JPH0580798A - Speech encoding and decoding device and sound source generating method - Google Patents

Abstract

PURPOSE:To provide a frame synchronizing process by stably interpreting and encoding an input speech with a small arithmetic quantity by using spectrum parameters and sound source parameters. CONSTITUTION:An AR preparatory selecting means 6 outputs a preparatory select AR code word 8 and an MA preparatory selecting means 16 outputs a preparatory select MA code word 18 selected from an MA code table 17 according to the distance to MA 15. A sound source preparatory selecting means 9 outputs a preparatory select sound source code word 11 selected preparatorily from a sound code book 10 according to a sound source code word series and a sound source position detecting means 2 detects a sound source position 3 in a frame; and a sound source generating means 12 generates a sound source 13 synchronized with the sound source position 3 from the preparatory select sound source code word 11 and a synthesizing means 19 synthesizes a synthesis speech 20 of the preparatory select AR code word 8, preparatory select MA code word 18, and sound source 13. An optimum code word selecting means selects the combination of an AR code word, an MA code word, and a sound source code word which minimize the distance between the synthesis speech 20 and an input speech 1 among the preparatory select AR code words 8, preparatory select MA code words 18, and preparatory select sound code words 11.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coding / decoding device used when digitally transmitting or storing voice.

[0002]

2. Description of the Related Art A conventional speech coding / decoding apparatus using a sound source signal of one pitch period (hereinafter referred to as a sound source) is, for example, "".
Estimation of ARMA parameters of speech using vocal cord source waveform model "Mats Junckvist, Hiroya Fujisaki Technical Report of IEICE SP86-49, PP39-4
5, 1986 ”. In this conventional one, an AR parameter (hereinafter abbreviated as AR) and an MA parameter (hereinafter abbreviated as MA) are used as spectrum parameters, and a sound source wave model defined on the differential waveform of the glottal sound source is used as a sound source. ..

FIG. 6 is a block diagram showing the structure of this conventional speech coding / decoding apparatus. FIG. 6 (a) is an analysis unit, and FIG.
(B) shows a synthesis part. First, the analysis unit shown in FIG. 6A will be described. The ARMA analysis means 44 obtains AR 45 and MA 46 from the input voice 1 of one pitch period and the sound source 13 generated by the sound source generation means 12, and outputs them to the synthesis means 19. The synthesizing means 19 generates a synthesized voice 20 of one pitch period from the sound source 13, AR 45, and MA 46. The distance calculating means 47 calculates the distance E1 between the synthetic voice 20 and the input voice 1.

When the distance E1 is less than the threshold value E0, the sound source parameters 48, AR49 and MA50 are output. Distance E1 is the threshold
If it is equal to or more than E0, a slight perturbation is given to one of the sound source parameters, and this is output to the sound source generation means 12 as the sound source parameter 48. The sound source generation means 12 generates the sound source 13 from the sound source parameter 48 and outputs it to the ARMA analysis means 44. This operation is repeated while reducing the perturbation given to the sound source parameter until the distance E1 becomes less than the threshold E0.

Next, the combining section shown in FIG. 6B will be described. The sound source generation means 40 generates a sound source 41 from the sound source parameter 48. The synthesizing means 42 includes the sound source 41 and the AR4.
9, the synthetic speech 43 is generated using the MA 50.

FIG. 7 is an explanatory diagram showing a source wave model used in the above-mentioned conventional speech coding / decoding apparatus, in which the horizontal axis represents time and the vertical axis represents amplitude. This source wave model g (n) represents a differential glottal source wave, and has variables A, B, C, D,
R, F, W and pitch period T are used as sound source parameters, and are defined by equation (1). In the formula, n is time. Further, α and β in the equation (1) are variables calculated by the equation (2) from the sound source parameter.

[0007]

[Equation 1]

[0008]

[Equation 2]

[0009]

The conventional speech coding / decoding apparatus is configured as described above, and the solution of the spectrum parameter and the excitation parameter is calculated for each parameter AbS.
Since it is performed by (Analysis by Synthesis), there is a large amount of calculation,
There is a problem that the obtained parameters fall into unstable solutions. Further, since the pitch period synchronization processing is performed, there is a problem that it is difficult to reduce the fixed bit rate and reduce the bit rate when encoding the excitation parameters.

