JP3597398B2 - Voice recognition device - Google Patents

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a speech recognition apparatus for a conversation performed through a communication path, and more particularly to recognizing a speech during a conversation, searching for information using the recognized speech as a keyword, and providing the searched information to a speaker. .
[0002]
[Prior art]
For example, Japanese Unexamined Patent Publication No. H8-24897 discloses a speech recognition method for providing information and operation support using speech recognition when a conversation such as receiving an order for a product or answering a product inquiry is performed by telephone. Is disclosed. The speech recognition method disclosed in Japanese Patent Application Laid-Open No. Hei 8-24897 discloses a speech recognition method in which a speech signal of only a speaker A is extracted from a communication path in which a speaker A and a speaker B have a conversation to perform speech recognition. The information to be provided to the speaker A is determined using the recognition result. For example, the speaker A on the side where the voice recognition device is placed recognizes the word "product C" from the utterance "Yes, the price of product C, please wait a moment", and displays the information of product C on the display. The speaker A who has displayed and confirmed it can reply the price of the product C to the speaker B.
[0003]
However, if the speech recognition device is always set to the operation mode, even if the speech recognition result is correct, the speech recognition device reacts when the speaker A does not intend, and unexpected screen switching may occur. . Therefore, a timing instructing means for instructing a timing for recognizing the voice is provided. For this reason, when the speaker A is talking with the speaker B, it is necessary to give a signal to start speech recognition with a switch or the like, and the operation becomes complicated and the dialogue with the speaker B is neglected. Cases arise.
[0004]
An object of the present invention is to improve the disadvantages and to provide a voice recognition device that can provide information and assist operation at an accurate timing even when the voice recognition is always in the operation mode.
[0005]
[Means for Solving the Problems]
A voice recognition device according to the present invention includes a voice recognition unit that performs voice recognition by extracting a voice signal from a communication path in which a speaker has a conversation, and a voice recognition result of an utterance of one of the speakers. Comparing the speech recognition results of the other speaker's utterances, and providing a recognition result comparison unit that provides information or operation support to the speaker only when the two speech recognition results satisfy predetermined conditions. Features.
[0006]
It is preferable that the recognition result comparing unit only provides information or operation support to the speaker only when the voice recognition result of the utterance of one speaker and the voice recognition result of the utterance of the other speaker are the same.
[0007]
Also, the recognition result comparison unit provides information or operation support to the speaker only when the voice recognition result of the utterance of one speaker and the voice recognition result of the utterance of the other speaker have the same meaning. Is also good.
[0008]
BEST MODE FOR CARRYING OUT THE INVENTION
The voice recognition device of the present invention includes a transmission voice input unit, a reception voice input unit, a transmission voice recognition unit, a reception voice recognition unit, an information storage unit, a recognition result comparison unit, and a display unit. Is connected to the voice input / output unit side of the speaker A who has a conversation via a communication path such as a public line by a voice input / output unit such as a telephone.
[0009]
For example, when the speaker B calls the voice input / output unit of the speaker A from the voice input / output unit and inquires about the product, the reception voice input unit extracts the reception voice from the speaker B, and receives the reception recognition unit. Send to The received voice recognition unit recognizes a keyword specifying the product from the received received voice and sends the recognized result to the recognition result comparison unit. On the other hand, if the speaker A who has listened to the received voice via the voice input / output unit of the speaker A determines that it is necessary to provide information, the speaker A utters a transmission voice of a response including a keyword specifying the product. . The transmission voice is extracted by the transmission voice input unit and sent to the transmission recognition unit. The transmission voice recognition unit recognizes a keyword for specifying a product from the transmitted transmission voice, and sends the recognized result to the recognition result comparison unit. The recognition result comparison unit compares the keyword recognized by the received voice recognition unit with the keyword recognized by the transmission voice recognition unit. Only when the result is the same, information such as the price and performance of the product is stored in the information storage unit. The information is read out, displayed on the display unit, and transmitted to the speaker A. Speaker A confirms the information of the product displayed on the display unit and conveys it to speaker B.
