JPS61175696A - Voice recognition responder - 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 [Technical Field] The present invention relates to a voice recognition response device, and particularly to a voice recognition response device that is used by unspecified speakers and requires reliable voice recognition.

[Prior art] Conventionally, perfect speech recognition has not been possible with this type of voice recognition response device, and in applications where recognition reliability is particularly required, it is necessary to use a voice response device or the like to output the recognition results as a response. and asked for confirmation.

As an example, for a balance inquiry in a banking service, the voice recognition response procedure is as shown in Table 1.
Margin table below 1 In this way, the customer side input and the bank side response are performed alternately,
In this process, the recognition result is also confirmed as ``yes'' or ``no''. That is, if the confirmation of the recognition result is positive, the process advances to the next step and a new input and response are performed.
In case of negative results, the same process was repeated and the next step could not proceed until it was recognized correctly.

However, simply increasing the number of recognition times does not always result in correct results; in fact, if the input cannot be recognized after 2 or 3 inputs, it often remains unrecognized no matter how many times the voice recognition response device is input. It was developed to free people from tedious key operations, but in reality it cannot be used by anyone, and if recognition is not successful, the user must input the voice over and over again until it is recognized correctly. However, it was thought that it would actually require more time and effort from the speaker.

[Objective] The present invention has been made in view of the above-mentioned drawbacks of the prior art, and its object is to provide a solution to the case where voice is not recognized no matter how many times it is input.

Another object of the present invention is to provide a voice recognition and response device that can sufficiently handle speakers with strong personalities who are difficult to recognize with this type of device.

[Embodiments] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a block diagram of a voice recognition response device according to an embodiment of the present invention. In FIG. The feature extractor 4 extracts the feature parameters of the audio signal, and is composed of a microprocessor including RAM and ROM, and includes a central processing unit (C) that recognizes input audio and performs audio response control based on it.
PU), 5 is a speaker from which the device outputs audio, and 6 is an external device (cash dispenser) that operates using the recognition result of the water device.

Inside the CPU 4, various functional blocks are shown that are realized by executing programs. 7 is a first recognition unit that recognizes inputs of predetermined types of sounds, such as "zero" to "cue" and "high", and 8 recognizes only specific sounds, such as "high", with high performance and high recognition rate. A second recognition section, 9 is a control section that advances control based on the recognition result, 10 is a voice response section that synthesizes and outputs a voice response signal, and 11 is a message memory that stores voice data of a response message.

FIG. 2 is a diagram showing the contents of the message memory 11. The message memory 11 stores various grouped messages, 12 is a guidance message storage area that stores guidance messages for the speaker, and 13 stores confirmation messages for input voice recognition results. Area 14 is an area that stores a suggested message to be proposed from the device to the speaker if recognition is not successful.

FIG. 3 is a flowchart illustrating the operating procedure of the embodiment. In step S1, a retry counter RC for counting the number of times that recognition has failed is initialized to 0, and an index register i for the guidance area is initialized to 1. In step S2, it is connected to the switch 5Wt-1 side. This is to recognize and identify a predetermined type of input voice. In step S3, the guidance message is accessed using the contents of the index register i (initially 1) and output to the speaker 5.

That is, a voice message saying "Please enter the first digit of your PIN number" is output. In step S4, a voice input from the speaker is waited for. If there is voice input, the process proceeds to step S5 where voice recognition is performed and the sign (number) of the recognition result is stored in register j. Step S6
Then, the confirmation message is accessed using the contents of register j (for example, 3) and output to the speaker 5. In other words, "Is it San?" is outputted as a voice. In step S7, the speaker waits for a reply, and when there is a reply, the input voice is recognized in step S8.
In step S9, it is determined whether "Yes" or not.

If "yes", it is confirmed that the recognition in step S5 is correct. The retry counter Re is set to 0 in step SIO, the index register i is incremented by 1 in step S11, and the code of the recognition result j is sent to the external device 6 in step S12. In step 513, it is determined whether the index register i is at the maximum (the number of input digits of the password is satisfied), and if it is satisfied, the process is terminated, and if it is not satisfied, the process returns to step S2 and the next guidance message is output.

