JPH0558559B2 - - Google Patents

Abstract

PURPOSE: To obtain a high recognition rate by utilizing the pattern of a vector field, performing a defocusing process (blurring process) by directions, and utilizing its result for voice recognition. CONSTITUTION: A time space pattern is converted into a vector field pattern having size and a direction at each grating point in a space by spatial differentiation, and as to the vector of the vector field pattern, its direction parameter is quantized into N values (N: integer). This quantized value is separated by vectors to generate N two-dimensional patterns having the sizes of the vectors as values at respective grating points, and the blurring process is performed by the directions of the two-dimensional patterns by the directions as to a time base and/or a space axis to extract a pattern as a feature of the voice. Consequently, high recognition rates is obtained even for voice recognition regarding large vocabulary and voice recognition regarding an unspecified speaker.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a speech feature extraction method used for speech recognition, etc. More specifically, the present invention utilizes vector field patterns and performs blurring processing ( The present invention provides a new method that can obtain a high recognition rate when used for speech recognition by applying blurring processing.

[Prior art]

Generally, in speech recognition, a standard pattern of speech obtained by extracting features from the word to be recognized is prepared for each word, and feature patterns extracted in the same way from the speech input as recognition target and multiple standards are prepared for each word. The standard pattern with the highest similarity is found by matching the patterns, and the word associated with this standard pattern is determined to have been input.
Conventionally, the spatio-temporal pattern itself of a scalar field with the horizontal axis as the time axis and the vertical axis as the spatial axis, which is obtained by analyzing the audio signal, has been used as the feature pattern. A typical spatio-temporal pattern of such a scalar field is a spectrum with frequency as its spatial axis, as well as a cepstrum with quefrency as its spatial axis, a PARCOR coefficient,
Various spatiotemporal patterns such as LSP coefficients and vocal tract cross-sectional area functions were used.

In addition, one of the issues to be solved in the field of speech recognition is dealing with multiple speakers or unspecified speakers, and improving the recognition rate by preparing a large number of standard patterns for one word. I was thinking about it. Furthermore, even if speakers are the same, their pronunciation speeds may differ, and a DP matching method that can absorb time axis fluctuations has been developed to cope with such cases.

[Problem to be solved by the invention]

Conventional methods that use the spatiotemporal pattern of the scalar field itself as a feature do not necessarily achieve a sufficient recognition rate when targeting large vocabularies or unspecified speakers. Many standard patterns are available or DP
Even using the matching method, these problems could not be fully resolved.

Therefore, the practical application of speech recognition systems for large vocabularies or for unspecified speakers has stalled. Therefore, one of the inventors of the present invention
Publication No. 59394 and "Comparison of effectiveness of spectral vector field and spectrum in speech recognition"
Journal of the Institute of Electronics and Communication Engineers (D), Vo1.J69−D.No.1P1704
(1986) proposed a method for spatially differentiating the spectrum of a scalar field, which is a time-frequency spatiotemporal pattern, to obtain a spectral vector field pattern, and using this pattern as a feature of speech.

Research using partial differentials of space-time points in past spectra as features was conducted by TB Martin.
“Practical applications of voice input to
machines" Proc. IEEE, 64-4 (1976). However, TBMartin uses the spatio-temporal pattern f(t, x) as ∂f(t,x)/∂t, ∂f(t,x). /∂x is calculated,
In this way, we construct a function that identifies 32 types of phonology for each frame, and divide the results into 32 types of phonology.
This method uses values expressed as values for word-by-word linear matching, which is different from the method described above that creates a spectral vector field from a spectral scalar field.

The main object of the present invention is to provide a speech feature extraction method that is improved from an engineering perspective and is suitable for practical use.

Another object of the present invention is to provide a speech feature extraction method that can obtain a high recognition rate even in speech recognition that targets a large vocabulary and speech recognition that targets unspecified speakers.

[Means to solve the problem]

The basic feature of the present invention is speech feature extraction that analyzes an audio signal to obtain a spatiotemporal pattern of a scalar field defined by a time axis and a spatial axis, and extracts speech features using the spatiotemporal pattern. In this method, the spatiotemporal pattern is spatially differentiated to convert it into a vector field pattern having a magnitude and direction at each grid point in space, and the direction parameter of the vector of the vector field pattern is set to N values (N: an integer). ), separate this quantized value into vectors with the same value, create N two-dimensional patterns for each direction with the size of the vector as the value of each grid point, and create a two-dimensional pattern for each direction. Depending on the direction of the pattern,
The purpose of this method is to extract a pattern obtained by blurring only the time axis or both the time and space axes as audio features.

