JPH11143488A - Voice recognition device - Google Patents

Abstract

(57) [Summary] [Problem] To eliminate discomfort to a speaker by constantly checking the correctness of the recognition result in order to prevent erroneous recognition and malfunction in a dialog between the speech recognition device and the speaker. , To provide a speech recognition service to a speaker smoothly and accurately. SOLUTION: In a voice recognition device 100, a recognition result for voice input from speakers 110 to 112 is determined,
The optimal voice guidance to the speakers 110 to 112 is achieved by selecting from a plurality of patterns in the database unit 108. By selecting the plurality of voice guidance patterns, it is possible to smoothly and accurately specify the information to be input next to the speaker.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition apparatus which is located in a communication network such as a public network and is suitable for providing a speech input service. The present invention relates to a voice recognition device that can determine and transmit appropriate guidance to a speaker.

[0002]

2. Description of the Related Art Hitherto, as a typical application of a speech recognition apparatus, practical studies have been made to apply it to a public network and to operate services of various contents. In applying the speech recognition device to a public network, the speech recognition device does not always recognize the input speech correctly due to factors such as differences in the surrounding environment (for example, noise) of the speaker and the type of speech input device (for example, telephone). However, it does not always meet the request (utterance) of the speaker 100%. For this reason, various countermeasures for the speaker in response to erroneous recognition have been studied.

One of them is disclosed in Japanese Patent Application Laid-Open No. 3-248199.
Japanese Patent Laid-Open Publication No. H10-15064 discloses a speech recognition apparatus that obtains a recognition result by comparing a voice input with a registered vocabulary, and transmits the recognition result as an operation instruction of another machine, and defines a first threshold and a second threshold. The sum D of the “city” distance between the input voice pattern and the reference pattern is taken, and the smallest one is taken as the recognition result,
If the reliability is larger than the first threshold, the recognition result is transmitted as an operation instruction if 1 / D or the like is a reliability, and if the reliability is larger than the second threshold and smaller than the first threshold, it is used. The recognition result is transmitted only when the user confirms the recognition result, and when the reliability is smaller than the second threshold value, the recognition result is invalidated. A speech recognition system that prevents the occurrence of a voice recognition is shown.

In general, a voice response (guidance output) function for confirming whether a recognition result is correct or not is applied between a voice recognition device and a speaker.

[0005] The flow of a dialog in a conventional speech recognition apparatus will be described with reference to FIG. The speech recognition device calculates the likelihood between the input word speeches S111 and S114 and all of the words registered in the device in advance, and calculates the likelihood in S10 of FIG.
2. As shown in S105, "Sales Department" and "Kato-san", which are the first candidates of the recognition result, are output as voices, and the speaker is requested to input "Yes" and "No" as confirmation. Often. Also, when the likelihood of the word to be checked is extremely small, it is determined that the likelihood of a correct answer is low, and the checking process is not performed, and as shown in S104 of FIG. Re-enter such as "Please enter."

[0006]

As described above, in the dialogue between the conventional voice recognition apparatus and the speaker, it is effective in that no malfunction occurs due to voice recognition, while the recognition result is correct. Since the confirmation is always repeated, it takes a long time and often causes the speaker to be uncomfortable psychologically. An object of the present invention is to solve the above-mentioned problems of the conventional technology.

That is, an object of the present invention is to increase the risk of providing erroneous services due to erroneous recognition,
It is to provide voice input and voice identification service smoothly, speedily, reliably and accurately.

[0008]

According to the present invention, a recognition result for a speech input from a speaker is calculated based on a result calculated by a reliability calculation unit provided in a speech recognition apparatus. This is achieved by providing means for selecting an appropriate voice guidance to the speaker from the plurality of guidance transition patterns prepared based on the reliability.

As described above, in the voice recognition system in which the speaker performs the next voice input in accordance with the voice guidance sent from the voice recognition device, the voice recognition device inputs the voice to the voice speaker next. Since the voice guidance for instructing the voice input information to be analyzed is a voice guidance pattern selected from a plurality of voice guidance patterns according to the analysis result of the voice state of the speaker, the voice input is performed. Service can be performed smoothly and accurately.

The guidance contents that can be output are broadly divided into the following four. (1) Dialogue promotion guidance For example, when one service is provided by a plurality of information inputs, guidance for prompting the (n + 1) th information input after the nth information input is completed.

(2) Reflection Guidance For example, "Yes" or "Yes" like "I'm OO"
Guidance that prompts for the input of positive words such as "No" and "Negative" such as "No."

(3) Listening Guidance When one service can be provided by a plurality of information inputs, after the n-th information input is completed, the n-th information is again input.
Guidance for prompting for second information.

(4) Recognizable Word Presentation Guidance Guidance to promptly output words prepared by the voice recognition at this stage as recognition candidates and to select and input from the recognized word group.

According to the present invention, when the return guidance is not required, for example, the dialogue promotion guidance is transmitted to the speaker, so that the speaker does not recognize the confirmation guidance and outputs the voice in the guidance content of the next process. Input becomes possible, and discomfort and annoyance are reduced.

