JPS6333797A - Voice recognition control - Google Patents

Abstract

(57) [Abstract] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a voice recognition control method, and more particularly to a voice recognition control method for correcting erroneous recognition when it occurs in voice input.

(Prior Art) Currently, voice response recognition devices using telephone lines have been introduced in the financial industry, distribution industry, etc., and services such as balance inquiries and order entry are provided to users. In other words, these services are performed by the user manually inputting a personal identification number, membership number, service code, etc. by voice from a telephone or the like.

FIG. 6 is a diagram showing a procedure for checking recognition results by a conventional voice response recognition device. As can be seen from the figure, the voice uttered by the user is recognized according to the synthesized voice guidance such as ``Please enter your password'' from the voice response recognition device. Then, check the recognition results and if there is a misrecognition,
As shown in the figure, the recognition results are corrected in order starting from the first digit.

(Problem to be solved by the invention) However, in the conventional method described above, even if the number of digits to be manually input is large and only one digit is misrecognized in the recognition result, the misrecognition is performed sequentially starting from the first digit. Since searching and making corrections is required, there is a problem in that it takes time to finalize the human resources, leading to a decline in service quality.

The present invention is intended to solve these problems, and when erroneous recognition occurs, it shortens the time it takes to confirm the recognition result by checking the recognition result, re-entering it for correction, etc.
The purpose of this invention is to provide a voice recognition control method that can improve services to users.

(Means for Solving the Problems) In order to solve the above problems, the present invention provides a word speech recognition unit that recognizes each word of input speech, and a word speech recognition unit based on speech data for guidance to the user. In a voice response recognition device that is equipped with a voice synthesis unit that outputs a synthesized voice, and that performs a process of confirming the recognition result for each word of the input voice using kaidance from the voice synthesis unit, the reliability of the recognition result for each word of the input voice is is calculated, and confirmation processing is performed based on this degree of certainty.

(Function) According to the present invention having the above configuration, the present invention functions as follows.

Each word of the input speech input by the user is speech recognized by the word speech recognition section. Then, confirmation processing of the recognition result of the input speech is performed according to guidance from the speech synthesis section. At this time, if a recognition error occurs in the recognition result, the confirmation process for each word is performed starting from the word with the lowest degree of certainty representing the accuracy rate of the recognition result. Then, when the correct re-input of the erroneously recognized word is completed, the process of confirming the recognition result for each word of the input speech is completed.

Therefore, the present invention can solve the above-mentioned problems, shorten the time required to confirm the recognition result, re-input for correction, etc., and improve the service provided to the user. A recognition control method can be provided.

(Example) Hereinafter, one embodiment of the present invention will be described based on the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention. In the figure, 1 is a telephone set, 2 is an exchange, 3 is a line control unit that controls the interface with a public telephone line, etc., 4 is a voice recognition unit that recognizes the voice uttered by the user, and 5 is a main control unit 6, which will be described later. 6 is a main control unit that controls each component of the voice response recognition device; 7 is a voice recognition unit 4;
This is a host computer that calculates the degree of certainty of the first candidate from the recognition results obtained in , that is, the first candidate, the second candidate, and each distance value, using a calculation formula described later. Note that the calculation of this degree of certainty may be performed by the main control section 6.

Further, FIG. 2 is a block diagram showing the speech recognition section 4 of FIG. 1. In the figure, 41 is a voice analysis circuit that analyzes the input voice uttered by the user and extracts feature parameters;
Reference numeral 42 denotes a speech dictionary in which feature parameters of speech in categories to be recognized are stored; reference numeral 43 compares the feature parameters from the speech analysis circuit 41 with each feature parameter stored in the speech dictionary 42, and calculates the distance between the two. A matching calculation circuit 44 calculates the first one in order of decreasing distance value based on the distance value of each feature parameter calculated by the matching calculation circuit 43.
This is a judgment circuit that selects a second candidate for comparison and outputs it as a recognition result.

Next, the recognition result confirmation process according to this embodiment will be explained.

Here, FIG. 3 is a flowchart showing the operation of this embodiment, and for example, FIG. 4 shows the relationship between recognition results, distance values, and certainty when the input voice is a 6-digit number. FIG. 5 shows the procedure for checking the recognition results.

First, the degree of certainty P calculated by the host computer 7 in FIG.
To explain how to calculate the recognition result of the first candidate, the certainty P of the recognition result of the first candidate is calculated using the formula (1
).

p=i□/JZI −(+) In other words, the higher the degree of certainty P is, the more the recognition result of the first candidate is a correct recognition of the input voice.

The voice response recognition device sends a guidance message such as "Please enter your PIN" to the user as shown in FIG. 5 via the voice synthesis unit 5 shown in FIG. , "cue" as shown in Figure 5, "
Hachi”, “Nana”.

``Roku'', ``Yon'', and ``Zero'' are recognized by the speech recognition unit 4 in Figure 1.
(steps 52 and 53). The host computer 7 in FIG. 1 calculates the degree of certainty for each digit of the recognition result using equation (1) from the first candidate, second candidate, and each distance value (step 61). The first candidate category for each digit is used as the recognition result, and the PIN number is 9, 8, 7, etc. as shown in Figure 5.
6,3. The voice synthesizer 5 shown in FIG. 1 outputs a synthesized voice to confirm the message, ``It's O, right?'' (step 62).
.