This blurring process only needs to be performed on the time axis when extracting voice features of only one gender, male or female.

When extracting features of voices of both sexes, blurring processing is also performed on the spatial axis, but the blurring processing on the time axis is emphasized more than the blurring processing on the spatial axis.

Further, these blurring processes are performed by performing a mask operation using a mask pattern having a predetermined weight value.

[Effect]

The input audio signal is quantized from the spatio-temporal pattern of the scalar field defined by the time and space axes, and the vector direction parameters are quantized into two-dimensional patterns separated for each quantized direction. converted. This directional two-dimensional pattern is then subjected to blurring processing, and directional pattern features are integrated. This results in enhancement and stabilization of voice features.

This integration performs a kind of structuring of the space-time point (t, x). In other words, this structuring is N
When considering the directional patterns in an integrated manner, a maximum of N vectors are added to the space-time point (t, x) (see Fig. 6). The effect of this on speech recognition lies in the formation of features that better represent phonology and their stable representation, and also corresponds to changes in the spectrum of spatiotemporal intervals in which phonology features exist.

This feature is first extracted microscopically in the spectral vector field, and then vectors in different directional intervals are considered as independent features, and then they are independently integrated at each space-time point. When integrating within a blurred mask pattern independently for each direction, more macroscopic features (audio features created by a wide spatiotemporal region) can be captured while the structural nature of the features is maintained. Moreover, the accumulation of this feature is the space-time point (t,
x), this audio feature is not a macroscopic feature formed only at a specific spatiotemporal point, but a wide range (especially in time) that differs slightly from time to time.
It will be stably formed over the area.

Therefore, by emphasizing and stabilizing the blurring process, phoneme differentiation and speaker normalization can be performed with higher accuracy than before.

〔Example〕

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail below based on drawings showing embodiments thereof.

FIG. 1 is a block diagram showing the configuration of an apparatus for implementing the method of the present invention.

In this embodiment, the analysis unit 2 spectrally analyzes the audio signal and uses a spectrum with the spatial axis as the frequency axis as the spatiotemporal pattern of the scalar field.

The input of the voice for standard pattern creation or the voice to be recognized is performed by the voice input unit 1 consisting of a voice detector such as a microphone and an A/D converter, and the voice signal obtained thereby is for multiple channels with different passing frequency bands (e.g.
10 to 30) bandpass filters connected in parallel. In the analysis section, a spatio-temporal pattern is obtained as a result of the analysis, and this pattern is segmented into each recognition unit word by the word section extraction section 3 and is provided to the feature extraction section 4 . As the word section cutting section 3, a conventionally known one may be used.

In the following description, a group of bandpass filters will be used as the analysis section 2 that divides the audio signal into frequency bands, as described above.
A fast Fourier transformer may also be used.

Now, the method of the present invention is characterized by the feature extraction section described below. The input pattern to the feature extraction unit 4 is a spatio-temporal pattern with the horizontal axis as the time axis and the vertical axis as the frequency, and the spatio-temporal pattern shown in FIG. (t, x)
(However, t is a number indicating the sampling time, x is a channel number of a bandpass filter or a number specifying a frequency band. 1≦t≦T, 1≦x≦L However, T and L are the maximum values of t and x, respectively. ).

The output of the word section extraction section 3 is input to the normalization section 41 of the feature extraction section 4, and the normalization section 41 linearly normalizes the time axis. This is to absorb the length of words, the length of the input voice, etc. to a certain extent, and the time axis is T.
frame to M frames (for example, about 16 to 32 frames). Specifically, if M≦T,
The normalized spatiotemporal pattern F(t, x) is as follows (1)
It is determined by the formula.