Further, since the conventional source wave model has a large number of parameters, there is a problem that a large amount of calculation is required for solution.

The present invention has been made to solve the above problems, and reduces the calculation amount of the solution of the spectrum parameter and the sound source parameter and stabilizes the solution of the parameter.
The object of the present invention is to realize high-quality synthetic speech generation and to achieve a fixed bit rate and a low bit rate by performing frame synchronization processing.

[0012]

A speech coding / decoding apparatus according to the present invention has a coding section, in which a plurality of spectrum analyzing means for analyzing input speech to extract spectrum parameters and a plurality of spectrum parameters as spectrum codewords. The spectrum preselection means for selecting from the spectrum codebook a finite L number of preselected spectrum codewords having a close distance between the spectrum codebook stored as a set and the spectrum parameter output by the spectrum analysis means, and existing in the frame from the input speech. A sound source position detecting means for detecting the start position of all sound sources of one pitch period and outputting it as a sound source position, and a parameter of a sound source model used as a sound source of one pitch period as a sound source parameter, and this sound source parameter as a sound source codeword Excitation codebook with multiple sets stored and excitation code selected in the past Source preselection means for selecting, from the source codebook, a finite number of preselected source codewords having a close distance on the source parameter from the source code source, and a source generating a source synchronized with the source position using the preselected source codeword. Generating means,
The preselection spectrum codeword is a synthesis means for generating synthetic speech from the preselected spectrum codeword and the sound source, and a combination of the spectrum codeword and the excitation codeword that minimizes the distance between the synthetic speech output from the synthesis means and the input speech. And an optimal codeword selecting means for selecting from the preselected excitation codeword,

In the decoding section, the same spectrum codebook and excitation codebook as in the coding section, spectrum dequantization means for outputting the spectrum codeword corresponding to the inputted spectrum codeword number from the spectrum codebook, and the input Source dequantization means for outputting the excitation codeword corresponding to the generated excitation codeword number from the excitation codebook, the same excitation generation means as the encoding part for generating the excitation signal from the excitation codeword, the excitation source, and the excitation source. It is provided with the same synthesizing means as the coder for generating synthetic speech from the spectrum code word.

The sound source generation method according to the present invention is
By using the variables A, B, C, L ₁ , L ₂ and the pitch period T, a sound source signal of one pitch period having a waveform g (n) of the following formula is generated. g (n) = An−Bn ² (0 ≦ n ≦ L ₁ ) g (n) = C (n−L ₂ ) ² (L ₁ <n ≦ L ₂ ) g (n) = 0 (L ₂ <n ≦ T) where n is time.

[0015]

According to the present invention, the spectrum preselection means preliminarily selects a finite number L of spectrum codewords having a small distance from the spectrum parameter obtained by the spectrum analysis means from the spectrum codebook, and the sound source preselection means has been selected in the past. A spectrum codeword that preliminarily selects a finite number M of excitation codewords having a short distance on the excitation parameter from the selected excitation codeword from the excitation codebook, and the optimum codeword selection means minimizes the distance between the synthesized speech and the input speech. By selecting a combination of the and excitation codewords from the preselected spectrum codeword and the preselected excitation codeword and outputting the respective numbers, stable encoding is performed with a small amount of computation, and the decoding unit selects the selected spectrum codeword. Proper decoding is performed based on the word number and the preselected excitation codeword number. Further, according to the sound source generation method according to the present invention, the sound source signal of one pitch period is satisfactorily generated with a small number of parameters.

[0016]

【Example】

Example 1. FIG. 1 is a configuration diagram of an encoding unit of a speech encoding / decoding device according to an embodiment of the present invention, and FIG. 2 is a configuration diagram of a decoding unit. The operation will be described below. Note that FIG.
2, the same parts as those in FIG. 6 are designated by the same reference numerals. First, the encoding unit in FIG. 1 will be described.

