JPS6347799A - Rhythm control system - Google Patents

Abstract

(57) [Abstract] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a prosody control method in the rule synthesis of speech.

In order to add natural prosody in round-the-clock speech synthesis, prosody control rules that control pitch, amplitude, rhythm, etc. are essential. A method for controlling the quantities that make up prosody using multivariate statistical analysis has already been proposed, but this method has the following drawbacks.

(b) The quantity that constitutes the prosody to be controlled is expressed as a linear sum of each factor, and other models cannot be handled.

(b) It is not possible to handle qualitative and quantitative parameters at the same time.

Purpose The present invention was made in view of the above-mentioned circumstances.
In particular, it was developed with the aim of generating highly natural prosodic patterns in the regular synthesis of speech.

Configuration In order to achieve the above object, the present invention is a speech rule synthesis method in which a parameter sequence of speech segments prepared in advance is read out according to an input character string, connected by a combination rule, and prosody is added by a prosody rule. This method is characterized by processing parameters using multivariate statistical analysis to obtain optimal control values. Hereinafter, the present invention will be explained based on examples.

1 and 2 are diagrams for explaining an embodiment of the prosody control method according to the present invention, but the present invention was made to improve the drawbacks of the prior art as described above. In order to improve the above-mentioned drawbacks, we added qualitative parameters through quantitative analysis to the target parameters (various constants that are thought to affect the quantities that make up the prosody), and created a model as an arbitrary function of them. is configured and controlled by that model.

In FIG. 1, 11 is a quantitative parameter section, 12 is a qualitative parameter section, 13 is a quantification type 1 analysis section, ], 4:
15 is the quantity part that makes up the prosody, where Zl is the predicted value of the quantity that makes up the prosody, and zj is the actual value, so that the following equations (1), (2), and (3) are satisfied. quantity (
ajk) and the coefficients of f.

Zj, = f(x,, xz”', xml, '/sr
8/21 "')'m2)"' (1)Σ= (Zi
-zi) 2 → minimum (3) c=1 Here, δ1(jk) is 1 when individual i responds to category K of factor item j. Otherwise, the function takes O, xj is m1 quantities obtained by quantification type I analysis, yl is m2 quantitative parameters that directly contribute to Z, R is the number of items, cj is the number of categories of item j , a jkj is the quantity for the category, and n is the number of data.

FIG. 2 is a diagram for explaining one embodiment of the present invention. In the figure, 21 is the reciprocal part of the mora number, 22 is the coefficient part, 23 is the coefficient C part, 24 is the addition part, and 25 is the front and rear parts. Qualitative parameters such as phonology, long sounds, consonants, etc.

26 is the quantification type 1 analysis part, 27 is the phonological duration length part,
Here, we are considering predicting the phoneme duration length as the quantity 2 that constitutes the prosody. The model uses qualitative parameters such as the preceding and following phonemes and whether or not it is a special phoneme such as a long vowel or consonant as a factor, and the quantity x1 obtained by performing a type I analysis of the utterance, and the mora of the sacrifice position of the utterance. reciprocal of number y
Consider a linear combination with At this time, the function f is f: (x□t yz) =x, +by, +c”
'(4), and the above equations (1) and (2) have the following form. Unknown constants (ajk), b, and c are obtained by partially differentiating equation (5) with respect to each constant using the principle of least squares method.

It can be easily obtained by solving simultaneous equations with 0 as 0.

With the model configured in this manner, the phoneme duration length can be determined by inputting the qualitative parameters and mora number of the phoneme to be determined.

Here, the number of moras is used as a quantitative parameter that directly contributes to Z, but it is also possible to use the mora position, the average power of the phoneme, the pitch, etc. Furthermore, various forms of the function f can be considered, such as a quadratic function of y or a product of X and y. However, if the simultaneous equations determining the unknown constants are nonlinear, it is necessary to introduce a numerical analysis method. Furthermore, it is also possible to predict other elements constituting prosody such as Z, pitch, and amplitude.

As is clear from the above explanation, according to the present invention, qualitative parameters are quantified using a-type I analysis, a model is constructed for quantitative parameters with this added, and naturalness is The prosody can be controlled so that a high synthesized sound can be obtained.

[Brief explanation of the drawing]

Figures 1 and 2 are configuration diagrams for explaining embodiments of the prosody control system according to the present invention. 11... Definite nine (material parameter part), 12... Qualitative parameter part, 13... Quantification ■ type analysis part, 14...
Arbitrary function part, 15... Quantity part forming prosody, 21.
・Reciprocal part of Mora number, 22 ・Coefficient part, 23...
Coefficient C part, 24... Addition part, 25... Qualitative parameter part, 26... Quantification type 1 analysis part, 27... Phoneme duration length part. 1st Figure 1!14 Figure 2

Claims (2)

[Claims] (1) In the speech rule synthesis method, which reads the parameter series of speech segments prepared in advance according to the input character string, connects them using connection rules, and adds prosody using prosody rules, various parameters are processed using multivariate statistical analysis. A prosody control method characterized by obtaining an optimal control value. (2) Express the quantities that make up prosody (phonological duration, pitch pattern, amplitude, etc.) as a function of several quantitative parameters, take at least one of them as an external standard, and use that quantitative parameter as a function. Claim (1), characterized in that a prediction model is set by performing a quantification type I analysis of qualitative parameters that are considered to contribute to factor items, and the quantities constituting the prosody are predicted by this model. Prosody control method described in.