JP3099844B2 - Audio encoding / decoding system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coding / decoding apparatus used for digitally transmitting or storing a speech signal, and more particularly to a speech signal which is separated into a sound source signal, an autoregressive coefficient and a moving average coefficient. The present invention relates to a speech coding / decoding system to be converted.

[0002]

2. Description of the Related Art A conventional speech coding / decoding system using a sound source signal, an autoregressive coefficient (hereinafter abbreviated as AR) and a moving average coefficient (hereinafter abbreviated as MA) is disclosed in the document "Speech using glottal excitation wave model". Examination of analytical synthesis method ”, Katsushi Seza, Tazaki
Hirohisa, Nakajima Kunio, Proceedings of the Acoustical Society of Japan, Fall Meeting, 1991-I-, 1-6-10, pp209-2
10, 1991. In this conventional method, a sound source wave model defined on a differential waveform of a glottal sound source wave is used as a sound source.

FIG. 2 is a block diagram showing a conventional speech coding / decoding system. First, the coding unit 100 shown in FIG. The excitation codebook 9 stores a plurality of excitation signals having one pitch period as excitation codewords.
The R codebook 13 stores a plurality of ARs (autoregressive coefficients) as AR codewords, and the MA codebook 6 stores a plurality of MAs (moving average coefficients) converted to spectral envelope parameters as MA codewords. It is. Pitch period extracting means 2
Extracts the pitch period 3 from the input audio signal 1 and outputs it. The sound source position detecting means 15 detects a peak position of the linear prediction residual signal of the input speech signal 1 for each pitch period, and outputs this as a sound source position 16. The sound source generation means 11
An excitation is generated using excitation codeword 10 output from excitation codebook 9 and pitch period 3 output from pitch period extraction means 2, and output as excitation signal 12. Synthesizing means 17
Is the sound source position 16 output from the sound source position detecting means 15
, An AR code word 14 output from the AR code book 13, an excitation signal 12 output from the excitation generating means 11, and a MA code word 8 output from the MA code book 6,
8 is generated. The optimal codeword selecting means 19 selects a combination of the AR codeword, the MA codeword, and the excitation codeword that minimizes the distance between the input speech signal 1 and the synthesized speech signal 18,
The AR codeword number 20, the excitation codeword number 21, and the MA codeword number 22 in the combination are output.

[0004] Next, a decoding unit 200 shown in FIG.
Will be described. The AR inverse quantization means 23 obtains an AR codeword 25 corresponding to the AR codeword number 20 from the encoding unit 100 from the AR codebook 24. MA inverse quantization means 31
Obtains the MA codeword 33 corresponding to the MA codeword number 22 from the encoding unit 100 from the MA codebook 32. Exciter inverse quantization means 26 similarly obtains, from excitation codebook 27, excitation codeword 28 corresponding to excitation codeword number 21 from encoding section 100. The sound source generating means 29 includes the encoding unit 1
An excitation signal 30 is generated from a pitch period 3 from 00 and an excitation codeword 28 from the excitation inverse quantization means 26. The synthesizing unit 38 receives the sound source signal 30 from the sound source generating unit 29 and A
The synthesized speech signal 39 is obtained from the AR code word 25 from the R inverse quantization means 23 and the MA code word 33 from the MA inverse quantization means 31.
Generate The encoding unit 100 and the decoding unit 200
, The AR codebook 13 and the AR codebook 24, the excitation codebook 9 and the excitation codebook 27, the MA codebook 6 and the MA codebook 32 are the same, respectively.

[0005]

In the case where an input speech signal is encoded using a one-pitch period sound source signal,
When the pitch period becomes longer, the MA, that is, the moving average coefficient works so as to compensate for the mismatch of the sound source signal, and the order of the MA becomes large. Therefore, although the optimal order of the MA differs depending on the input audio signal, the conventional audio coding / decoding system is configured as described above, and the quality of the synthesized audio is deteriorated because the order of the MA is fixed. There was a problem that.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and has as its object to improve the quality of synthesized speech by determining the order of MA in accordance with an input speech signal.

[0007]

According to the speech encoding / decoding method of the present invention, an encoding section is provided with a pitch loop of an input speech signal.
Order determining means for determining the order K of the moving average coefficient in accordance with the period and outputting it as the MA order, and MA conversion for converting the MA code word output from the MA codebook into the K-order MA and outputting it to the combining means Means for converting the MA codeword output from the MA inverse quantization means into the same MA order as the coding section, that is, K-order MA, and outputting to the synthesizing means. Things.

