JP2011203349A - Speech recognition system and automatic retrieving system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speech recognition system and an automatic retrieving system, capable of reducing incorrect recognition of a keyword at natural communication time.SOLUTION: An automatic retrieving system 1 includes a telephone unit 3 installed in a vehicle P, a controller unit 4 installed in a call center Q, a telephone unit 5, and an operator terminal 6. The controller unit 4 includes: a speech recognition processor 7 which obtains utterance information of a driver and an operator from the telephone unit 5, and which extracts the keyword relating to intention of the driver from utterance contents before and after the characteristic utterance, when it is determined that characteristic utterance is included in the utterance information; and an automatic retrieving processor 8 for performing automatic point of interest (POI) retrieving by using the keyword extracted by the speech recognition processor 7.

Description

  The present invention relates to a voice recognition system that performs voice recognition during a call between a driver and an operator, and an automatic search system using the voice recognition system.

  As a conventional speech recognition system, for example, as described in Patent Document 1, when a speaker's speech is subjected to speech recognition processing, a keyword is detected from the recognition result, and a condition corresponding to the keyword is satisfied In addition, there is known one that outputs information related to the keyword.

JP 2005-215726 A

  By the way, during a natural call without a scenario, the amount of words that the speaker utters increases, making it difficult to extract arbitrary words, so it is possible to get the necessary important words (keywords) and pick up extra words There is.

  An object of the present invention is to provide a voice recognition system and an automatic search system that can reduce misrecognition of a keyword during a natural call.

  The speech recognition system according to the present invention determines utterance information acquisition means for acquiring utterance information during a call between a driver and an operator, and determines whether or not a characteristic utterance is included in the utterance information acquired by the utterance information acquisition means. Characteristic utterance determination means, and keyword extraction means for extracting a keyword from utterance contents before and after the characteristic utterance when the characteristic utterance determination means determines that the characteristic utterance is included. It is what.

  As described above, in the speech recognition system of the present invention, the utterance information at the time of a natural call between the driver and the operator is acquired, and when it is determined that the utterance information includes the characteristic utterance, the characteristic utterance Extract keywords from previous and next utterance contents. Important keywords intended by the driver are likely to appear before and after characteristic utterances by the operator. This makes it possible to reduce misrecognition of keywords by specifying characteristic utterances in which keywords are likely to appear before and after utterances even during natural calls with a large amount of words uttered by drivers and operators.

  Preferably, the apparatus further includes dictionary switching means for switching a voice recognition dictionary used for recognition of voices of a driver and an operator. In this case, by switching to an appropriate speech recognition dictionary at an appropriate timing, the keyword recognition time can be shortened while further reducing the erroneous keyword recognition.

  At this time, it is preferable to further include a dialogue phase discriminating unit for discriminating the dialogue phase of the characteristic utterance, and the dictionary switching unit includes a unit for switching the voice recognition dictionary according to the dialogue phase discriminated by the dialogue phase discriminating unit. The dialogue phase includes business visits and conditions. By using a speech recognition dictionary suitable for such a dialogue phase, keyword misrecognition can be more reliably reduced.

  The dictionary switching means may include means for switching the speech recognition dictionary according to words appearing before and after the characteristic utterance. In this case, misrecognition of keywords can be more reliably reduced by using a speech recognition dictionary suitable for words appearing before and after a characteristic utterance.

  Furthermore, a search operation means for an operator to perform an operation related to information search is further provided, and the dictionary switching means responds to the search word when the search word is input by the search operation means and the start of the search process is instructed. You may have a means to switch a speech recognition dictionary. When an operator operates search operation means to execute search processing, the use of a speech recognition dictionary suitable for the search word input by the search operation means can more reliably reduce keyword misrecognition. Can do.

  The automatic search system according to the present invention determines utterance information acquisition means for acquiring utterance information during a call between a driver and an operator, and whether or not a characteristic utterance is included in the utterance information acquired by the utterance information acquisition means. A characteristic utterance judging means, a keyword extracting means for extracting a keyword from utterance contents before and after the characteristic utterance when the characteristic utterance judging means judges that the characteristic utterance is included, and a keyword extracting means Search means for searching for information based on the extracted keyword is provided.

  As described above, in the automatic search system of the present invention, the utterance information at the time of natural call between the driver and the operator is acquired, and when it is determined that the utterance information includes the characteristic utterance, the characteristic utterance Extract keywords from previous and next utterance contents. Important keywords intended by the driver are likely to appear before and after characteristic utterances by the operator. This makes it possible to reduce misrecognition of keywords by specifying characteristic utterances in which keywords are likely to appear before and after utterances even during natural calls with a large amount of words uttered by drivers and operators. In addition, the information search is automatically performed based on an appropriate keyword with few misrecognitions, so that the operator does not have to input a search word. Therefore, the operator's work efficiency can be improved.

