JPS5988798A - Voice recognition processing 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] [Field of Application of the Invention] The present invention is applicable to each of a plurality of sets of standard speech buttons prepared for each word to be recognized, which have the greatest similarity to the main speech. Is there a reliable voice analysis of the input voice in a voice recognition device that makes a judgment and outputs it as a complete recognition result? The present invention relates to a voice recognition processing method for improving the recognition efficiency of the device by performing voice recognition processing after the recognition.

[Prior art]

In this type of speech recognition device, the conventional speech recognition processing method only asks the speaker about the recognition result? My friend did not let me confirm any of the voice analysis results of Daiki's voice.

However, if the voice analysis of powerful voices is not performed reliably, even if the subsequent recognition process is performed thoroughly and reliably, and the recognition results are checked, a good recognition rate will not be obtained in the end. Not only that, but the serviceability deteriorated due to repeated input.

[Purpose of the invention]

An object of the present invention is to eliminate the above-mentioned drawbacks of the prior art, perform voice recognition processing after performing reliable voice analysis on input voice, improve the recognition rate, and provide a voice recognition processing method with good serviceability. be.

[Summary of the invention]

The configuration of the speech recognition processing method according to the present invention is to store a plurality of sets of standard speech button data corresponding to each word to be recognized, perform a speech analysis on the powerful speech and extract the characteristic data, In a speech recognition device that has a function of performing a bang matching process between the feature data and each of the above standard voice bang data, and determining and outputting the one with the highest degree of similarity as a recognition result, Based on the results of the analysis, the voice corresponding to the input voice is synthesized and debaked using the same method, and then the subsequent voice recognition processing is performed based on the results of the above-mentioned voice analysis. , or performs control processing to force the re-input of the audio.

In short, after confirming that the voice analysis of the input voice was appropriate, a predetermined voice recognition process is performed based on the result to improve the recognition rate.

[Embodiments of the invention]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of a speech recognition apparatus using a speech recognition processing method according to the present invention, and FIG. 2 is a flowchart thereof.

Here, 1 is a microphone related to human voice input, 2 is an input unit that performs gain adjustment, band limitation, and other necessary preprocessing on the input audio signal, and then converts it into a digital signal. 3
4 is an analysis unit that performs audio analysis of large-scale audio based on the input digital audio signal and extracts characteristic data;
5 is the voice section detection ■S where the voice section of the input voice is detected according to a predetermined threshold for voice section detection, and 5 is the voice where the bang matching process (similarity calculation process) between the input voice and each standard voice bang is performed. A recognition unit 6 is a determination unit that determines the ranking of each degree of similarity for the powerful voice based on the processing result, 7
is a standard back door button memory that stores (memorizes) multiple sets of standard voice button data for each word to be recognized; 8 is a standard voice button selection unit that controls the selection; 9 is a large voice analysis result/recognition unit. The speech synthesis unit 10 is responsible for displaying and confirming results, giving voice instructions and other necessary displays and instructions.
In the speaker, 11 is a console unit for displaying and confirming input voice analysis results and recognition results, voice commands, and other necessary display and operations; 12 is a control unit for controlling each of the above units and other necessary processing; and 13 is a host device that performs desired service processing based on voice recognition results.

First, prior to voice recognition processing, the control unit 12Ff prepares for human voice input unit 21, analysis unit 3. Instructs the speech section detection unit 4 and also indicates the classification of the word to be recognized at that time (for example, number, service type name, product name,
A standard voice button selection unit 8 is configured to select all sets of standard voice buttons (classifications such as place names, etc.) from the standard voice button memory 7.
(Process 21 in FIG. 2).

Once these preparations are complete, the f! The sound is emitted from speaker 0 via 159 (same process 22).

・4 As a result, when a voice is input from the speaker or the microphone 1 (same processing 23), the input unit 2 performs digital conversion, etc., and the analysis unit 3 performs voice analysis on the digital voice signal and determines the relevant characteristics. Data etc. are extracted (processing 24).

In order to have the voice itself confirm whether or not the result of the voice analysis of the powerful voice was appropriate, the control unit 12 conversely analyzes the input voice using the same method based on the result of the voice analysis (feature data, etc.) iC in analysis s3. A confirmation voice corresponding to is synthesized (for example, PARCOR voice synthesis is performed based on the PARCOR voice analysis result), and this voice is emitted from the speaker 10 (process 25).