[0008]

In the present invention, the order determining means determines the order of the MA in accordance with the pitch period of the input speech signal, and the MA converting means converts the MA code word, which is a spectrum envelope parameter, to the order of the order determined by the order determining means. It converts to MA and the synthesizing means generates a synthesized speech signal using this MA.
The distance between the input speech signal and the synthesized speech signal in the encoding unit is reduced, and the quality of the synthesized speech is improved.

[0009]

【Example】

Embodiment 1 FIG. FIG. 1 is a block diagram showing an embodiment of a speech encoding / decoding system according to the present invention. In FIG. 1, the same parts as those in FIG. 2 are denoted by the same reference numerals, and description thereof will be omitted. First, in the encoding unit 100 shown in FIG. 1A, reference numeral 4 denotes an MA order K corresponding to the input speech signal 1 based on the pitch period 3 output from the pitch period extracting means 2, and the MA order The order determining means 7 for outputting as 5 and the MA converting means 7 for converting the MA code word 8 output from the MA codebook 6 into the MA of the MA degree 5 (K order) and outputting the same to the synthesizing means 17. Also, in the decoding unit 200 shown in FIG. 1B, 34 determines the order K of the MA according to the input speech signal 1 based on the pitch period 3 sent from the encoding unit 100, and 35
The degree determining means 36 outputs the MA code word 33 output from the MA inverse quantization means 31 as the MA degree 35
(Kth) MA that is converted to MA and output to combining means 38
Conversion means. These degree determining means 4, 34 and M which are newly provided in the encoding section 100 and the decoding section 200
The A conversion means 7 and 36 are similar.

The operation of the present system will be described below with reference to FIG. The description of the same parts as in the conventional example of FIG. 2 will be omitted. First, the encoding unit 100 shown in FIG. 1A will be described. The order determining means 4 determines the order of the MA to be, for example, 1/4 of the pitch period 3 of the input audio signal 1, and outputs the result as the MA order 5. It is assumed that, for example, a cepstrum coefficient is used as an MA codeword stored in the MA codebook 6. The MA conversion means 7 converts the MA in the MA codebook 6
The codeword (cepstrum) is converted into an MA of MA degree 5 using the following equation (1), and is converted as conversion MA8a by combining means 17:
Output to In the equation (1), C _n represents the n-th order of the cepstrum coefficient, and α _n represents the n-th order of the linear prediction coefficient.

[0011]

(Equation 1)

Next, the decoding unit 200 shown in FIG.
Will be described. The MA inverse quantization means 31 obtains the MA codeword corresponding to the MA codeword number 22 from the encoding unit 100 from the MA codebook 32 and outputs it as the MA codeword 33. The degree determining means 34 determines the degree of the MA at １ ／ of the pitch period 3 of the input speech signal 1 in the same manner as the encoding section 100, and outputs it as the MA degree 35. MA conversion means 36
Is the MA code word 3 output from the MA inverse quantization means 31
3 in the same manner as the encoding unit 100,
The signal is converted to an MA of degree 35 and output to the synthesizing means 38 as a converted MA 33a. As described above, the distance between the input speech signal and the synthesized speech signal in the encoding unit is reduced by determining the order of the MA in accordance with the pitch period of the input speech signal without fixing the order of the MA, and the quality of the synthesized speech is reduced. Is improved.

Embodiment 2 FIG. In the first embodiment, the decoding unit 2
Also in 00, the degree determining means 34 determines the MA degree in accordance with the pitch period 3 of the input voice signal 1, but the optimal codeword selecting means 19 of the coding unit 100 minimizes the distortion between the original voice and the decoded voice. It is also possible to select the order of the MA to be executed and the MA codeword in this order, transmit the order of the MA from the encoding unit 100 to the decoding unit 200, and input the MA order to the MA conversion means 36. The degree determining means 34 of the unit 200 becomes unnecessary.

Embodiment 3 FIG. Further, in the first embodiment, the example in which the cepstrum coefficient is used as the MA code word has been described.
It is also possible to use the spectral envelope parameters of LSP and PARCOR as codewords.

Embodiment 4 FIG. Further, as the AR codeword of the first embodiment, the spectral envelope parameters of LSP, PARCOR, and LPC cepstrum can be used.