Further, in order to make it possible to finally confirm the guidance content, it is possible to store the guidance data in the guidance data storage unit and confirm the connection destination. Hereinafter, the present invention will be described in detail with reference to the embodiments shown in FIGS.

The structure and operation of the voice recognition apparatus according to the present invention for selecting and transmitting voice guidance will be specifically described with reference to the embodiments shown in FIGS.

[0017]

FIG. 2 is a diagram showing a hardware configuration of the present invention. A speech recognition apparatus 100 stores input speech from speech input devices 110 to 112 and a dictionary data storage unit 103 for recognition. A speech recognition unit 102 is provided for determining a score between the stored dictionary data for recognition and determining a candidate word for input speech. The score may be, for example, the distance between the input voice and the dictionary data for recognition, and the reciprocal of the sum.

Further, the speech recognition apparatus 100 transmits the speech data from the speech recognition section 102 in addition to the recognition dictionary data storage section 103 for storing the recognition dictionary data to be referred to when the speech recognition section 102 recognizes the input speech. A reliability calculation unit 101 that calculates the reliability of a candidate word to be calculated. The reliability can be obtained, for example, by calculating the difference between the score of the first candidate of the recognition result and the score of the second candidate of the recognition result.

The speech recognition apparatus 100 further receives the code of the first candidate word determined by the speech recognition unit 102 and the reliability calculated by the reliability calculation unit 101, and outputs the next guidance data and the next guidance data. A control unit 107 for designating an address on a database of recognition dictionary data to be used for recognition and transferring them to a guidance data storage unit 105 and a recognition dictionary data storage unit 103, respectively; a recognition dictionary data 108D used for speech recognition; A database unit 108 for storing guidance data 108G used at the time of voice synthesis, a voice synthesis unit 104 for converting the guidance data into voice data and sending it to the speaker, and a guidance data storage unit 105 for storing guidance data used at the time of voice synthesis. And

FIG. 3 is a diagram showing the flow of the processing operation of the voice recognition apparatus shown in FIG. 2 together with voice (dashed arrow line) and control data (solid arrow line).

When a line is connected between the speaker and the voice recognition device, the control section 107 selects guidance data for prompting the speaker to input a voice, and uses the corresponding guidance data address (C301). , The guidance data is read out from the database unit 108 and the storage unit 105 is read out.
The guidance data is transferred to (C302).

The voice synthesis unit 104 receives the guidance data stored in the guidance data storage unit 105, converts the guidance data into guidance voice, and transmits the guidance voice to the speaker (A301).

In the meantime, the control unit 107 stores the dictionary data for recognition used for voice recognition based on the dictionary data address for recognition (C304).
Is transferred to the recognition dictionary data storage unit 103 (C30).
5).

When there is a voice input from the speaker for the output guidance (A302), the voice recognition unit 102 inputs the input voice (A302) and the recognition dictionary data (C306) sent to the voice recognition unit 102. Is calculated (C307), and candidate words for the input voice are determined.

The speech recognition section 102 sends the obtained score of the recognition result (C307) to the reliability calculation section 101 to obtain the reliability of the recognition result (C308), and controls the recognition result and the reliability in the control section. It is transmitted to C107 (C309).

The control unit 107 selects the guidance data to be used next and the selection dictionary data for recognition in accordance with the value of the reliability.
Then, the guidance data address (C310) and the dictionary data address for recognition (C313) are transmitted and transferred to the guidance data storage unit 105 and the dictionary data storage unit for recognition 103 (C311) and (C314). The guidance data is further sent to the speech synthesis unit 104 (C31
2) It is converted into guidance voice and transmitted to voice input devices 110 to 112 such as telephones (A30)
3). The dictionary data for recognition is transferred to the voice recognition unit 102.

Next, a case where the dialogue promotion guidance is output will be described with reference to FIGS. 4, 5 and 7. FIG.

As shown in FIG. 7, the dialogue promotion guidance (S701) from the voice recognition device 100, for example, in response to "Please enter your department", the input voice of the speaker (S71).
1) is “sales department”, and the reliability R for this is greater than the threshold th1 as shown in item 1 of FIG. 4 (S5).
06), as shown in FIG. 5, the control unit 107 selects the next guidance data and recognition dictionary data for recognizing the next necessary information without performing the reflection (S50).
7) and (S508).

As a result, as shown in FIG. 7, the guidance output to the speaker is the dialogue promotion guidance for recognizing the second necessary information (S702), "Please name the person in charge." .

Next, when the reliability R of the recognition result of the input voice of the speaker is smaller than the threshold th1 and larger than the threshold th2 as shown in item No. 2 of FIG. (S509), (S510), (S511), (S
As indicated by 512), the next guidance data and recognition dictionary data are selected so as to output the reflection guidance. The voice recognition device performs the operations of (S102) and (S102) of FIG.
Reflection guidance as shown in 105) can be output.

