JPH0330159B2 - - Google Patents

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a speech recognition method, and more specifically, to a novel method that can obtain a high recognition rate using vector field patterns and its implementation. It provides the equipment to be used.

[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 standard patterns 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 sound path 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 for this purpose, it is possible to improve 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 that the invention seeks to solve]

Conventional methods that use the spatio-temporal pattern of the scalar field as a feature do not necessarily achieve a sufficient recognition rate when targeting large vocabularies or unspecified speakers. A large number of 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 unspecified speakers and large vocabularies has stalled. Therefore, one of the inventors of the present invention obtained a spectral vector field pattern by differentiating the time-frequency spatiotemporal pattern in the spectral space of a scalar field, and used this pattern as a feature. However, this application takes this method one step further from an engineering perspective, and proposes an improved speech method that is easy to calculate, can be performed in a short time, is suitable for practical use, and has a higher recognition rate. The purpose of this invention is to provide a recognition method and a device used for its implementation.

[Means for solving problems]

The speech recognition method according to the present invention extracts a feature pattern from a speech signal input as a recognition target,
Matching the characteristic pattern with the standard pattern,
In a speech recognition method that identifies input speech, an audio signal is analyzed to obtain a spatiotemporal pattern of a scalar field defined by a time axis and a spatial axis, and each grid point in space is determined by spatially differentiating the spatiotemporal pattern. Convert it to a vector field pattern with magnitude and direction,
Quantize the direction parameter of the vector of the vector field pattern into N values (N: integer),
This quantized value is separated into vectors having the same value, and N two-dimensional patterns for each direction are created with the size of the vector as the value of each grid point, and the two-dimensional patterns for each direction are used as the characteristic pattern. It is characterized by:

[Effect]

The input audio signal is quantized from the spatiotemporal pattern of the scalar field defined by the time and space axes, and the vector direction parameters are quantized and converted into multiple directional two-dimensional patterns separated for each quantized direction. As a result, this pattern is recognized as a characteristic pattern. This pattern is composed of the spatial differential of the spatiotemporal pattern, that is, the spatiotemporal change information, so it represents speech phonology well and is not easily affected by speaker fluctuations, etc. Also, by quantizing the direction parameter, the vector field absorb fluctuations in Furthermore, complex number operations that must be performed when the vector field pattern itself is used as a feature pattern are no longer necessary, and calculations are simplified.

〔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 section spectrally analyzes the audio signal and uses a spectrum with the frequency axis as the spatial 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 has multiple channels (for example, 10
~30) are input to the analysis section 2 which is formed by connecting bandpass filters 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 described above as an analysis unit that divides the audio signal into frequency bands, but a fast Fourier transformer may also be used.

Now, the method of the present invention is characterized by the feature extractor described below. The input pattern to the feature extractor 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, and x is a channel number of a bandpass filter or a number specifying a frequency band. 1≦t≦T, 1≦t≦L).

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 23 frames). Specifically, in the case of M≦T, the normalized spatiotemporal pattern F(t, x) is obtained by the following equation (1).

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

Claims (1)

[Claims] 1. A spatiotemporal pattern of a scalar field defined on a time axis and a spatial axis is obtained from an audio signal input as a recognition target, and a feature pattern based on the spatiotemporal pattern is matched with a standard pattern. , in a speech recognition method that identifies input speech, the speech signal is analyzed to obtain the spatiotemporal pattern of the scalar field, and the spatiotemporal pattern is spatially differentiated to generate a vector field having a magnitude and direction at each grid point in the space. For the vector of the vector field pattern, its direction parameter is N
quantize it into a value (N: an integer), separate this quantized value into the same vectors, and create N two-dimensional patterns for each direction with the size of the vector as the value of each grid point, A speech recognition method characterized in that the two-dimensional pattern for each direction is used as the characteristic pattern. 2 Obtain a spatio-temporal pattern of a scalar field defined by the time and space axes from the audio signal input as a recognition target, match the feature pattern based on the spatio-temporal pattern with a standard pattern, and identify the input audio. The speech recognition device includes: an analysis unit that analyzes an input speech signal to obtain a spatiotemporal pattern of the scalar field; a normalization unit that normalizes the spatiotemporal pattern with respect to time; and a spatial differentiation of the normalized spatiotemporal pattern. a vector field extraction unit that extracts a vector field pattern having a magnitude and direction at each grid point in space; a direction-specific two-dimensional pattern creation unit that separates each vector having the same conversion value and creates N direction-specific two-dimensional patterns with the magnitude of the vector as the value of each grid point; A speech recognition device characterized in that an output of a direction-based two-dimensional pattern creation section is used as the characteristic pattern. 3. Obtain a spatio-temporal pattern of a scalar field defined by the temporal and spatial axes from the audio signal input as a recognition target, match the characteristic pattern based on the spatio-temporal pattern with a standard pattern, and identify the input audio. The speech recognition device includes: an analysis unit that analyzes an input speech signal to obtain a spatiotemporal pattern of the scalar field; and a spatial differentiation of the spatiotemporal pattern for each of a plurality of frames in the time axis direction sequentially output from the analysis unit. a vector field pattern extraction unit that sequentially extracts a vector field pattern; and a vector that quantizes a direction parameter of the sequentially extracted vectors of the vector field pattern to an N value (N: an integer) and has the same quantized value. a first direction-specific two-dimensional pattern creation unit that sequentially creates N direction-specific two-dimensional patterns with the magnitude of the vector as the value of each grid point; Find the average value of two-dimensional patterns for each direction of multiple frames that are sequentially created in a separate two-dimensional pattern creation section, or
Select one pattern from among the dimensional patterns,
and a second direction-based two-dimensional pattern creation section that sequentially creates direction-specific two-dimensional patterns for one frame, and the output of the second direction-based two-dimensional pattern creation section is configured to be used as the characteristic pattern. A speech recognition device characterized by: