JPS58189694A - Voice recognition system - 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 The present invention relates to an improvement in a speech recognition method, a method for recognizing continuous word speech uttered by an unspecified speaker.

Conventionally, methods for recognizing speech uttered by unspecified speakers include a learning method that examines the characteristics of the input speech and transforms a standard pattern to match the characteristics, and a learning method that transforms the input speech into a standard pattern to match the characteristics. A normalization method that performs deformation, a multi-standard method that predicts the range of speech deformation due to different speakers and arranges a large number of standard patterns within that range of variation, and a discriminant function method that combines with an appropriate preprocessing method. etc. have been proposed. Among these, the ones that currently have practical level recognition ability are based on the multi-standard method and the discriminant function method. Considering that the ability can be expanded to include hidden and continuous word recognition, the multi-standard method is almost the only realistic method.

However, in continuous word recognition, when considering possible combinations of consecutive words, even if techniques such as the two-stage DP method or the continuous DP method are used, the amount of processing for recognition increases significantly. Therefore, in a system that performs speaker-independent continuous word recognition while maintaining the multi-standard system, the membranes such as pattern and chiming parts become extremely large, making it unrealistic from an economic point of view.

The present invention aims to improve such problems.

Those who are inputting information to a speech recognition device should note that, if they pay attention to a certain use situation, changes such as changing gender from male to female or changing from adult to child cannot occur during use.

That is, in the present invention, the state of the recognition device is divided into a recognition state (state 1) of discrete utterances that requires a relatively small amount of recognition processing;
It is divided into continuous speech recognition states (state 2) that require a large amount of processing. First, the input speech is recognized in state l, and after limiting the personality of the speaker, a standard button of a common set of the personality is used, and state 2 is recognized. The aim is to reduce the amount of processing in state 2 by recognizing the following.

Hereinafter, the present invention will be explained based on Examples.

FIG. 1 shows an example of the configuration of a telephone information service system to which the present invention is applied. Consisting of a system control unit 1, a voice recognition unit 2 according to the present invention, a voice response unit 3, a hybrid coil 4, and a subscriber telephone 5, only the parts necessary for explaining the present invention are extracted from the telephone information service system and written down. It is.

FIG. 2 is a configuration example of a speech recognition device for explaining the present invention. In FIG. 2, a control section 21 not only sends and receives commands and results 27 from the system control section 1 of FIG. 1, but also controls the voice recognition section 2. The similarity calculation unit 23 calculates the degree of similarity between the input voice analyzed by the analysis unit 22 and the standard bang data in the standard bang memory 24, and the continuous bang matching unit 25 calculates the optimum matching value between each standard bang data. is calculated. The result is determined by the determination section 26, and the determination result is sent to the control section 21. Since the configuration of a recognition device that performs continuous bump matching processing is already known (see Japanese Patent Application No. 1982-2205), a description thereof will be omitted. In the example of this device, the input button is always compared with the specified standard button, so if it is known in advance that the input is a discrete utterance, the output of the matching section will be the same as that of the input discrete utterance word. In this case, continuous word input is judged as a continuous word, and the operation of the continuous slam matching section 25 is the same. The following explanation can be applied in exactly the same manner in a system in which the operation of the matching section 25 is switched between discrete vocalization and continuous vocalization (see, for example, Japanese Patent Application No. 158296/1983).

It is assumed that the types of words currently registered are "yes", "no", and numbers from 0 to 9. Further, it is assumed that five types of standard clicks for each word and number, ie, 3×5−15, are registered for each word/number/man/woman/child, taking into consideration the differences among speakers. Taking the example of balance inquiry at a bank, going back to Figure 1, when a call from a user enters the system,
First, the voice response unit 3 asks the user, “Do you want to inquire about your balance?” At the same time, the voice recognition unit 2 recognizes two types of words, “yes” or “no,” as discrete inputs based on commands from the system control unit l. Wait for input at the mote (state l). When the user answers "yes" or "no", the recognition unit 2 only has to compare the two words "yes" and "no" with a total of 30 standard bangs, 15 each. As a result, the best matching standard batan or the male group (or female or child group)
If so, then in state 2 (state of continuous word recognition), the control unit 21 issues a division command so that only standard bangs with numbers belonging to men (or women, or children) are used. In the next step, the voice response device 3 outputs a voice to the user saying, "Please enter your password," and the recognition unit 2 changes to state 2, making it possible to recognize consecutive numbers.

For example, when the user inputs a password such as 1'-1234J by voice, the recognition unit 2 only needs to match it with 10X5-50 number standard buttons belonging to the male (or female, or child) group. Become. Therefore, the matching ability of the recognition unit 3 is sufficient to match at most 50 standard batons. On the other hand, if recognition is performed in state 1 without determining the set, it will be necessary to check with 5 to 150 standard drums of 10XI.

As described above, according to the present invention, it is possible to economically realize a system for recognizing continuous voices uttered by an unspecified speaker, and the effects thereof are significant.

[Brief explanation of the drawing]

FIG. 1 shows an example of the configuration of a telephone information service system to which the present invention is applied, and FIG. 2 shows a block configuration of a speech recognition device according to the present invention. Figure 1 Criminal Figure 2

Claims (1)

[Claims] A first recognition state that recognizes discrete utterances and recognizes continuous utterances using multiple sets of standard patterns prepared for multiple types of speakers with different characteristics for voice patterns having the same meaning. In a speech recognition method having a second recognition state, the characteristics of the input speech are recognized in the first recognition state, and a standard pattern to be used in the second recognition state is determined based on the recognized characteristics. A speech recognition method characterized by limiting the number of pairs.