The AR analysis means 4 analyzes the input voice 1 by AR,
Output AR5. The AR preselection means 6 uses, for example, the square distance as a distance measure, selects a finite number of AR codewords having a short distance between parameters from the AR 5 from the AR codebook 7, and outputs these as preselected AR codewords 8. To do.

The sound source position detecting means 2 detects, for example, a peak position for each pitch cycle of the LPC residual signal of the input voice 1 and outputs it as a sound source position 3.

The excitation preselection means 9 uses, for example, a weighted square distance between excitation parameters as a distance measure, and a finite M number of excitation codewords having a small distance from the excitation codeword selected in the previous frame are extracted from the excitation codebook 10. It is selected and output as the preselected excitation codeword 11. The sound source generation means 12 uses the preselected sound source codeword 11 to generate a sound source synchronized with the sound source position 3 and outputs it as the sound source 13.

The MA calculating means 14 calculates the MA 15 using the preselected AR codeword 8 and the sound source 13. MA preliminary selection means 16
Uses, for example, the square distance between parameters as a distance measure, selects a finite number of N MA codewords that are close to the MA 15 from the MA codebook 17, and outputs these as preselected MA codewords 18.

The synthesizing means 19 generates a synthesized speech 20 from the preselected AR codeword 8, the preselected MA codeword 18 and the sound source 13. The optimum codeword selecting means 21 selects a combination of an AR codeword, an MA codeword, and an excitation codeword in which the distance between the input speech 1 and the synthesized speech 20 is the smallest, and selects the combination.
The AR codeword number 22, the MA codeword number 23, and the excitation codeword number 24 are output.

FIG. 3 illustrates an example of the operation of the optimum codeword selecting means. First, the input voice (solid line) and the synthetic voice (broken line) in the distance calculation range a including several pitch periods before and after.
The distance E1 is selected to be a combination of an AR code word, an MA code word, and an excitation code word, and the distance E1 is a predetermined threshold value E0.
Select this in the following cases.

When the distance E1 exceeds a predetermined threshold value E0, the several pitch cycle length with the large power of the input voice is set as the distance calculation range b (b <a), and the input voice and the synthesized voice in this range are calculated. A combination of AR codeword, MA codeword, and excitation codeword that minimizes the distance is selected.

The AR codebook 7, the excitation codebook 10 and the MA codebook 17 are A-parameter-specific parameters for a large amount of learning speech.
It is created by clustering the AR parameter, sound source parameter, and MA parameter, which are obtained by bS until a stable solution is obtained, for example, by the LBG algorithm.

Next, the decoding unit shown in FIG. 2 will be described. AR
The inverse quantization means 25 obtains the AR code word 27 corresponding to the AR code word number 22 from the AR code book 26.

The MA inverse quantization means 30 obtains the MA code word 32 corresponding to the MA code word number 23 from the MA code book 31. The excitation dequantization means 35 obtains the excitation codeword 37 corresponding to the excitation codeword number 24 from the excitation codebook 36.

FIG. 4 is an explanatory diagram showing an interpolating method for AR code words, MA code words, and excitation code words. In the figure, V, W, X, Y, Z
Is a synthetic interval of one pitch period. The AR interpolation means 28
For example, the AR codeword 27 of the current frame and the AR codeword of the previous frame are linearly interpolated for each section and output as an interpolated AR 29.

The MA interpolating means 32 linearly interpolates the MA codeword 32 of the current frame and the MA codeword of the previous frame, for example, for each section, and outputs the interpolated MA 34. Sound source interpolation means 3
Reference numeral 8 linearly interpolates the excitation codeword 37 of the current frame and the codeword of the previous frame for each section, and outputs the interpolation excitation parameter 39. The sound source generation means 40 generates a sound source 41 from the interpolation sound source parameter 39. Synthetic means 4
2 generates a synthetic voice 43 from the sound source 41, the interpolated AR 29, and the interpolated MA 34.

