JPS6214200A - 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 speech recognition technology, and particularly relates to a method for preventing deterioration in recognition performance due to changes in the emotional state of a speaker.

[Conventional technology]

In conventional speech recognition devices, when comparing data patterns at the time of registration or recognition, only parameters that change relatively little are extracted and matched.

[Problem that the invention seeks to solve]

The conventional speech recognition device described above performs recognition by extracting components that have relatively little influence on changes in the speaker's emotion, but this does not mean that there is no influence, and it is possible to This is one of the reasons why this occurs.

Particularly, when the speaker has a tense BK, the recognition rate drops significantly.

[Failure to solve the problem]

In the recognition device of the present invention, nine components indicating the speaker's tension level are
1. Extracting means from the uttered voice and its share at the time of registration. A means to make a comparative judgment with the time of recognition.

It has means for feeding back the results of comparison and judgment to the speaker.

〔Example〕

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

Figure i1 is a block diagram of one embodiment of the invention.

1 is an audio signal input terminal, 2 is an audio signal pattern data analysis section, 3 is a basic pitch period detection section, 4 is an output signal changeover switch of 2, 5 is an output signal changeover switch of 3, and 6 is an audio Matching processing section using pattern data,
7#i Comparison/judgment unit of the basic pitch period at the time of registration and the basic pitch period at the time of recognition.

Reference numeral 8 denotes a storage unit for storing the pattern data and basic pitch period of the voice signal, and 9 denotes an output terminal for the voice recognition result.

Reference numeral 10 denotes an acoustic transducer for feeding back the comparison and judgment result of the fundamental pitch period in the form of melodies or voice.

Next, the operation will be explained.

In this embodiment, the basic pitch period of the input audio signal is used as a measure indicating the speaker's nervousness.

When registering, switch 4. and the switch 5 are both connected to the storage section 84AllK, and the data extracted from the input voice number 4g by the audio pattern/data analysis section 2 and the basic pitch period detection section 3 are both stored in the storage section 8.

At the time of recognition, the switches 4 and 5 are connected to the matching processing section and the 21PF non-pitch period comparison section, respectively, and the voice pattern data analysis section 2 extracts a V voice pattern from the input terminal l to p voice signal. The data is extracted, and the matching processing unit 6 compares it with the voice pattern data in the storage unit 8 and outputs the recognition result to the output terminal 1.
Output from 0.

At the same time as the above operation, from the input audio signal.

It is extracted by the basic pitch period detection section 3. The basic pitch period data is compared with the data in the storage section 8 by the basic pitch period comparison section 7, and if the change is more than one step, the nervousness of one utterance 8 is relieved. For,! From meal transducer 10.

Output in the form of melodies or voice messages.

〔Effect of the invention〕

As explained above, the present invention provides an approach that is highly dependent on the emotion of a speaker, which is not used in conventional speech recognition devices.
By extracting + from the input speech and feeding it back to the two speakers, there is an effect of improving the average speech recognition rate or shortening the time required for training.

[Brief explanation of drawings]

FIG. 1 is a block diagram of one embodiment of the present invention. DESCRIPTION OF SYMBOLS 1...Audio signal input terminal, 2...Audio pattern data analysis section, 3...Basic pitch cycle detection section, 4...Audio pattern data changeover switch, 5...Basic pitch cycle changeover switch, 6 ...Matching processing section, 7.Basic pitch period comparison section, 8.Storage section, 9.Result output terminal, 10.Acoustic transducer agent, patent attorney Hara Uchi. Sound=S＼＼−0゛′

Claims (1)

[Claims] In a speech recognition device that performs training before performing a recognition operation, in addition to the means of analyzing the speech signal input during the registration period and extracting pattern data, there is also a method of extracting a portion that is correlated with the speaker's tension level, It has means for storing, and further has means for performing pattern matching between the analysis pattern data and input speech during speech recognition, as well as means for extracting components correlated with the degree of tension and comparing and determining them. , specific speaker speech recognition characterized by having a feedback means to relieve the speaker's tension when there is a change of more than a certain amount between a component correlated with the tension level at the time of registration and the same component at the time of recognition. Device.