Embodiment 5 FIG. Further, in the first embodiment, the preliminary selection of each code word is not performed in the optimal code word selecting unit 19 of the encoding unit 100, but the preliminary selection can be performed for each code word.

[0017]

According to the present invention described above, according to the present invention, the coding unit, and the order determining means for outputting a movement determines the order K of the average coefficients MA orders according to the pitch periodic input audio signal, The MA codeword output from the MA codebook
MA converting means for converting to the next MA and outputting to the synthesizing means;
Convert the A codeword MA of the same order as the coding unit, by providing the MA conversion means for outputting the combining means, able to determine the order of the MA in response to an input audio signal, or
And, because also widely adaptation range of MA order, the quality of the synthesized speech can be improved.

[Brief description of the drawings]

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

FIG. 2 is a block diagram showing a conventional speech encoding / decoding system.

[Explanation of symbols]

 Reference Signs List 1 input speech signal 2 pitch period extracting means 3 pitch period 4 order determining means 5 MA order 6 MA codebook 7 MA conversion means 8 MA codeword 8a conversion MA 9 sound source codebook 10 sound source codeword 11 sound source generating means 12 sound source signal 13 AR codebook 14 AR codeword 15 sound source position detecting means 16 sound source position 17 combining means 18 synthesized speech signal 19 optimal codeword selecting means 20 AR codeword number 21 sound source codeword number 22 MA codeword number 23 AR inverse quantizing means 24 AR codebook 25 AR codeword 26 excitation dequantization means 27 excitation codebook 28 excitation codeword 29 excitation generation means 30 excitation signal 31 MA inverse quantization means 32 MA codebook 33 MA codeword 33a conversion MA 34 order determination means 35 MA degree 36 MA conversion means 38 synthesis means 39 synthesized speech signal 100 encoder 2 00 Decryption unit

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-63-244100 (JP, A) JP-A-61-167999 (JP, A) JP-A-63-143598 (JP, A) 80798 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 11/00-21/06 H03M 7/30 H04B 14/04 JICST file (JOIS)

Claims (1)

(57) [Claims] 1. A speech encoding / decoding system for separating and encoding an input speech signal into a sound source signal, an autoregressive coefficient and a moving average coefficient, comprising: A plurality of excitation codebooks stored as a plurality, an autoregression coefficient codebook storing a plurality of autoregression coefficients as autoregression coefficient codewords, and a plurality of moving average coefficients converted to spectral envelope parameters stored as a moving average coefficient codeword the moving average coefficient codebook, and the order determining means for determining the order of the moving average coefficients according to the pitch periodic input audio signal, the order determining unit moving average coefficient code word output from the moving average coefficient codebook Moving average coefficient conversion means for converting to a moving average coefficient of the order determined in the above, excitation generation means for generating an excitation signal from an excitation codeword output from an excitation codebook, and excitation generation Synthesizing means for generating a synthetic speech signal from the sound source signal output from the stage, the autoregressive coefficient codeword output from the autoregressive coefficient codebook, and the moving average coefficient output from the moving average coefficient converting means, and output from the synthesizing means. The combination of the excitation codeword, the autoregression coefficient codeword, and the moving average coefficient codeword when the distance between the synthesized speech signal to be input and the input speech signal is the smallest is selected, and the codeword number is designated as the excitation codeword number, autoregression. An optimal codeword selection means for outputting as a coefficient codeword number and a moving average coefficient codeword number is provided. A decoding section selects an excitation codeword corresponding to the excitation codeword number transmitted from the encoding section from the excitation codebook. Sound source inverse quantization means, sound source generation means for generating a sound source signal from the selected sound source codeword, and an autoregressive coefficient corresponding to the autoregressive coefficient codeword number transmitted from the encoding unit Auto-regression coefficient inverse quantization means for selecting a word from the auto-regression coefficient codebook, and selecting a moving average coefficient code word corresponding to the moving average coefficient code word number transmitted from the encoding section from the moving average coefficient code book. A moving average coefficient inverse quantization means, a moving average coefficient conversion means for converting the selected moving average coefficient codeword into a moving average coefficient of the same order as that of the encoding section, a sound source signal output from the sound source generation means and autoregression A speech coding / decoding system comprising a synthesizing means for generating a synthesized speech signal from an autoregressive coefficient codeword output from a coefficient inverse quantization means and a moving average coefficient output from a moving average coefficient conversion means.