As described above, by performing frame synchronization processing by synthesizing while interpolating with the code words of the preceding and following frames, it is possible to achieve a low bit rate and a fixed bit rate. The AR codebook 7 and the AR codebook 26, the excitation codebook 10 and the excitation codebook 36, and the MA codebook 17
And the MA codebook 31 are the same.

FIG. 5 is an explanatory diagram showing an embodiment of a sound source wave model for explaining the sound source generation method of the present invention.
In the figure, the vertical axis represents the time derivative of the sound source wave, and the horizontal axis represents time.
Section a is the time from the glottal opening point to the minimum point, and section b is
Is the time obtained by subtracting the section a from the pitch cycle T, the section c is the time from the minimum point to the zero crossing, and the section d is the time from the glottal opening point to the first zero crossing.

This source wave model is defined on the differential waveform of the glottal source wave, and the differential glottal source wave has a pitch period T, amplitude AM, OQ (the ratio of the section a to the pitch period), OP. (Ratio of section d to section a), CT
It is calculated from Equation (3) using five sound source parameters (the ratio of section c to section b). In the formula, n is time. In the formula (3), A, B, C, and L are formula (4).
Is a variable defined by.

[0032]

[Equation 3]

[0033]

[Equation 4]

Example 2. In the first embodiment, one set of AR code word, MA code word, and excitation code word is selected for one frame, but a plurality of code words can be selected for each parameter.

Example 3. Although AR and MA are used as the spectrum parameters in the first embodiment, it is also possible to use only AR.

Example 4. In the first embodiment, the synthesis means generates synthetic speech from the spectrum parameter and the sound source parameter, but it is also possible to generate the synthetic speech while interpolating the spectrum parameter and the sound source parameter and calculate the distance between the synthetic speech and the input speech. It is possible.

Example 5. In the optimum codeword selection means of the first embodiment, in a frame in which the distance between the synthesized voice and the input voice is large, it is possible to interpolate the spectrum parameter and the sound source parameter from the preceding and succeeding frames to obtain the parameter of the current frame.

Example 6. In the first embodiment, the excitation codeword includes the pitch period T and the amplitude AM.
It is also possible to exclude the excitation code AM from the excitation codeword and perform clustering to create an excitation codebook, and separately encode and decode the pitch period and amplitude.

[0039]

As described above, according to the present invention,
The combination of the spectrum codeword and the excitation codeword that minimizes the distance between the input speech and the synthesized speech is selected from the previously obtained stable codewords to stabilize the solution of the spectrum parameter and the excitation parameter, and the spectrum code Pre-selection of words and excitation code words has the effect of reducing the amount of calculation in the solution of spectrum parameters and excitation parameters.

The encoding unit selects a set of spectrum codeword and excitation codeword for each frame, and the decoding unit interpolates the spectrum codeword and excitation codeword between the codewords of the preceding and succeeding frames, respectively. By performing the frame synchronization processing by synthesizing the signals, the low bit rate and the fixed bit rate can be achieved.

Further, by using the sound source generation method of the present invention, a sound source of one pitch period can be satisfactorily expressed with a small number of parameters, and an effect of reducing the amount of calculation in finding a sound source parameter can be obtained.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an encoding unit of a speech encoding / decoding device showing an embodiment of the present invention.

FIG. 2 is a configuration diagram of a decoding unit of a speech encoding / decoding apparatus showing an embodiment of the present invention.

FIG. 3 is an operation explanatory view of the optimum codeword selecting means in the embodiment of the present invention.

FIG. 4 is an explanatory diagram of a method of interpolating excitation codewords, AR codewords, and MA codewords in the embodiment of the present invention.

FIG. 5 is an explanatory diagram of a sound source wave model according to the sound source generation method of the present invention.

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

FIG. 7 is an explanatory diagram of a conventional source wave model.