Next, when the reliability R of the recognition result of the input voice of the speaker is smaller than the threshold th2 as shown in item No. 3 of FIG. 4, the control unit 107 proceeds to (S513) and (S51) of FIG.
As shown in 4), the next guidance data and recognition dictionary data are selected so as to output the re-listening guidance. Thus, the guidance output to the speaker side is, for example, "Please ask again" as shown in (S104) of FIG.

Next, a case where the input of the speaker is not registered in the dictionary data for recognition will be described with reference to FIGS. 4, 5 and 8. FIG. For example, as shown in FIG. 8, when the input voice of the speaker is “Electronic parts sales department” and the score A of the recognition result of the input voice is smaller than the threshold D as in item No. 4 in FIG. The control unit 107 selects the guidance data and the dictionary data for recognition in the next order. Thus, the guidance output to the speaker side is, for example, the recognizable word presentation guidance as shown in (S802) of FIG. 8 "Please select from the human resources department, the planning department, and the sales department".

Next, a description will be given of a case where all recognized contents are provided with guidance at the end of voice input. The recognition result is stored in the guidance data storage unit 105 in FIG. 2, and the guidance data is selected according to the flow of the process in FIG. For example, "I will connect to Kato of the sales department" shown in FIG.

[0034]

As described above, according to the present invention, since a plurality of patterns of guidance are sent to the speaker based on the numerical value of the reliability calculation unit, it is possible to omit useless repetitions, etc., and to facilitate the dialogue. Make it possible.

Thus, there is an effect that the discomfort of the speaker can be reduced and the service requested by the speaker can be provided accurately.

[Brief description of the drawings]

FIG. 1 is a transition diagram showing an example of a dialogue between a conventional voice recognition device and a speaker.

FIG. 2 is a hardware configuration diagram of a speech recognition device according to the present invention.

FIG. 3 is a sequence diagram showing an operation sequence in the voice recognition device according to the present invention.

FIG. 4 is a table showing a list of judgment conditions for the next guidance based on the reliability of the speech recognition apparatus according to the present invention.

FIG. 5 is a flowchart of a next guidance selection process of the voice recognition device according to the present invention.

FIG. 6 is a flowchart of a next guidance selection process of the voice recognition device according to the present invention.

FIG. 7 is a transition diagram showing an example of a dialog realized by the speech recognition device according to the present invention.

FIG. 8 is a transition diagram showing an example of a dialog realized by the speech recognition device according to the present invention.

[Explanation of symbols]

101: reliability calculation unit, 102: speech recognition unit, 103:
Recognition dictionary data storage unit, 104 ... Speech synthesis unit, 105
... Guidance data storage unit, 106 ... Interface unit, 107 ... Control unit, 108 ... Database unit, 110
~ 112 ... voice input device

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Masahiro Kosaka 216 Totsuka-cho, Totsuka-ku, Yokohama-shi, Kanagawa Prefecture Inside the Hitachi, Ltd.Information and Communications Department (72) Inventor Hana Igarashi 180 Totsuka-cho, Totsuka-ku, Yokohama-shi, Kanagawa Hitachi Communication Systems Co., Ltd.

Claims (9)

[Claims] 1. A vocabulary candidate group corresponding to a function and service for receiving a voice input of a speaker, which is located in a communication network,
It has a voice recognition function for selecting one or more vocabulary candidates based on the voice input and outputting information on its certainty, and a response function for storing a plurality of types of response data to be transmitted to the speaker. Outputting a transmission guidance corresponding to response data selected from a plurality of types of response data stored in the response function, using information on the likelihood output from the voice recognition function. Recognition device. 2. A plurality of types of response data, wherein at least two of data for dialogue promotion guidance, data for return guidance, data for repeat guidance, and data for guidance on presentation of recognizable words are stored. The speech recognition device according to claim 1, wherein 3. The speech recognition apparatus according to claim 2, wherein the dialog promoting guidance is output when the information on the probability of the recognition result exceeds a certain value. 4. The speech recognition apparatus according to claim 1, wherein the next information can be input to the speaker without outputting guidance for confirming the recognition result. 5. The apparatus has response data for instructing a speaker to re-input the same voice as the voice input at the time of first recognition, and the distance data and the distance difference data between the two are below a certain value. 2. The speech recognition device according to claim 1, wherein the output is performed in the case of. 6. The speech recognition apparatus according to claim 1, further comprising response data for explaining service contents that can be provided to the speaker, and outputting the response data when the distance data is equal to or less than a predetermined value. 7. The speech recognition apparatus according to claim 1, wherein said speech recognition apparatus has a function of presenting a target word to said speaker. 8. When the voice recognition device transmits response data for promoting dialogue, finally, when the voice input is completed, all recognized contents are output as response data, and the speaker is asked to confirm. The speech recognition device according to claim 2, wherein: 9. The voice recognition according to claim 2, wherein the content of the voice recognition is provided before transmitting the recognition result to the destination (the name of the service to be connected). apparatus.