[0010]
【Example】
FIG. 1 is a block diagram showing the configuration of one embodiment of the present invention. As shown in the figure, a speaker A and a speaker B have a conversation via a communication path 2 such as a public line by voice input / output units 1a and 1b such as telephones. Speaker A is, for example, a party who receives an order for a product or receives a question, and speaker B orders an order for a product or asks a question. A speech recognition device 3 is connected to the speech input / output unit 1a of the speaker A. The voice recognition device 3 includes a transmission voice input unit 4, a reception voice input unit 5, a transmission voice recognition unit 6, a reception voice recognition unit 7, an information storage unit 8, a recognition result comparison unit 9, and a display unit 10. The transmission voice input unit 4 is composed of a voice microphone or the like, and extracts and inputs a transmission voice when the speaker A talks with the voice input / output unit 1a. The transmission voice input unit 5 is connected to the communication path 2, When B talks on the voice input / output unit 1b, it extracts and inputs the received voice received via the communication path 2. The transmission voice recognition unit 6 recognizes the transmission voice input from the transmission voice input unit 4, and since the speaker A can be specified in advance, the voice recognition tuned to the speaker A by the specific speaker method or the speaker adaptation. To improve recognition performance. The received voice recognition unit 7 recognizes the received voice input from the received voice input unit 5, and since the speaker cannot be specified, the voice recognition method of the unspecified speaker system capable of recognizing the voice without depending on the speaker. To recognize the voice. For example, information of various products is stored in the information storage unit 8 in advance. The recognition result comparing unit 9 reads out information corresponding to the voice recognition result from the information storage unit 8 and displays the information only when the voice recognition results of the transmission voice recognition unit 6 and the reception voice recognition unit 8 satisfy predetermined conditions. Display at 10.
[0011]
The operation of the speech recognition apparatus 3 configured as described above when the speaker B inquires the speaker A of, for example, a product will be described.
[0012]
The speaker B makes a call from the voice input / output unit 1b to the voice input / output unit 1a, and the communication path is connected. For example, the speaker B asks the speaker A "I want to tell the price of the product C." When such an utterance is made, the receiving voice input unit 5 extracts the receiving voice of the speaker B, "I want to know the price of the product C," and sends it to the reception recognizing unit 7. The received voice recognition unit 7 recognizes the keyword “commodity C” from the received received voice and sends the recognized result to the recognition result comparison unit 9. On the other hand, if the speaker A who hears the voice "I want to know the price of the product C" via the voice input / output unit 1a determines that information provision is necessary, "Yes, Please wait for a moment. " The transmission voice is extracted by the transmission voice input unit 4 and sent to the transmission recognition unit 6. The transmission voice recognition unit 6 recognizes the keyword “commodity C” from the transmitted transmission voice, and sends the recognized result to the recognition result comparison unit 9. A language model such as a grammar for voice recognition by the reception voice recognition unit 5 and the transmission voice recognition unit 4 uses, for example, a language model 21 displaying a product name or the like as shown in FIG. For example, word spotting, that is, automatically extracting only predetermined words, recognizing words and syllables by ignoring others, and continuous speech recognition recognizing speech that uttered words continuously After recognizing the entire utterance, a product name as shown in FIG. 2 is extracted.
[0013]
The recognition result comparison unit 9 compares the keyword “product C” recognized by the received voice recognition unit 7 with the keyword “product C” recognized by the transmission voice recognition unit 6, and only when the result is the same, the information storage unit 8, information such as the price and performance of “commodity C” is read out, displayed on the display unit 10 and transmitted to the speaker A. The speaker A confirms the information of the product displayed on the display unit 10 and transmits the information to the speaker B.