[Description of Codes] 1 input speech 2 sound source position detection means 4 AR analysis means 6 AR preliminary selection means 7 AR codebook 8 preliminary selection AR codeword 9 excitation preliminary selection means 10 excitation codebook 11 preliminary selection excitation codeword 12 excitation generation Means 14 MA calculation means 16 MA preliminary selection means 17 MA codebook 18 preselected MA codeword 19 combining means 21 optimal codeword selection means 25 AR dequantization means 26 AR codebook 28 AR interpolation means 30 MA dequantization means 31 MA codebook 32 MA codeword 33 MA interpolation means 35 excitation dequantization means 36 excitation codebook 38 excitation interpolation means 40 excitation generation means 42 synthesis means

Claims (3)

[Claims] 1. A speech coding / decoding device for separating a speech signal into spectral parameters representing its frequency spectrum characteristic and a sound source signal for coding, and analyzing the input speech to a coding section to extract spectral parameters. Spectrum analyzing means, a spectrum codebook in which a plurality of sets of spectrum parameters are stored as spectrum codewords,
Spectral preselection means for selecting, from the spectrum codebook, a finite number of preselected spectrum codewords that are close in distance to the spectrum parameter output by the spectrum analysis means, and an excitation parameter representing an excitation signal of one pitch period as the excitation codeword. Excitation source preselecting means for selecting, from the excitation codebook, a finite number of preselected excitation codewords that have a close distance on excitation parameters between the excitation codebook containing a plurality of sets and the excitation codewords selected in the past, and this excitation source. A one-pitch period excitation generating unit that generates an excitation signal of one pitch period from the pre-selected excitation code word selected by the preliminary selection unit, and a synthesized voice from the pre-selected spectrum code word and the excitation signal of this one pitch period. The synthesizing means and the combination of the spectrum code word and the excitation code word that minimizes the distance between the synthesized speech and the input speech are A pre-selected spectrum code word and the pre-selected excitation code word are selected, and a one-pitch cycle optimum code word selecting means for outputting the spectrum code word number and the excitation code word number of the selected combination is provided, and the decoding section is provided. , The same spectrum codebook as the coding unit, the same excitation codebook as the coding unit, spectrum dequantization means for outputting the spectrum codeword corresponding to the inputted spectrum codeword number from the spectrum codebook, and Source excitation codeword number corresponding to the excitation codeword number from the excitation codebook, and excitation pitch dequantization means, the same one-pitch period excitation generation means as the encoding unit that generates the excitation signal from the excitation codeword, and one pitch A voice code comprising the same synthesizing unit as a coding unit for generating a synthetic voice from the excitation signal output from the periodic sound source generating unit and the spectrum code word. Decoding apparatus. 2. A sound source position detecting means for detecting the start points of all the sound source signals of one pitch period existing in an analysis frame of a fixed time from the input voice and outputting the sound source position to the encoding unit, and the sound source spare. Sound source generation means for generating a sound source signal synchronized with the sound source position output by the sound source position detection means using the preselected sound source codeword selected by the selection means, and a synthesized speech from this sound source signal and the preselected spectrum codeword. And a combination of the spectrum code word and the excitation code word that minimizes the distance between the synthesized speech and the input speech within a range of several pitch periods in three frames including the preceding and following frames. And the optimum codeword selecting means for selecting from among the preselected excitation codewords, and the decoding unit is provided with the current frame obtained by the spectrum dequantizing means. A spectrum code word and a spectrum code word of the previous frame are interpolated for each pitch period, and a spectrum interpolating means for outputting the obtained interpolated spectrum parameter, and an excitation code word of the current frame obtained by the excitation dequantization means The same excitation as the excitation interpolating unit that interpolates the excitation codeword selected in the previous frame for each pitch period and outputs the obtained interpolation excitation parameter, and the encoding unit that generates the excitation signal in the frame from the interpolation excitation parameter The speech coding / decoding apparatus according to claim 1, further comprising a generation unit. 3. A sound source generation method for generating a sound source signal of one pitch period composed of a waveform g (n) of the following equation, using variables A, B, C, L ₁ , L ₂ and a pitch period T. g (n) = An−Bn ² (0 ≦ n ≦ L ₁ ) g (n) = C (n−L ₂ ) ² (L ₁ <n ≦ L ₂ ) g (n) = 0 (L ₂ <n ≦ T) where n is time.