JPS62141596A - Voice recognition - 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 that allows efficient confirmation of the recognition results of input speech.

[Background of the invention]

For example, JP-A-58-16296 and JP-A-57-1
As described in Japanese Patent No. 50895, correction of erroneously recognized codes is considered in these documents. However, in these cases, one of the input techniques is
If a digit is misrecognized, the system waits for all finger inputs to be completed and then checks and corrects the digit again starting from the first digit. For this reason, there is a problem in that the recognition results cannot be checked efficiently.

[Purpose of the invention]

SUMMARY OF THE INVENTION An object of the present invention is to provide a speech recognition method that can improve speech recognition services by efficiently checking the recognition results of input speech.

[Summary of the invention]

For this purpose, the present invention is designed to reduce the probability of erroneous recognition when the number of digits input is large, and also to prevent users who are likely to get rejected (the approximation value is constant when matching with the standard 7% pattern). Focusing on the high possibility of input processing (when the input audio is below the specified value), the input processing that is normally performed in a series is confirmed and corrected at that point when a rejection or timeout occurs (when the input audio cannot be considered within a certain period of time). It was designed to do this.

In other words, every time an error (reject or timeout) is detected during input of multiple digits, the system checks and corrects the unconfirmed content that has already been recognized at that time, and then inputs subsequent digits. In this way, when all digits have been entered, only the contents that have not been confirmed yet are confirmed.

For users who tend to make mistakes, they are asked to enter multiple digits in parts.

[Embodiments of the invention]

The present invention will be explained below with reference to FIGS. 1 and 2.

FIG. 1 shows the flow of speech recognition processing according to the present invention.

Further, FIG. 2 relates to the present invention. This figure shows the configuration and relay method of an example of a voice response device installed at a voice response service center.

Now, to explain the present invention with reference to FIGS. 1 and 2, when a user wants to receive a voice response service, a voice response service center 27 is called using a telephone 1 installed at home. In the voice response control device 6, the exchange 2
When the trunk section 3 detects a call from a transit user, the voice output device i! outputs an input instruction voice to the user. 5 to output an input instruction voice (processing block 11). After outputting this voice, the voice response control device 6 activates the voice recognition device 7 to start recognition (processing block 12), and also switches the switch 4 of the trunk section 3 to the voice recognition device! ? The device is connected to the side and preparations for audio capture are made.

The voice input from the user is matched with the standard pattern on the standard pattern memory 9 in the voice recognition unit 8, and the one with the most probable approximate value is reported to the voice response control device 6 via the control unit 10 as a recognition result. This is how it is. As a result, the voice response control device 6 determines whether the report is an input report (processing block 13), but if it is an input report, the process includes updating the number of input digits (processing block 14), and storing the input data (processing block 14). Processing block 15) is performed. Such processing is performed every time a digit is input, and is repeated until the input of the previous digit is completed.

Here, if the input voice for a certain item from the user is 0 digits, it will be rejected during input, for example at the 3rd digit,
Or, assuming that a timeout is detected, the voice response control bag v! is sent immediately after the third digit is input. 1
6 is now reported as an abnormality. If there is an abnormality report in the voice response control device 6, the voice output device i! 5 to output a confirmation voice for the unconfirmed content up to the third digit that has already been input (processing block 1
7) It has become a thing. Judgment processing (processing block 18)
If the confirmation result is normal, the number of confirmed digits is set (processing block 20) and the audio output device 5 is instructed to output the next input digit. Voice recognition for purpose! The recognition start is activated for the A position 7 (processing block 12). On the other hand, if the confirmation result is incorrect, after the digit-by-digit confirmation/correction process (processing block 19) is performed for unconfirmed items until the number of input digits is low, processing is performed in the same way as when the confirmation result is normal. It has become something that

If all digits have been input in the determination process (processing block 16) through the process described above, then a determination is made as to whether or not a code that has already been confirmed exists (processing block 22). It has become.

If a confirmed code exists, the voice output device 5 is instructed to output a voice confirming the input code remaining after excluding the confirmed code from all digits (processing block 23).
If there is no confirmed code, the voice output device 5 is instructed how to output a voice confirming all input codes (processing block 24). In either case, confirmation from the user is required. After waiting (processing block 25), processing according to the contents is performed. That is, if there is an error in the confirmation content, the confirmation correction (block 26) is performed for each digit.
) to establish it as the final input code, and if there is no error, it is established as the final input code as is.

As mentioned above, when inputting multiple digits, for users who are easy to input, input them all at once and then check all digits, and for users who are difficult to input, inputting multiple digits will result in digits being input. Since the verification and correction is performed after being divided, the verification and correction of the speech recognition result can be performed efficiently, and serviceability is improved.

〔Effect of the invention〕

As described above, according to the present invention, the recognition results of input speech can be checked efficiently.

This has the effect of improving voice recognition services.

[Brief explanation of drawings]

FIG. 1 is a diagram showing a flow of voice recognition processing according to the present invention, and FIG. 2 is a diagram showing a configuration and relay method of an example of a voice response device according to the present invention. DESCRIPTION OF SYMBOLS 1...Telephone, 2...Switchboard, 3...Trunk section, 4...Switch, 5...Audio output device, 6...
Voice response control device, 7... Voice recognition device, 8... Voice recognition unit, 9... Standard pattern memory, 10... Control unit, 27... Voice response service center.

Claims (1)

[Claims] In a voice recognition method that inputs voice for each item at once using a voice response device and then confirms it, each time an abnormality is detected in the input voice during input, only the unconfirmed results among the already recognized results are checked and corrected. After performing this, inputting subsequent digits is continued, and when all digits have been inputted, only unconfirmed results among the already recognized results are confirmed.