JPH03222000A - Voice analyzing device using vocal cord sound source wave model - Google Patents

Abstract

PURPOSE:To eliminate the useless search of a parameter, to decrease the arithmetic quantity and to shorten the analytical time by preliminarily setting the range of a sound source parameter by a duty ratio at the time of deriving the sound source parameter by an estimating circuit. CONSTITUTION:The pitch period of a sound signal inputted to a buffer device 102 is derived by an extracting device 103, and in accordance therewith, the range of a duty ratio of glottis closing and glottis opening sections of a vocal cord sound source wave is set by a setting circuit 104. Within the set range of this set sound source parameter, the sound source parameter is estimated by an estimating circuit 106 and a voice meatus parameter is estimated by an analyzing circuit 105. At the time, by a sound source wave generating circuit 108, a vocal cord sound source wave is generated from the sound source parameter. Subsequently, the analysis by the circuit 106 and the voice meatus parameter and the sound source parameter are outputted from a port 107. In such a way, the arithmetic quantity is decreased and the analytical time can be shortened.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a speech analysis device that simultaneously estimates sound source parameters and vocal tract parameters of an input speech signal.

(Prior Art) In a speech synthesis method, more natural synthesized speech can be generated by driving a vocal tract filter with a vocal cord sound source wave. This synthesis method requires an analysis method to estimate the sound source and vocal tract parameters from the speech signal. An analysis method that attempts to estimate both vocal tract characteristics and sound source characteristics from speech signals using a vocal cord sound source wave model has been proposed by Fujisaki et al. (IEICE Journal J72-DI [N
.

8.1109-1117. (September 1989) The six-vocal fold sound source wave model is a model in which the vocal cord waves are formulated using a mathematical formula consisting of several sound source parameters. Speech analysis using a vocal cord sound source wave model assumes vocal cord waves generated by a vocal cord sound source wave model, aligns the vocal cord waves and the voice signal, and performs AR analysis or ARM A analysis of the known input. The AR coefficient or ARMA coefficient is determined by expressing the characteristics of the vocal tract. An analysis method that optimally estimates both the vocal fold wave and the vocal tract characteristics using the vocal fold wave with the smallest analysis error among all vocal fold waves and the AR, coefficient or ARMA coefficient estimated at that time as vocal tract parameters. It is.

(Problems to be Solved by the Invention) In conventional speech analysis methods, pitch is first extracted, and the sound source parameters and vocal tract parameters are estimated by assuming vocal cord sound source waves over the entire pitch period range.
The number of assumed vocal cord sound source waves is quite large, and therefore the amount of calculation required is large and the analysis takes time.

The present invention aims to reduce the amount of calculation required to estimate both vocal fold waves and vocal tract characteristics, and to speed up the analysis.

(Means for Solving the Problems) Means provided by the first invention of the present application in order to solve the above-mentioned problems is a speech analysis device that estimates sound source parameters and vocal tract parameters of input speech signals. means for extracting a pitch period; means for setting a value range of the ratio of the open interval to the closed interval of the vocal cord wave set from the pitch period extracted by the extracting means; means for estimating the value of the sound source parameter within the range of the ratio;
The source parameter estimating means includes means for generating a vocal cord sound source wave from the sound source parameter estimated by the estimating means, and means for estimating a vocal tract parameter of the speech signal using the vocal cord sound source wave, and the sound source parameter estimating means The present invention is a speech analysis device using a vocal cord sound source wave model, characterized in that the sound source parameters are modified within a range of values.

Means provided by the second invention of the present application in order to solve the above-mentioned problem is a speech analysis device that estimates sound source parameters and vocal tract parameters of an input speech signal, and includes means for extracting a pitch period. , a means for setting the range of the ratio of the open interval to the closed interval of the vocal cord wave set from the pitch period extracted by this extraction means, and a finite number of values obtained by quantizing the sound source parameters within a certain range. means for storing a sound source code consisting of a combination; means for setting a range of sound source calls from the value range of the ratio of the open section to the closed section set by the setting means; and setting by the sound source code range setting means. means for selecting the sound source code from the storage means within the selected range; means for generating a vocal cord wave of the sound source code selected by the selection means; A vocal cord sound source wave model comprising: means for analyzing vocal tract parameters of a signal; and means for determining a sound source code that minimizes an analysis error by the analysis means within the sound source codes selected by the selection means. This is a speech analysis device using

(Function) In the first invention, when obtaining sound source parameters using a sound source parameter estimation circuit using the steepest descent method, etc.
By setting the range of sound source parameters in advance using the duty ratio, unnecessary searches for parameters can be considerably reduced. Therefore, as the number of parameter combinations decreases, convergence to the optimal solution becomes faster and the analysis time becomes shorter.

In the second invention, the number of chords to be searched is considerably reduced by setting the range of sound source parameters, so the amount of calculation required for analysis is reduced and the analysis time is shortened.

It is known from physiological observations that the duty ratio itself changes depending on the pitch period. Taking this into consideration, the duty ratio is also controlled based on the pitch period.

