JPH0256680B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field of the Invention The present invention relates to a voice pattern file management/editing method in a system using a voice recognition device.

Background of the technology In order to use this type of conventional system, each user registers in advance the information that should be output when the user's voice is correctly recognized. Even if the information to be output is common to many users, it is necessary to manage it individually. (For example, Shuzo Saito: Fundamentals of speech information processing, Ohmsha,
1981.11) Conventional technology and problems In the conventional method, as the amount of registrations increases, users have to register more themselves, the number of registrations increases,
This has resulted in disadvantages such as an increase in the burden on users and an increase in the amount of files managed by the system.

Furthermore, in voice recognition using already registered voice patterns, if there are many false recognitions due to changes in the voice uttered by the speaker over time, etc., there is a drawback that the user must re-register the voice patterns.

Purpose of the Invention In order to solve these drawbacks, the present invention prepares information that can be shared together with voice patterns in a system,
In addition to making it possible for users to retrieve the sound without registering it in advance, the system also allows users to manage recognition accuracy and manipulate sound patterns to eliminate the need to re-register sounds due to an increase in misrecognitions.The drawings are detailed below. explain.

Embodiment of the Invention The figure shows a specific configuration example of an embodiment for explaining the audio pattern editing method according to the invention, in which 1 is an audio input terminal, 2 is an audio analysis section, 3 is an input audio pattern memory, and 4 is a recognition unit. 5 is a processing unit, 5 is a control unit, 6 is an input/output interface, 7 is an audio pattern file, 7
1 and 72 constitute the audio pattern file 7, 71 is a common audio pattern file, 72
8 is an individual voice pattern file, and 8 is an output information file registered corresponding to the voice pattern. Note that the speech analysis section 2 is an LSI-based speech analyzer that is currently available to the general public, and the recognition processing section 4 and control section 5 are ordinary microprocessors.

When speech is input from the speech input terminal 1, it is converted into an input speech pattern by the speech analysis section 2 and stored in the input speech pattern memory 3. At the same time as the storage is completed, the control unit 5 saves the portion of the individual voice pattern file 72 corresponding to the speaker and the input voice pattern memory 3.
instructs the recognition processing unit 4 to compare the contents of
As a recognition result by the recognition processing unit 4, an identification number indicating which voice pattern it is and a recognition processing result of recognition accuracy, that is, similarity in recognition are received. The control unit 5 requests the speaker to confirm the recognition result received from the recognition processing unit 4 through the input/output interface 6, or checks the validity of the recognition result according to criteria determined in advance by the system. , if it is determined to be correct, searches the output information file 8 using the identification number, and outputs the output information as a result of the search to the outside through the input/output interface 6.

In the voice pattern file management/editing operation described above, if it is determined that there is no corresponding answer in the individual voice pattern file 72 as a result of voice recognition, the control section 5 instructs the recognition processing section 4 to read the input voice pattern memory 3. Common audio pattern file 71 for comparison with the content
The recognition processing unit 4 is instructed to perform the comparison between the two and receives the comparison results. In the control unit 5, if the recognition result is determined to be correct by the confirmation means similar to the above-mentioned operation, the output information file 8 based on the identification number is
output the search results to the outside, and add the identification number to the contents of the input voice pattern memory 3,
It is additionally registered in the portion corresponding to the speaker of the individual voice pattern file 72. If it is determined that there is no corresponding solution in the common speech pattern file 71, the speaker is notified of this and the process is terminated.

In the voice pattern file management/editing operation described above, when a correct answer is obtained from the individual voice pattern file 72 and the result is notified to the speaker, the control unit 5 uses the recognition output from the recognition processing unit 4 as a recognition result. The accuracy is statistically processed retrospectively, and if the result is lower than the judgment criteria predetermined by the system, the audio pattern selected as the correct answer is manipulated (for example, replaced with the input audio pattern), and the individual audio Update the pattern file 72.

That is, if the input voice pattern of the user who is the speaker is found to match in the individual voice pattern file 72, the input voice pattern of the user is traced back to the past of the corresponding voice pattern in the individual voice pattern file 72. For example, when the statistical value of the variation in recognition accuracy of similarity in recognition and the value of similarity in recognition when there is a matching solution exceed a predetermined threshold, for example, the individual speech pattern file 7
The voice pattern selected as the correct answer is replaced with the individual voice pattern of the corresponding input voice pattern in 2, and the individual voice pattern file 72 is updated.

Note that in the configuration of this embodiment, an input speech pattern memory is provided as a means for extracting input speech patterns, but in other embodiments, a means for storing and reproducing input speech is provided, and the speech recognition device recognizes the input speech pattern. After the process is complete, switch to registration mode, create and retrieve the input audio pattern using the playback sound, prepare two speech recognition devices, set one to recognition mode and the other to registration mode, and retrieve the input audio pattern. be.

Effects of the Invention As explained above, by collectively managing information common to many users in a system and having files with corresponding audio patterns, the common information can be used without the need for users to register in advance. This has the advantage of improving user operability.

Additionally, since common information is managed collectively in the system, duplicate information can be removed and the file size can be reduced.

In addition, by managing the recognition accuracy when a correct answer is obtained, it is possible to automatically detect a decrease in the recognition rate of speech recognition due to existing speech patterns, and further compare the recognition accuracy with a reference value to manipulate the existing speech patterns. This has the advantage of omitting the user's voice re-registration.

In addition, by providing a means to extract the input voice pattern, it can be used when automatically generating a personal voice pattern, or when replacing the input voice pattern as a voice pattern operation in response to changes in existing personal voice patterns over time. This has the advantage of omitting the need for the person to re-speak.

[Brief explanation of drawings]

The figure is a configuration diagram of an embodiment of the present invention. DESCRIPTION OF SYMBOLS 1... Audio input terminal, 2... Audio analysis section, 3... Input audio pattern memory, 4... Recognition processing section, 5... Control section, 6... Input/output interface, 7... Audio pattern file, 71... Common audio pattern file, 72
...Individual audio pattern file, 8...Output information file.

Claims (1)

[Claims] 1. The speech to be recognized is patterned in advance, the speech uttered by the user is patterned, and the degree of similarity is compared for each of the speech pattern groups that have been patterned in advance. and, as a result of the comparison, a pattern with the highest degree of similarity is output as a recognition result.In a speech pattern editing method in a speech recognition system, the pattern having the highest degree of similarity is output as a recognition result. A voice pattern file consisting of a common voice pattern file as a voice, a voice pattern file consisting of an individual voice pattern file of the voice uttered by an individual user, and an input voice that analyzes the voice uttered by the user and converts the voice-recognized input voice into patterns and temporarily stores them. comprising a voice pattern storage means, extracting a patterned input voice pattern uttered by a user from the input voice pattern storage means, and extracting the extracted input voice pattern uttered by the user and the user's individual voice of the voice pattern file; information to be output when the retrieved input voice pattern uttered by the user does not have a match in the individual voice pattern file of the individual and there is a match in the common voice pattern file. A voice pattern editing method comprising: associating the input voice pattern with an input voice pattern, and adding the associated input voice pattern to an individual voice pattern file of the individual. 2 The speech to be recognized is patterned in advance, the speech uttered by the user is patterned, the similarity is compared for each of the voice pattern groups that have been patterned in advance, and as a result of the comparison, A voice pattern editing method in a voice recognition system that edits a voice pattern output that outputs a pattern with the highest degree of similarity as a recognition result, comprising: a common voice pattern file as a common pattern group of all voices to be recognized; , a voice pattern file consisting of an individual voice pattern file of the voice uttered by an individual user, and an input voice pattern storage means that analyzes the voice uttered by the user, converts the voice-recognized input voice into a pattern, and temporarily stores it. , extracting a patterned input voice pattern uttered by the user from the input voice pattern storage means, and comparing the extracted input voice pattern uttered by the user with the user's individual voice pattern file of the voice pattern file; and information to be output when the retrieved input voice pattern uttered by the user does not have a match in the individual voice pattern file of the individual and there is a match in the common voice pattern file, and the input voice pattern. Add the matched input voice pattern to the individual voice pattern file of the individual, and if the retrieved input voice pattern uttered by the user matches the individual voice pattern file of the individual, The input voice pattern uttered by the user is calculated based on the statistical value of the variation in similarity in the past recognition of the corresponding voice pattern in the individual voice pattern file, and the recognition in the case where there is a matching solution. A voice pattern editing method characterized in that when a similarity value exceeds a predetermined threshold, the corresponding input voice pattern in the individual voice pattern file is replaced with an individual voice pattern.