[0014]
In addition, speaker B makes an utterance, for example, "Do you have a product with a price of 1,000,000 yen or less?", While speaker A says, "If it is product C, it can be purchased for 980,000 yen." In the case of a conversation, the received voice recognition unit 7 has no keyword to recognize, and the transmitted voice recognition unit 6 recognizes the keyword “product C”. Since the recognition result of the transmission voice recognition unit 6 is different, the information of "product C" is not displayed on the display unit 10.
[0015]
The above embodiment has described the case where the corresponding information is displayed on the display unit 10 from the recognition result comparison unit 9 when the result recognized by the reception voice recognition unit 7 and the result recognized by the transmission voice recognition unit 6 are the same. When the result recognized by the reception voice recognition unit 7 and the result recognized by the transmission voice recognition unit 6 have the same meaning, the corresponding information may be displayed on the display unit 10 from the recognition result comparison unit 9.
[0016]
For example, as shown in FIG. 3, when the product having the official name “NT95” is given the nickname “Otoboke-kun” in the recognition result comparison unit 9 so that consumers can easily understand the product, “NTG” is used. , "NT 90" and "Otoboke-kun" are provided in advance with a meaning-reading conversion table 91 such that they represent the same meaning (product) "NT95". If the result recognized by the received voice recognition unit 7 is "NTN", and the result recognized by the transmission voice recognition unit 6 of the utterance of the speaker A is "NTN", the reading is different. The recognition result comparing unit 9 determines that the same meaning “NT95” is represented, and displays the information “NT95” on the display unit 10. In this way, appropriate information can be transmitted from speaker A to speaker B.
[0017]
In the above embodiment, the transmission voice input unit 4 and the reception voice input unit 5 are provided separately, and the transmission voice recognition unit 6 and the reception voice recognition unit 7 are provided separately. The receiving voice input unit 5 is made common, the transmitting voice recognizing unit 6 and the receiving voice recognizing unit 7 are made common, and the received voice and the transmitting voice are recognized by the unspecified speaker system. B may be used for switching. In this way, the device can be simplified.
[0018]
【The invention's effect】
As described above, the present invention compares the speech recognition result of one speaker's speech with the speech recognition result of the other speaker's speech, and only when the two speech recognition results are under predetermined conditions. Since information provision or operation support is provided, the voice recognition device is always set to the operation mode, and information provision and operation support can be performed only when information provision is necessary without instructing the recognition timing. it can.
[0019]
Also, only when the speech recognition result of one speaker's utterance and the speech recognition result of the other speaker's utterance are the same, erroneous information provision etc. Can be prevented, and accurate information can be provided.
[0020]
Furthermore, if the speech recognition result of one speaker's speech and the speech recognition result of the other speaker's speech have the same meaning, information or operation support is provided to the speaker, so that the name or reading In the case where there are multiple words, even if the readings of the speakers' utterances are different, if the meaning is common, it is possible to provide information etc. at the right timing. Even when it is done, accurate information can be provided.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention.
FIG. 2 is an explanatory diagram showing a language model.
FIG. 3 is a configuration diagram of a meaning-reading conversion table.
[Explanation of symbols]
Reference Signs List 1 voice input / output unit 2 communication path 3 voice recognition device 4 transmission voice input unit 5 reception voice input unit 6 transmission voice recognition unit 7 reception voice recognition unit 8 information storage unit 9 recognition result comparison unit 10 display unit

Claims (3)

A speech recognition unit that extracts a speech signal from a communication path in which a speaker has a conversation and performs speech recognition, and a speech recognition result of one speaker's speech and a speech recognition of the other speaker's speech. A speech recognition apparatus, comprising: a recognition result comparing unit that compares results and provides information or assists operation to a speaker only when two speech recognition results satisfy predetermined conditions. The said recognition result comparison part provides information or operation support to a speaker only when the speech recognition result of the speech of one speaker and the speech recognition result of the speech of the other speaker are the same. Voice recognition device. The said recognition result comparison part provides information or operation support to a speaker only when the speech recognition result of the speech of one speaker and the speech recognition result of the speech of the other speaker have the same meaning. The speech recognition device according to claim 1.

Images