JPS61235900A - Voice recognition method - 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 Application of the Invention] The present invention relates to a speech recognition method, and particularly to a speech recognition method suitable for efficiently correcting input speech when it is erroneously recognized.

[Background of the invention]

In the speech recognition method, it is desirable that the recognition rate is 100 yen. However, in the current situation, it is inevitable that many erroneous recognitions will occur, especially when recognizing the voices of a large number of unspecified people. Therefore, how to easily and quickly correct misrecognition has become an important technical issue.

For this reason, various correction methods have been proposed. For example, the method disclosed in Japanese Patent Publication No. 47-21921 prepares the most probable pattern, the next most probable pattern, and the next most probable pattern for a certain input voice. If the first pattern is not confirmed to be correct, it is corrected by comparing it with the next pattern.

In addition, the method disclosed in Japanese Patent Application Laid-open No. 58-16296 removes patterns related to misrecognition and patterns judged to be unnecessary from standard speech patterns, and then re-inputs speech from the remaining standard patterns. This method selects a pattern that matches the above and increases the recognition rate upon re-input.

However, each of the above-mentioned conventional methods has disadvantages in that the recognition process is performed only at the time of input, which imposes a large burden on the user during correction, and the correction speed is slow.

[Purpose of the invention]

An object of the present invention is to provide a highly serviceable speech recognition method that can easily and quickly correct erroneous recognition.

[Summary of the invention]

In speech recognition processing, even if the same content of speech is re-inputted at the time of erroneous recognition and the same processing is repeated, there is a high probability that the erroneous US will be used again. In other words, when the service is repeatedly provided, there are many cases in which people misrecognize a specific word as a certain word (for example, the number 1 is always identified as the number 2).

The present invention has been made in view of the above-mentioned points, and if, at the time of erroneous recognition, there are the same words, numbers, etc. that were incorrectly recognized in the past input, the words, numbers, etc. that were corrected in the past will not reach the predetermined recognition rate when re-entering. It is determined whether a predetermined recognition rate has been reached, and if a predetermined recognition rate has been reached, identification is made based on the content.

[Embodiments of the invention]

Hereinafter, one embodiment of the present invention will be described with reference to the drawings. The figure is a block diagram of a voice recognition system to which an embodiment of the present invention is applied. In figure 0, numeral 1 denotes a telephone as an example of a terminal device.

2 is an exchange, 5 is a network control device, 4 is a voice recognition device, 9 is a processing device, and 10 is a voice output unit.

The speech recognition device 4 includes n/4 recognition units 6 for the number of lines n, and aims to make effective use of the recognition units 6. Each recognition unit 6 is connected to the network control device 3 by a complete link 5. Then, each recognition section 6 is connected to an allocation control section 7, and a memory 8 is connected to this allocation control section 7.

The allocation control section 7 connects the line where the input event has occurred to the vacant recognition section 6 to perform recognition processing. If an erroneous recognition occurs in the recognition process, the control unit 7 stores the erroneously recognized numbers and the corrected numbers in the memory 8.

Furthermore, the control unit 7 controls the processing device 9 when an erroneous recognition occurs.
Check the contents of Mesosu 8 for the line specified by , and if there are numbers that were misrecognized in the past and corrected, the recognition rate when re-entering the numbers after correction is within the specified range. Identify whether or not.

As a result, if the recognition rate is within a predetermined range, that number is used as the recognition result even if the recognition rate is not the number one.

With this configuration, the voice input from the telephone 1 through the exchange 2 and the network control device 3 to the voice recognition device fIt4 is recognized by the recognition section 6 selected under the control of the control section 7. If erroneous recognition is made at this time, the control unit 7 examines the contents of the memory 8 and performs correction processing according to instructions from the processing device 9. For example, if the number 1 is incorrectly recognized as the number 2 when the number 1 is input by voice, the control unit 7 checks the contents of the memory 8 and confirms that the recognition rate when the number 2 was corrected to the number 1 in the past is within a predetermined range. It is determined whether or not it exists, and when it is within a predetermined range, Equation 2 is corrected to Equation 1.

Through the above processing, misrecognition that occurs due to the characteristics of the user's voice and its compatibility with the voice recognition device can be quickly corrected.

〔Effect of the invention〕

According to the present invention, when a certain erroneous recognition occurs due to the characteristics of the user's voice (distortion of the frequency of a specific number, etc.) during voice input, it can be quickly corrected, which not only improves the service but also allows calls to be put on hold. Time is saved and system throughput can be increased.

[Brief explanation of drawings]

The figure is a system configuration diagram that uses a speech recognition device to which an embodiment of the present invention is applied.

Claims (1)

[Claims] 1. When input speech is misrecognized, check past correction examples of the misrecognized pattern, and if the recognition rate of the correction is within a predetermined range, output the corrected pattern as the input speech pattern. A speech recognition method characterized by: