampled at a constant time interval and supplied to the speech input section 1 is input to the linear prediction analysis section 2 and the linear prediction coefficient α is obtained. This linear prediction coefficient α is compared with one element of the synthetic speech LPC codebook 15, and the result is sent to the error minimization unit A11 as an error. The error minimization unit A11 operates the index of the synthetic speech LPC codebook 15 while operating the index iα ′ of the synthetic speech LPC codebook 15 having the minimum error.
Ask for. The linear predictor 4 calculates and outputs the predicted speech x _v using the element indicated by the index iα ′ (linear prediction coefficient α ′) and the excitation vector which is the element of the excitation codebook 5.

Next, the error minimization unit B12 receives the error between the input speech x obtained by the differencer 21 and the predicted speech x _v , operates the index of the excitation codebook 5 and predicts the speech with the minimum error. x _v and the index i _e of the excitation codebook 5 at that time are obtained. Then, to the speech decoding device 30, the index i α ′ of the synthetic speech LPC codebook 15 and the excitation codebook 5 are transmitted.
The index i _{e of} is sent. Also, the predicted speech x _v with the smallest error is sent from the error minimization unit B12 to the linear prediction analysis unit 2, converted into the linear prediction coefficient α ″ again, and newly registered as an element of the synthetic speech LPC codebook 15. .

FIG. 5 shows an example of the decoding device of the second embodiment. From the speech coding apparatus 20, the index iα 'of the synthetic speech LPC codebook 15 and the index i _e of the excitation codebook 5 are sent. First, the element α ′ of the synthetic speech LPC codebook 15 indicated by the index iα ′ and the excitation codebook 5 indicated by the index i _e.
Is supplied to the linear predictor 4, and the synthesized speech x _v is calculated and output. Further, the linear prediction analysis unit 2 obtains the linear prediction coefficient α ″ again for the synthesized speech x _v , and the synthesized speech LPC is obtained in the same manner as the speech coder.
It is registered as an element of the codebook 15. Since this embodiment corresponds to adaptive vector quantization of linear prediction coefficients,
Quantization efficiency is higher than that of conventional scalar quantization, and the linear prediction coefficient is only inside the device (that is, the linear prediction coefficient itself is not transmitted), so it has sufficient analysis order and quantization accuracy. It is possible.

[0018]

As described above, the speech coding apparatus of the present invention utilizes the correlation (similarity) with the past synthesized speech, which has not been used conventionally, to improve the sound quality and reduce the bit rate. Can be realized.

[Brief description of drawings]

FIG. 1 is a conceptual diagram of an embodiment of a speech encoding apparatus according to the present invention.

FIG. 2 is a diagram showing the configuration of a speech encoding apparatus according to the first embodiment of the present invention.

FIG. 3 is a diagram showing a configuration of a decoding device according to a first embodiment.

FIG. 4 is a diagram showing a configuration of a speech encoding apparatus according to a second embodiment.

FIG. 5 is a diagram showing a configuration of a decoding device according to a second embodiment.

FIG. 6 is a diagram showing a configuration of a conventional speech encoding device.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Speech input part, 2 ... Linear prediction analysis part, 5 ... Excitation codebook, 9 ... Synthetic speech codebook, 10 ... AbS processing part, 30 ... Speech decoding device.

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] April 2, 1993

[Procedure Amendment 1]

[Document name to be amended] Statement

[Correction target item name] 0012

[Correction method] Change

[Correction content]

Next, in the error minimization section B12, the difference calculator 21
The error between the difference speech x ″ obtained in step S1 and the predicted speech x _v ″ is received, and the predicted speech x _v ″ with the smallest error is manipulated while operating the index of the excitation codebook 5.
And index i _e of the excitation codebook 5 at that time
Ask for. Then, the linear prediction coefficient α, the index i _e of the excitation codebook 5 and the index i _{x ′} of the synthetic speech codebook 5 are sent to the speech decoding device 30. Further, the predicted speech x _v ″ with the smallest error is added in the adder 13 to the element indicated by the index i _{x ′} of the synthesized speech codebook 9 obtained by the error minimization unit A 11, and the resultant synthesized speech x _v is Registered as some elements of the synthetic voice codebook 9.

Claims (1)

[Claims] 1. A synthetic speech codebook composed of synthetic speech resynthesized based on past input speech, each element of this synthetic speech codebook is compared with the input speech, and the element having the highest correlation is input speech. A linear prediction analysis unit that obtains a linear prediction coefficient for the differential speech subtracted from, and an AbS processing unit that minimizes a linear prediction error with the differential speech by using an excitation codebook composed of excitation vectors. A speech coding apparatus characterized by: