JPH0365560B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (A) Technical Field of the Invention The present invention provides a speech analysis and synthesis device, in particular, a method for converting a power spectrum to 1/n in a form corresponding to linear prediction coefficients.
In a speech analysis and synthesis device that obtains deformed prediction coefficients by performing multiplication compression, instead of using a filter configured using the linear prediction coefficients as coefficients, the filter used for the deformed prediction coefficients is The present invention relates to a speech analysis and synthesis device that employs a configuration that allows signals to pass through in series.

(B) Technical Background and Problems Conventionally, linear prediction coefficients have been extracted to extract parameters used in speech synthesis, speech recognition, etc. In the above-mentioned speech synthesis and speech recognition, the spectral envelope information of the input speech signal is obtained from the above-mentioned linear prediction coefficients by, for example, treating the prediction coefficient itself as a time function and performing Fourier transform to calculate the inverse spectrum of the spectrum. , or further use the spectral envelope information to obtain formant frequencies and the like.

However, in the case of the conventionally known method for extracting spectral envelope information, there is a problem that the obtained spectral envelope information is affected by the pitch frequency of the input voice. In order to solve this problem, the present inventors previously applied for a patent application filed in 1983-
No. 188060, Special Application No. 188061, Special Application No. 1880-
No. 50431 and others, we proposed a system that compresses the power spectrum extracted from input speech and then extracts the "deformation" prediction coefficient α'. That is, in general, this type of speech analysis and synthesis apparatus adopts the configuration shown in FIG. 1, and improves the point that the linear prediction coefficient α is obtained and, for example, spectral envelope information P^(w) is extracted. , second
As shown in the figure, it is clear that the "deformation" prediction coefficient α' is obtained, then the deformed spectrum envelope information P^(w) is obtained, and then the spectrum envelope information P^(w) is obtained by decompression. is being used. In FIG. 1, 1 is a Fourier transform processing unit that performs a Fourier transform on a discrete input audio signal S(n);
2 is a square value extraction unit that extracts the power spectrum P(w) of the input voice, and 3 is a Fourier inverse transform processing unit that extracts the power spectrum P(w).
4 is a linear prediction coefficient calculation unit which calculates the linear prediction coefficient α( based on the autocorrelation coefficient R(n). 5 is a Fourier transform processing unit that performs Fourier transform by regarding the linear prediction coefficient α(n) as a time function;
6 represents a square value extraction section, and 7 represents a reciprocal number processing section. The Fourier transform processing section 5, the square extraction section 6, and the reciprocal processing section 7 are used to calculate the linear prediction coefficient α.
Spectral envelope information P^ of the input audio signal from (n)
It can be thought of as extracting (w). and the second
In the figure, the numbers 1 to 7 and S in the figure
(n), P(w), and P^(w) correspond to those in FIG. 1, 8 represents a conversion processing section, and 9 represents an inverse conversion processing section, which is received in FIG.

In FIG. 2, the power spectrum P(w) of the input voice is obtained by the square value extraction unit 2. For example, P′(w)=[P (w)] A conversion processing unit 8 that provides a conversion of ^1/n −(1) is inserted. In the case shown in Figure 2, the input audio signal S(n)
After performing the transformation shown in equation (1) on the power spectrum P(w) whose absolute value was obtained by Fourier transformation, the modified autocorrelation coefficient R′(n) and the modified prediction coefficient α′(n), deformed spectrum envelope information P^(w)
Then, the inverse transformation of the transformation of the above equation (1) is performed in the inverse transformation processing section 9. That is,
In the frequency domain after the Fourier transform of the input audio signal S(n) and before the inverse transform by the Fourier inverse transform processor 3, the transform shown in equation (1) is performed to obtain spectral envelope information. In extracting P^(w), the inverse transformation P^(w)=P^'(w) ⁿ −(2) is performed.

As shown in Fig. 2 above, many advantages can be enjoyed by extracting the deformation prediction coefficient α'.
Using the spectral envelope information P^(w) obtained from the inverse transform processing unit 9 shown in the figure, the linear prediction coefficient α without the above compression processing (or the result after the above compression/expansion is performed) Therefore, it is necessary to reproduce the linear prediction coefficient α″) and use a filter that uses the linear prediction coefficient as a coefficient.

(C) Object and Structure of the Invention In view of the above-mentioned points, the present invention improves the point of constructing a filter by regenerating the linear prediction coefficient α″ from the obtained deformation prediction coefficient α′, and The purpose of the present invention is to configure a filter (referred to as a deformation filter in this specification) that uses ′ as a coefficient to perform desired processing.For this reason, the speech analysis and synthesis device of the present invention has the following features: , at least means for determining a power spectrum from an input voice, and 1/n for the determined power spectrum.
a conversion processing unit that performs multiplication compression; and a prediction coefficient calculation unit that calculates a prediction coefficient based on the output from the conversion processing unit; In a speech synthesis device that obtains a deformation prediction coefficient corresponding to the multiplication result and outputs the deformation prediction coefficient and uses it for speech analysis and synthesis, the deformation prediction coefficient α' is used as a coefficient,
The present invention is characterized in that it is equipped with a filter having a width Σα _i ′Z ^-i , and a signal is made to pass through the filter n times in cascade to obtain a filter output. This will be explained below with reference to the drawings.