(Example) An example of the first invention is not shown in FIG. First, the operation of each module will be explained. Buffer device 102 is a device that stores audio signals. The pitch extraction device 103 is a device for extracting a pitch period from an audio signal, and uses a conventional device. The sound source parameter setting circuit 104 includes 10
This circuit sets the duty ratio of the glottal closure section and the glottis [1 section] of the vocal cord sound source wave from the pitch period extracted by step 3, and sets the variable range of the sound source parameter. The sound source wave generation circuit 108 is a circuit that generates vocal cord sound source waves from sound source parameters. An analysis circuit 105 performs speech analysis using the vocal cord sound source wave generated in step 108 to estimate vocal tract parameters. A conventional analysis circuit 105 is used. A sound source parameter estimation circuit 106 estimates sound source parameters. Here, conventional sound source parameter estimation algorithms such as the steepest descent method and the hill climbing method are used.

Next, the overall flow of the first embodiment will be explained. signal line 10
1, the voice signal is inputted, the voice signal is stored in the buffer device 102, the pitch period of the voice signal is obtained in the pitch extraction device 103, and the glottal closure period and the glottis opening period of the vocal cord sound source wave are determined according to the pitch period. The setting circuit 104 sets the value range of the duty ratio. A sound source parameter estimation circuit 106 estimates the sound source parameters within the sound source parameter setting range set in step 104, and an analysis circuit 105 estimates vocal tract parameters. At that time, a vocal cord sound source wave is generated from the sound source parameters at 108 . Analysis circuit 10 for each sound source parameter estimated by sound source parameter estimation circuit 106
After repeating the above analysis, the final vocal tract parameters and sound source parameters are sent to port 107 as analysis results.
is output.

An embodiment of the second invention is not shown in FIG. First each modi, l
- Explain the operation of the key. The buffer device 102, pitch extraction device 1o3, and analysis device 105 are the same as in FIG. A chord range setting circuit 201 sets the duty ratio of the glottal closing section and the glottal opening section of the vocal cord wave from a pitch period of 103. The chord control circuit 202 selects a chord that satisfies the chord range set in step 201 as a sound source code 20.
4 and sends the code information to the sound source wave generation circuit 108. When the analysis of all codes within the range is completed, analysis completion information is sent to the code determination circuit 203. After completing the analysis of all the codes selected in step 202, the code determination circuit 203 determines the code that minimizes the evaluation function such as analysis error.

Next, the overall flow of the second embodiment will be explained. Signal line 101
The input voice signal is stored in the buffer device 102, and the pitch extraction circuit 103 extracts the pitch period of the voice signal. A duty ratio is set in step 201 based on the pitch period, and a code range is set. A chord is selected in step 202 according to the set chord range, and a vocal cord sound source wave for the selected chord is generated in step 108.
5 and perform the analysis. After completing the analysis using all the selected chords, the chord determination circuit 203 selects the optimal chord, and the vocal tract parameters and chord corresponding to the optimal chord are sent to the boat 107.
output as analysis results.

(Effects of the Invention) As described above, according to the present invention, the amount of calculation can be reduced to one fourth or less when performing analysis using a vocal cord sound source wave model. Furthermore, it is no longer necessary to estimate vocal cord waves that are actually impossible, and the reliability of the analysis is improved.

[Brief explanation of drawings]

FIG. 1 is a configuration diagram showing an embodiment of the first invention of the present application, and FIG.
The figure is a configuration diagram showing an embodiment of the second invention of the present application. 101... To audio hole capo 1, 102... Buffer device, 103... Pitch extraction device, 104... Sound source parameter setting circuit, 105... Analysis circuit, 106...
...Sound source parameter estimation circuit, 107...Output port for analysis results, 108...Sound source wave generation circuit, 201...
- Code range setting circuit, 202... code control circuit,
203...Code determination circuit, 204...Sound source code.

Claims (2)

[Claims] (1) In a speech analysis device that estimates the sound source parameters and vocal tract parameters of an input speech signal, there is a means for extracting a pitch period, and an aperture section of a vocal cord wave that is set from the pitch period extracted by the extraction means. means for setting a value range of the ratio of the closed interval to and means for estimating a vocal tract parameter of the speech signal using the vocal cord sound source wave, and the sound source parameter estimating means corrects the sound source parameter within the range of the value of the ratio. A speech analysis device using a featured vocal cord sound source wave model. (2) In a speech analysis device that estimates the sound source parameters and vocal tract parameters of an input speech signal, there is provided a means for extracting a pitch period, and an aperture section of a vocal cord wave that is set from the pitch period extracted by the extracting means. means for setting a range of values of the ratio of the closed interval to means for setting a range of sound source codes from a range of values of ratios of intervals and closed intervals; means for selecting said sound source code from said storage means within the range set by said sound source code range setting means; and said selection means. means for generating vocal cord waves of the sound source code selected by the vocal cord wave generating means; means for analyzing vocal tract parameters of the voice signal using the vocal cord waves generated by the vocal cord wave generating means; and means for determining a sound source code that minimizes an analysis error by the analysis means.