The speaker listens to this a lot and decides whether or not it is appropriate for his or her own voice input. In the case where the head is "I" or has fallen off and there is only "J", the judgment/confirmation result is input from the console section 11 (same process 26). This may be performed by voice input and the voice recognition may be performed.

As a result, if you enter an input that is inappropriate,
Regulation? The 1111 unit 12 issues a reminder to input the voice again (process 27), and repeats the same process as described above.

By human power (or by unmanned power within a predetermined time) that the result of Kamigumi voice analysis was appropriate, the voice recognition unit 5 takes the characteristics of the input voice and the selected Yanagi-kei voice. A slam matching process is performed on the basis of the bang data, and the degree of similarity of each upper Hα standard tone self-pattern to the input voice is transmitted to the determining unit 6 (same process 28).

Judgment unit 6 has the highest degree of similarity (acids are also likely)
Convey the thing to system #1 section 12 as an import result (N processing 2
9).

-C@ also has a low probability similarity value with respect to the input sound j, and in the case of a misjudge, it is doubtful to output Somu as the R1a result, the control unit 12 selects the standard sound button selection capital 8. The user is instructed to select the same button as before (Iwl process 32), and a message is emitted from the speaker 10 via the voice synthesis section 9 to remind the user to input again.

If the recognition result is not rejected, the control unit 12 sends a confirmation request Mevse V from the speaker 10 via the speech synthesis unit 9 as a display for the speaker to confirm whether or not the recognition result is correct. Make a sound (same process 30).

Note that the above display may be a lamp display on the console section 11 or the like.

The speaker listens to this, learns whether his or her own input voice was recognized correctly or incorrectly, and outputs the confirmation result from the console unit 11 to the control unit 12 (process 31).
).

The above-mentioned confirmation result input to the control s12 does not necessarily have to be performed by an operation on the console unit 11, and may be performed by a powerful confirmation voice from the microphone 1. In this case, it is desirable that each of the steps be simple and difficult to misrecognize so as to ensure proper recognition, and the same applies to the above-mentioned 7. Process 26.

If the above-mentioned recognition candidate is correct according to the confirmation information in the second part, the control unit 12 sends it to the host device 13 as a recognition result, finishes the processing for one input voice, and then starts the next one. Prepare for input.

-If you receive confirmation that it was a misidentification,
Processes 32 and 33 are performed in the same way as in the case of the above-mentioned project, and this process is repeated until correct recognition is obtained. When correct recognition is obtained, the recognition result is sent to the host device 13 in the same manner as described above. and the series of processing ends.

In this way, after having the speaker confirm the synthesized speech that is compared with the input speech based on the results of speech analysis,
Since subsequent speech recognition processing is performed based on the result of the speech analysis, it is possible to prevent erroneous recognition caused by inappropriate speech analysis.

[Effects of the Invention] As explained above in detail, according to the present invention, the sound is generated after performing a reliable voice analysis on the input voice.

Since voice recognition processing is performed, the recognition performance of this type of voice recognition device can be greatly improved, and its reliability.

A remarkable effect can be obtained in improving serviceability.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an embodiment of a speech recognition device using a speech recognition processing method according to the present invention, and FIG. 2 is a flowchart thereof. DESCRIPTION OF SYMBOLS 1...Microphone, 2...Input section, 3...Analysis section, 4...Speech section detection section, 5...Speech recognition section,
6... Judgment unit, 7... Standard voice button memory, 8.
...Standard voice button selection section, 9...Speech synthesis section, 10
...Speaker, 11...Console section, 12...
Control unit, 13... host device.

Claims (1)

[Claims] 1. Store all multiple sets of standard voice button data corresponding to each word to be recognized, perform voice analysis on the input voice to extract the feature data, and combine the feature data with the above standard voice button data. A speech recognition device that has the function of performing slam matching processing and determining and outputting all the recognition results with the highest degree of similarity, based on the results of voice analysis of the human voice, uses the same method to After synthesizing and debating the voice corresponding to the voice, based on the results of the confirmation to 7n, perform subsequent voice recognition processing based on the result of the voice analysis, or control and process so as to cause the voice to be re-entered. This is a voice recognition processing method with all the features.