In response to this confirmation gun dance, the user utters a message indicating that the input was not correct, such as "No" as shown in Figure 5, because the content is different from what was input manually, and the voice response recognition device recognizes this input voice. Recognize (step 63). It is determined whether or not erroneous recognition has occurred (step 64). If erroneous recognition has occurred, the following confirmation process for each digit is performed.

This confirmation process for each digit is performed in the order of the digits with the lowest degree of certainty calculated by the host computer in FIG.

Therefore, as shown in FIG. 4, confirmation processing is performed for the fifth digit with the lowest degree of certainty, that is, the first place in the degree of uncertainty rank r (step 71).

The voice response recognition device outputs a confirmation response to the user such as "The fifth digit is 3" as shown in FIG. 5 by the speech synthesis unit 5 shown in FIG. 1 (step 72). In response to this guidance, the user considers the fifth digit, and as shown in Figure 5, the user says something like "No" to indicate that the input was incorrect because it differs from the content manually inputted, and the voice response is recognized. The device recognizes this input audio (step 7
3).

Therefore, the voice response recognition device proceeds to step 62 if the recognition result of the input voice is an instruction to end the process such as "Owari", and if it is "no", it determines that there was an erroneous recognition in the fifth digit. 74.75), performs the re-input process for the 5th digit.

This re-input process is performed as follows.

The voice response device outputs a re-input guidance such as "Please enter the 5th digit again" as shown in FIG. The uttered voice "yon" is recognized (step 82). In the re-input speech recognition, it is better to exclude the result of the previous erroneous recognition, that is, "san" in this embodiment, from the recognition target.

Hereinafter, in this example, the recognition result of the re-input voice of the 5th digit is "
As shown in Figure 5, the voice response recognition device outputs confirmation guidance such as ``The 5th digit is 4'' by the voice synthesis unit 5 in Figure 1. (step 83). The user responds to this guidance by saying "yes" as shown in Figure 5, as it matches the voice input.
The voice response recognition device recognizes this input voice (step 84), and determines whether the re-input of the fifth digit has been correctly corrected (step 85). In this embodiment, it is correct, so the process advances to step 76. However, if erroneous recognition occurs again due to re-input, the process advances to step 81 and the re-input process is performed again.

Next, in step 76, it is determined whether the confirmation process for each digit has been completed for a predetermined number of digits (six digits in this embodiment). Since we have only processed the fifth digit, step 77
Then, confirmation processing is performed on the second digit whose uncertainty level is the next one, r=2. Therefore, the voice response recognition device outputs confirmation guidance such as "The second digit is 8" as shown in FIG. 5 by the speech synthesis unit 5 shown in FIG. 1 (step 72). In response to this guidance, the user does not need to correct any digits other than the 5th digit that have been correctly entered by re-entering, so the user may not be able to correct the digits shown in Figure 5.
Then, the voice response recognition device recognizes this input voice (step 73), and determines whether the subsequent confirmation process for each digit has ended (step 74). In this embodiment, the process now proceeds to step 62.

Therefore, based on the final result, the voice response recognition device provides guidance to confirm all digits, such as "Your password is 9,8.7°6.4.O," as shown in Figure 5, as shown in Figure 1. is outputted by the speech synthesis section 5 (step 62). In response to this guidance, the user says that all digits match the voice input, so as shown in Figure 5, the user says something like "hi" to indicate that the voice input was correct, and the voice response recognition device recognizes this guidance. The input voice is recognized (step 63), the input voice of all digits is determined to be correct (step 64), and the recognition result confirmation process is ended.

(Effects of the Invention) As explained above, according to the present invention, when erroneous recognition of input speech occurs, confirmation processing for each word is performed starting from the digit with the lowest recognition certainty, and the digit that was erroneously recognized is corrected. Since the confirmation process for each word is finished when the re-input is finished, the time and effort required for the confirmation process can be reduced, and it is possible to provide a speech recognition i4J control method capable of improving the service to the user.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a diagram showing the configuration of the speech recognition section shown in FIG. 1, FIG. 3 is a flowchart showing the operation of this embodiment, and FIG. A diagram showing the relationship between recognition results, distance values, and degrees of certainty in this embodiment; FIG. 5 is a diagram showing a procedure for confirming recognition results in this embodiment; and FIG. 6 is a diagram showing confirmation of recognition results by a conventional voice response recognition device. It is a figure which shows a procedure. ■...Telephone, 2... Switchboard, 3... Line control section, 4... Speech recognition section, 5-- Speech synthesis section, 6... Main control section, 7... Host computer, 4
1...Speech analysis circuit, 42-...B voice dictionary, 4
3--Verification calculation circuit, 44... Judgment circuit.

Claims (2)

[Claims] (1) It is equipped with a word speech recognition section that recognizes each word of input speech, and a speech synthesis section that outputs synthesized speech based on speech data for guidance to the user, In a voice response recognition device that performs confirmation processing of recognition results for each word of input speech using guidance, the degree of certainty of recognition results for each word of input speech is calculated, and confirmation processing of the recognition results of input speech is performed to prevent misrecognition. If this occurs, the confirmation process for each word is performed starting from the words with the lowest degree of certainty, and the confirmation process for the recognition results for each word of the input voice is completed when the correct re-input of the erroneously recognized word is completed. Characteristic voice recognition control method. (2) The degree of certainty is the first recognition result for each word of the input speech.
The voice recognition control method according to claim 1, wherein the distance value of the candidate is l_1 and the distance value of the second candidate is l_2, and the calculation is performed by l_2/l_1.