Next, the process proceeds from step S3 to step S9 in the same manner as above, and if the speaker's response is not "yes", it means that the recognition result j was incorrect. ”, address A-
Q (8) Speaker 0 who outputs the voice "Is it a bee?" is wrong, so input "No". Since it is not "high", the flow advances to step S14, where the retry counter RC is incremented by one. In step 315, the retry counter RC is checked, and if the content is not 2, the process returns to step S2. In this way, in the embodiment, the voice is re-input and confirmed using the same method only once.

If it is determined in step S15 that the retry counter RC=2, the process proceeds to step 516 and the switch SW is switched to the 2 side. This is to use the second recognition section. The second recognition unit 8 of the embodiment is configured to be able to recognize only "yes" from all types of speakers with high performance and high reliability. In this sense, the first recognition section and the second recognition section are shown separately in this embodiment.

Now, in step S17, the content of the proposal counter is initialized to 1. In step 518, the proposed message is accessed based on the contents of the proposal counter and output as audio to the speaker 5. That is, "If it's 1, please answer yes" is outputted as a voice. In step S9, a reply from the speaker is awaited. In step S20, the voice input of the reply is recognized. In step S21, it is checked whether the recognition result is "high". If "yes", the phonetic digit that the speaker wanted to input is equal to the content in the proposal counter, so in step S24 the content of k is transferred to index register i, and the process proceeds to step S10.
This is to enter the next digit of the password.

Further, if it is not "high" in step S21, it means that the number is not the number intended by the speaker. The flow is step 522
Proceed to step S2, add 1 to the proposal counter k, and proceed to step S2.
3, it is determined whether the proposal counter k is the maximum. If it is not the maximum, the process returns to step 318 and proposes the next number, and if it is the maximum, the process returns to step 317 and starts from 1.

In the above-mentioned embodiment, the first recognition unit indicates "high" or "
If "No" can be recognized with high performance, there is no need to separately provide a second recognition section.

In addition, if it is difficult for the second recognition unit to recognize “hi” or “no” with high performance, a simple method that simply detects the presence or absence of voice (or sound) instead of recognizing “yes” or “no” may be used. It may be confirmed by a reliable method.

Furthermore, in the embodiment described above, the control unit counts the number of errors when the voice input by the speaker is incorrectly recognized, and changes the question-and-answer format when the number of errors exceeds a certain value (two times in the embodiment). Ta. In this case, the recognition result response for the speech input by one speaker is outputted to the speaker for confirmation, with the word with the highest score (similarity) as the first candidate. As a result of outputting the gl candidate as a recognition result and recognizing the confirmation voice,
If the number is incorrect, instead of requesting the speaker to input the number aloud again, the device immediately makes a suggestion response using the second and third choices in the device, making the operation more efficient. I can do it.

[Effect J As described above, according to the present invention, the number of erroneous recognitions is counted, and if the number exceeds a certain value, the process moves from the voice input/recognition-response confirmation process to the proposal output-response recognition process. By adding switching control, it is possible to provide a voice recognition response device that can perform efficient and accurate recognition.

Furthermore, by determining the input to the second recognition unit based on whether or not mere voice or sound is input, it is possible to provide a voice recognition response device that anyone can operate more reliably. Also,
If the present invention is applied to a voice typewriter, a high speed and highly reliable typewriter can be realized.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a voice recognition response device according to an embodiment of the present invention, FIG. 2 is a diagram showing the contents of a message memory, and FIG. 3 is a flowchart illustrating the operating procedure of the embodiment. Here, 1...microphone, 2...feature extraction unit,
3... A/D converter, 4... Central processing unit (CPU), 5... Speaker, 6...
external device, 7... first recognition unit, 8... second recognition unit,
9...Control unit, 10...Voice response unit, 11...Message memory.

Claims (3)

[Claims] (1) recognition means for recognizing voice input; response means for outputting the recognition result of the recognition means; confirmation means for determining whether the recognition result is correct by determining the response input to the output of the response means; a word suggestion means for outputting the word selected by a predetermined method as a voice when the confirmation means detects an error in the recognition result; and a word identification means for identifying the word intended to be input by determining a response input to the output of the word suggestion means. A voice recognition response device comprising: means. (2) The voice recognition response device according to claim 1, wherein the word suggestion means sequentially selects words within a predetermined group. (3) The voice recognition response device according to claim 1, wherein the word suggestion means selects words in order of similarity to the recognition result.