(D) Embodiments of the Invention Fig. 3 shows an example of a conventional configuration in which speech synthesis is performed using the pitch period, power, and voiced/unvoiced information obtained based on the residual signal. An embodiment of the present invention corresponding to FIG. 3, FIG. 5 is an example of a conventional configuration used in a waveform encoding method, and FIG.
The figure shows an embodiment of the invention corresponding to FIG.

In FIG. 3, 10 is a noise component, 11 is a pitch periodic component, 12 is a power component, 13 is a filter configured using linear prediction coefficient α (or α″) as a coefficient, and S(n ) represents synthesized speech.

In the conventional configuration, the linear prediction coefficient α (or α″) obtained under the state in which the compression process shown in equation (1) above is canceled is used as the coefficient in the filter 13 shown in FIG. Z represents the time delay unit. However, as stated at the beginning of this specification, the linear prediction coefficient α'' is generated again from the state where the deformation prediction coefficient α' has been obtained with the configuration shown in FIG. It is undesirable and complicated to use.

In the present invention, the above-mentioned deformation prediction coefficient α′, deformation spectrum envelope information P^(w), linear prediction coefficient α,
The relationship between spastle inclusive information P^(w) is
P^′(w)←……→α′ [P^′(w)] ⁿ = P^(w) P^(w)←……→α″ Focusing on the relationship, (1 /1+Σα _i ″Z ^-i )←……→(1/1+Σα′ _i Z
^-i ) It is assumed that it corresponds to ⁿ , and is configured accordingly. Fig. 4 shows its configuration, with illustration symbols 10, 11, 12,
(n) corresponds to FIG. 3, and 14 represents one of the modified filters according to the present invention. Although FIG. 4 shows that n deformable filters 14 are connected in series, for example, only the deformable filter 14-1 is used to connect the filter 14-1.
It goes without saying that the output of the filter 14-1 may be returned to the input side by using only the filter 14-1, so that the signal passes n times in cascade.

FIG. 5 shows an example of a conventional configuration used in the waveform encoding method, where χ is an audio signal, ε is a residual signal, and 15
16 is a transmitting side filter, which is configured using linear prediction coefficient α (or α″) as a coefficient, and 16 is a receiving side filter, which is configured using linear prediction coefficient α (or α″).
It represents what is constructed using the coefficients.

In the embodiment of the present invention shown in FIG. 6, a deformation filter 19 constructed using the above-mentioned deformation prediction coefficient α' as a coefficient is used as the transmitting side filter 17 and the receiving side filter 18.
Then, the filters 15 and 16 and the filter 1
The relationship between 7 and 18 is such that (1+Σα _i ″Z ^-i )←……→(1+Σα _i ′Z ^-i ) ⁿ .
In the case shown in FIG. 6 as well, it is not necessary to connect n deformation filters 19 in series, and it is sufficient if the signal passes through the deformation filters 19 n times.

(E) Effects of the Invention As explained above, according to the present invention, it is possible to use a filter that uses the deformation prediction coefficient α' as it is as a coefficient.

[Brief explanation of drawings]

Figures 1 and 2 are explanatory diagrams explaining the prerequisite problem of the present invention, and Figure 3 is an explanatory diagram for explaining the prerequisite problem of the present invention. An example of a conventional configuration for performing synthesis, FIG. 4 is an embodiment of the present invention corresponding to FIG. 3, FIG. 5 is an example of a conventional configuration used in a waveform encoding method, and FIG. 6 corresponds to FIG. 5. An embodiment of the present invention is shown below. In the figure, α or α″ is a linear prediction coefficient, α′ is a deformation prediction coefficient, 13, 15, 16 are filters, 14, 1
9 represents a deformation filter.

Claims (1)

[Claims] 1. At least means for determining a power spectrum from an input voice; and 1/n for the determined power spectrum.
a conversion processing unit that performs compression to multiply the power spectrum; and a prediction coefficient calculation unit that calculates a prediction coefficient based on the output from the conversion processing unit; In a speech analysis and synthesis device that obtains a deformation prediction coefficient corresponding to the result and outputs the deformation prediction coefficient and uses it for speech analysis and synthesis, the deformation prediction coefficient α' is used as a coefficient,
1. A speech analysis and synthesis device comprising a filter having a width Σα _i ′Z ^-i and passing a signal through the filter n times in cascade to obtain a filter output.