JP3290444B2 - Backward code excitation linear predictive decoder - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a backward type code excitation linear predictive decoder.

[0002]

2. Description of the Related Art FIG. 2 is an explanatory diagram of a conventional backward code excitation linear predictive decoder.

The backward-type code-excited linear predictive encoder generates a vocal tract prediction coefficient and an excitation gain by backward adaptation and transmits only information on the excitation code, and is described in the following paper. ing.

Reference: JHChen, “High-quality 16 kb /
s speech coding with a one-waydelay less than 2 m
s, "Proc.IEEE Int.Conf.Acoust., Speech, Signal Proce
ssing, (April 1990). In FIG. 2, 101 is a synthesis filter, 102 is a vocal tract analysis circuit, 103 is an adder, 104 is a gain control circuit, 1
05 is a multiplier, 106 is a statistical codebook, 107 is an inverse quantizer, 108 is an input terminal of a received input code, 109
Is an output terminal of the synthesized speech vector.

[0005]

In the code-excited linear predictive decoder shown in FIG. 2, the entire system is composed of vocal tract prediction coefficients Ai.
Is calculated from the past synthesized speech vector. Therefore, when an error occurs in the synthesized speech vector, an error occurs in the vocal tract prediction coefficient, which has a drawback that the effect extends for a long time.

[0006]

SUMMARY OF THE INVENTION The present invention provides a backward-type code-excited linear predictive decoder which employs a relatively low-order impulse response approximation acyclic filter (hereinafter referred to as FI).
R filter) is separately provided in order to obtain a synthesized speech vector for calculating vocal tract prediction coefficients.

[0007]

Since a relatively low-order FIR filter for calculating vocal tract prediction coefficients is provided separately from the synthesis filter for speech decoding, the influence of errors in vocal tract prediction coefficients can be converged in a short period of time. It is possible to provide a backward type code excitation linear prediction decoder that is resistant to errors.

[0008]

FIG. 1 is a block diagram of an embodiment of a backward type code excitation linear prediction decoder according to the present invention.
0 is an FIR filter, 211 is an adder, 201-209
Has the same configuration as that in FIG. Incidentally, the synthesis filter 201 is used to recursive filter may be used LPC other linear predictor.

Referring to FIG. 1, a received input code C input from an input terminal 208 is converted into a statistical excitation code index I by an inverse quantizer 207 and transmitted to a statistical code book 206. Statistical code book 206
Outputs a statistical excitation code vector E corresponding to the index I. Gain control circuit 204 calculates an excitation gain G for the current frame from the past excitation vector e _x. Statistical excitation codevector E is excitation vectors e _x of the current frame is multiplied is excited gain G by the multiplier 205
Becomes

The synthesis filter 201 predicts the speech synthesis vector S1 of the current frame based on the speech synthesis vector S1 of the past frame.
The vocal tract prediction coefficient of 01 is updated every frame time according to the vocal tract prediction coefficient Ai given from the vocal tract analysis circuit 201.

[0011] The adder 203 adds the excitation vector e _x and output of the synthesis filter 201, the add vector and outputs from an output terminal 209 as a synthesized speech vector S1 of the current frame.

[0012] FIR filter 210 is to predict the speech synthesis vector S2 of the current frame based on the excitation vector e _x of the past frame, the FIR filter 21
The tap coefficient of 0 is updated for each frame time according to the impulse response Hi given from the vocal tract analysis circuit 202.

[0013] The vocal tract analysis circuit 202 detects the past synthesized speech S
2, the vocal tract prediction coefficient Ai for the current frame is calculated using a linear prediction analysis (for example, an autocorrelation method), and output to the synthesis filter 201. Further, the vocal tract analysis circuit 202 calculates the impulse response Hi of the synthesis filter 201 from the vocal tract prediction coefficient Ai just obtained, and outputs it to the FIR filter 210.

In the configuration of FIG. 1, the vocal tract analysis circuit 202
By setting the order of the FIR filter 210 to a relatively low order in accordance with the predetermined analysis accuracy of the vocal tract prediction coefficient Ai in, even if an error occurs between the encoders and the decoders, The effects of errors in the synthesized speech signal and the vocal tract prediction coefficients converge in a certain short period. Further, the synthesis filter 201 can perform prediction based on a synthesized voice signal for a sufficiently long period, as in the related art.

[0015]

According to the present invention, the synthesized speech signal for calculating the vocal tract prediction coefficients is obtained by the FIR type synthesis filter.
Even if an error occurs between the encoders and the decoders, the errors in the synthesized speech signal and the vocal tract prediction coefficients due to the error always converge in a certain short period. Therefore, it is possible to realize an error-resistant backward-type code-excitation linear prediction decoder.

[Brief description of the drawings]

FIG. 1 is a block diagram of a backward type code excitation linear prediction decoder showing an embodiment of the present invention.

FIG. 2 is an explanatory diagram of a conventional technique.

[Explanation of symbols]

 Reference Signs List 201 synthesis filter 202 vocal tract analysis circuit 203 adder 204 gain control circuit 205 multiplier 206 statistical codebook 207 inverse quantizer 210 FIR filter 211 adder

──────────────────────────────────────────────────続 き Continuation of front page (72) Inventor Yoshihiro Ariyama 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (56) Reference JP-A-2-231825 (JP, A) Kaihei 3-2900 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 19/00-19/14 H03M 7/30 H04B 14/04

Claims (1)

(57) [Claims] 1. A backward type code excitation linear prediction decoder for outputting a first addition vector of an excitation vector and an output of a synthesis filter as a synthesized speech vector of a current frame, wherein the excitation is based on a received input code. Means for creating a vector, and updating its own vocal tract prediction coefficient in accordance with the vocal tract prediction coefficient given from the vocal tract analysis circuit, wherein the current frame is updated based on a synthesized speech vector of a past frame. said synthesis filter to synthetic speech vector prediction to the output of, in accordance with the impulse response provided from the vocal tract analysis circuit,
A non-recursive filter that updates its own tap coefficient and predicts and outputs a synthesized speech vector of the current frame based on an excitation vector of a past frame, and the excitation vector and the acyclic filter A vocal tract analysis circuit that receives a second sum vector with the output as an input.
For example, the vocal tract content prayers circuit is for giving the composite Firutahe calculates the vocal tract prediction coefficient in the second adder vector, and the non-recursive calculate the impulse response at the second addition vectors A backward-type code-excited linear predictive decoder, which is provided to a filter.