F(t,x)= _(T/M)

Claims (1)

[Claims] 1. A speech feature extraction method that analyzes a speech signal to obtain a spatiotemporal pattern of a scalar field defined by a time axis and a spatial axis, and extracts speech features using the spatiotemporal pattern. When extracting the features of voices of only one gender, male or female, the spatiotemporal pattern is transformed into a vector field pattern having a magnitude and direction at each grid point in space by spatial differentiation, and the vector field pattern is For vectors, quantize their direction parameters into N values (N: integer), separate these quantized values for each vector,
N directional two-dimensional patterns are created with the size of the vector as the value of each grid point, and the pattern is obtained by blurring only the time axis for each direction of the two-dimensional pattern. A voice feature extraction method characterized by extracting as follows. 2. In a speech feature extraction method in which a speech signal is analyzed to obtain a spatiotemporal pattern of a scalar field defined by a temporal axis and a spatial axis, and speech features are extracted using the spatiotemporal pattern, the spatiotemporal pattern is is transformed into a vector field pattern that has a magnitude and direction at each grid point in space by spatially differentiating, and the direction parameter of the vector of the vector field pattern is quantized to N values (N: integer), and this quantum Separate each vector with the same value, create N directional two-dimensional patterns with the size of the vector as the value of each grid point, and create a time axis for each direction of the two-dimensional pattern. and extracting a pattern obtained by performing blurring processing on the spatial axis as a feature of the sound, and the blurring processing is performed by emphasizing the blurring processing on the temporal axis more than the blurring processing on the spatial axis. method. 3. In a speech feature extraction method in which a speech signal is analyzed to obtain a spatiotemporal pattern of a scalar field defined by a temporal axis and a spatial axis, and speech features are extracted using the spatiotemporal pattern, the spatiotemporal pattern is is transformed into a vector field pattern that has a magnitude and direction at each grid point in space by spatially differentiating, and the direction parameter of the vector of the vector field pattern is quantized to N values (N: integer), and this quantum Separate each vector with the same value, create N directional two-dimensional patterns with the size of the vector as the value of each grid point, and create a time axis for each direction of the two-dimensional pattern. and a pattern obtained by performing blurring processing on the spatial axis is extracted as a voice feature, and the blurring processing is performed by emphasizing the blurring processing on the time axis more than the blurring processing on the spatial axis, and the features of voices of both sexes are extracted. A voice feature extraction method characterized in that the blurring processing on the spatial axis in the case of is performed more emphatically than the blurring processing on the spatial axis in the case of extracting features of voices of only one gender. 4. In a speech feature extraction method in which a speech signal is analyzed to obtain a spatiotemporal pattern of a scalar field defined by a temporal axis and a spatial axis, and speech features are extracted using the spatiotemporal pattern, the spatiotemporal pattern is is transformed into a vector field pattern that has a magnitude and direction at each grid point in space by spatially differentiating, and the direction parameter of the vector of the vector field pattern is quantized to N values (N: integer), and this quantum Separate each vector with the same value, and create N two-dimensional patterns for each direction with the size of the vector as the value of each grid point, and at least time for each direction of the two-dimensional pattern for each direction A pattern formed by performing blurring processing on the axis is extracted as a sound feature, and the blurring processing has a center point corresponding to the grid point and a center point from the center point for each grid point of the two-dimensional pattern for each direction. A voice feature extraction method characterized in that the process performs a mask calculation on a mask pattern having a spread of two or more grid points in both directions of a time axis and having a predetermined weight value. 5. The audio feature extraction method according to claim 4, wherein the weight values are all "1". 6. In a speech feature extraction method in which a speech signal is analyzed to obtain a spatiotemporal pattern of a scalar field defined by a temporal axis and a spatial axis, and speech features are extracted using the spatiotemporal pattern, the spatiotemporal pattern is is transformed into a vector field pattern that has a magnitude and direction at each grid point in space by spatially differentiating, and the direction parameter of the vector of the vector field pattern is quantized to N values (N: integer), and this quantum Separate each vector with the same value, create N directional two-dimensional patterns with the size of the vector as the value of each grid point, and create a time axis for each direction of the two-dimensional pattern. and/or a pattern formed by performing blurring processing on the spatial axis is extracted as a sound feature, and the blurring processing has a central point corresponding to the grid point for each grid point of the two-dimensional pattern for each direction, and A mask pattern that extends from a center point by two or more grid points in each direction on the time axis, and one grid point or more in each direction from the center point in both directions along the spatial axis, and further has a predetermined weight value. A voice feature extraction method characterized by performing a mask operation on the voice. 7. The audio feature extraction method according to claim 6, wherein the spread of the mask pattern in the time axis direction is larger than the spread in the spatial axis direction. 8. The audio feature extraction method according to claim 6, wherein the center point of the mask pattern and the weight value in the time axis direction are all "1", and the weight value in the spatial axis direction is smaller than "1".