JPS61183696A - Specified speaker voice recognition equipment - Google Patents

Abstract

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

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a speech recognition device by matching acoustic speech signals, and more particularly to a speech recognition device for a specific speaker.

[Conventional technology]

Conventionally, in this type of specific speaker speech recognition device, speech analysis patterns of a specific speaker are registered in advance, and recognition operations are performed on these fixed speech data patterns during recognition. Problems to be Solved] The conventional speaker-specific speech recognition device described above performs speech recognition operations on fixed speech data patterns, so it is difficult to recognize changes in the uttered speech due to changes in the speaker's emotions, changes in the environment, etc. There was a drawback that the recognition rate decreased.

[Means for solving problems]

In the specific speaker speech recognition device of the present invention, if a speech data pattern that satisfies the recognition judgment condition is found among the registered speech data patterns during recognition operation, the latest input speech data and the recognized registered speech data are used. It has means for generating and re-registering a new audio data pattern from the existing audio data pattern and the correspondence on nine time frames obtained by dynamic pattern matching.

〔Example〕

Next, the present invention will be explained with reference to the drawings.

FIG. 1 is a block diagram of one embodiment of the present invention.

A voice input terminal 1 is connected to a voice analysis section 2.

The speech analysis section 2 divides the signal into a plurality of frequency components using a bandpass filter bank every seven frames, and outputs the divided frequency components to the analysis signal output line 3.

The analysis signal output line 3 transmits the analyzed signal to the analysis signal buffer 4 and the distance calculation section 7. At the time of first voice registration, the buffer output line 5 is passed from the analysis signal buffer 4,
The analyzed signal is registered in the storage unit 6.

Next, during the recognition operation, the distance calculation section 7
A dynamic pattern matching calculation is performed on the input signal and the registered voice data pattern obtained by passing through the entire memory unit output line 8, and each distance between each registered voice pattern and the analysis signal of the input voice is calculated. is determined and output to the comparison/judgment section 10 through the distance output line 9. The comparison/determination unit 10 determines the registered voice data pattern having the minimum distance less than or equal to a preset recognition determination value as the recognized voice data pattern, and outputs the recognized voice data pattern from the recognition result output line 11 .

Next, during the re-registration operation, the distance calculation unit 7 performs the dynamic pattern matching calculation again using the recognized voice data pattern specified from the recognition result output line 11 and the analysis signal in the analysis signal buffer as input, and compares the two The time frame corresponding output line 13 shows the time frame correspondence of the two input signals.
The data is output to the new audio data pattern calculation unit 12 through t-. The new audio data pattern calculation unit 12 generates a new audio data pattern from the data indicating time frame correspondence of the output line 13, the recognized audio data pattern, and the analysis signal, and re-registers it through the new audio data pattern registration line 14t. I do.

Thereafter, similar processing is repeated during recognition operations.

FIG. 2 is a diagram showing the time frame relationship between the audio data pattern and the analysis signal in the embodiment of FIG. 1. 1
5 is a registered voice data pattern.

16 to 27 indicate each time frame. Reference numeral 41 indicates an analysis signal, and 42 to 49 indicate its respective time frames. 28
40 is a line showing the correspondence relationship between the two.

The procedure for generating a new audio data pattern is shown below.

First, the arithmetic mean of the same frequency components is calculated for all time frames in a corresponding relationship, and this is used as a temporary new audio data pattern.Next, the number of time frames is determined for this temporary new audio data pattern.

Next, time frames are thinned out from the temporary new audio data pattern so that this number of time frames is the same as the arithmetic mean value of the number of time frames of both original data. The time frame shall be as follows.

〔Effect of the invention〕

As explained above, the present invention re-registers the recognized input speech signal as a new speech data pattern during speech recognition, so that the latest speech data pattern is always maintained and changes in the speaker's utterance over time are realized. This has the effect of improving the recognition rate.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, and FIG. 2 is a diagram showing the correspondence of time frames. 1...Audio input terminal, 2...Audio analysis section, 3...Analysis signal output line, 4...Buffer, 5...... Buffer output line, 6...
・Storage section, 7...-Distance calculation section, 8...
Memory unit output line, 9...Distance output line, 10...-
. . . Comparison/determination section, 11 . . . Recognition result output line,
12... New audio data pattern calculation section, 13...
... Time frame compatible output line, 14... New audio data pattern registration line, 15... Audio data pattern, 16-27... Time frame, 2
8-40... Line showing correspondence, 41...
Analysis signal, 42-49...time frame. =1・≦−kopa・,

Claims (1)

[Claims] An analysis unit that divides an input audio signal into multiple frequency components and time frames of a constant period, a storage unit that registers multiple audio data patterns of a specific speaker, an inter-pattern distance calculation unit, and an inter-pattern distance calculation unit a determination unit that determines audio data with the highest degree of similarity; a recognized audio data pattern, a signal obtained by analyzing the input audio signal, and a time frame correspondence relationship between the two obtained by the inter-pattern distance calculation unit; 1. A speech recognition device comprising: means for generating a new speech data pattern from a given speech data; and means for re-registering the new speech data pattern.