  According to the present invention, it is possible to reduce misrecognition of keywords during a natural call. This makes it possible to appropriately perform automatic search for information using the keyword.

It is a block diagram which shows schematic structure of 1st Embodiment of the automatic search system concerning this invention. It is a flowchart which shows the detail of the procedure of the speech recognition process performed by the speech recognition process part shown in FIG. It is a table | surface which shows an example of characteristic utterance. It is a flowchart which shows an example of the telephone call image of a driver and an operator. It is a block diagram which shows schematic structure of 2nd Embodiment of the automatic search system concerning this invention. It is a flowchart which shows the detail of the procedure of the speech recognition dictionary switching process performed by the dictionary switching process part shown in FIG. It is a table | surface which shows an example of a dialogue phase and characteristic utterance. It is a flowchart which shows an example which switches a speech recognition dictionary in the telephone call of a driver and an operator.

  DESCRIPTION OF EMBODIMENTS Hereinafter, preferred embodiments of a speech recognition system and an automatic search system according to the present invention will be described in detail with reference to the drawings.

  FIG. 1 is a block diagram showing a schematic configuration of a first embodiment of an automatic search system according to the present invention. In the figure, an automatic search system 1 of the present embodiment is a system that automatically performs a point of interest (POI) search from a natural call between a driver of a vehicle P and an operator of a call center Q, and includes a voice recognition system. Yes. The POI search is a search for facility information such as restaurants, hotels, convenience stores, gas stations, and the like.

  The automatic search system 1 includes a car navigation (car navigation) 2 and a telephone 3 mounted on a vehicle P, a controller unit 4, a telephone 5 and an operator terminal 6 installed in a call center Q.

  The telephone 3 is a mobile phone or in-vehicle wireless device in which a driver makes a voice call with an operator, and is connected to the car navigation 2.

  The telephone 5 and the operator terminal 6 are connected to the controller unit 4. The telephone 5 is used by an operator to make a voice call with a driver. The operator terminal 6 is used by an operator to perform operations related to POI search and display search results on the screen.

  The controller unit 4 includes a voice recognition processing unit 7 and an automatic search processing unit 8. The voice recognition processing unit 7 inputs the speech information of the driver and the operator from the telephone 5, performs a predetermined voice recognition process using the voice recognition dictionary 9, and outputs the recognition result to the automatic search processing unit 8. The automatic search processing unit 8 executes an automatic POI search process using the facility information database 10 and displays the search result on the operator terminal 6 on the screen.

  FIG. 2 is a flowchart showing details of the procedure of the speech recognition processing executed by the speech recognition processing unit 7. In the figure, first, the utterance information of the driver and the operator is acquired from the telephone 5 (step S51).

  Subsequently, it is determined whether or not the acquired utterance information includes a characteristic utterance prepared in advance (step S52). A characteristic utterance is a word that can be expected to acquire keywords related to the driver's intention before and after the utterance.

  As the characteristic utterances here, for example, as shown in Fig. 3, there is a request for a question such as "Would you like to meet you today?" Can be mentioned. For example, the understanding expression is a word that is uttered when the operator understands the driver's story. For this reason, before and after the utterance of the understanding expression, words that allow the operator to confirm the driver's intention are likely to appear.

  When it is determined that the acquired utterance information does not include a characteristic utterance, the process returns to step S51. When it is determined that a characteristic utterance is included in the acquired utterance information, the appearance position of the characteristic utterance is retained, and keywords related to the driver's intention are extracted from the utterance contents before and after the characteristic utterance. (Procedure S53).

  Subsequently, it is determined whether or not the extracted keyword includes a search keyword such as an area or a genre word (step S54). When it is determined that no area or genre word exists in the extracted keyword, the process returns to step S51. If it is determined that an area or genre word exists in the extracted keyword, it is determined whether or not an area or genre word that can be a search keyword is prepared (step S55). If it is determined that there are no areas or genre words that can be used as search keywords, the process returns to step S51.

  If it is determined that the search keyword area and genre words have been prepared, the search keyword is sent to the automatic search processing unit 8 to request execution of an automatic POI search (step S56).

  Then, the automatic search processing unit 8 automatically performs a POI search using the search keyword from the voice recognition processing unit 7 and displays the search result on the operator terminal 6 to present it to the operator. Thereafter, the operator reports the search result to the driver through a call.

  FIG. 4 shows an example of a call image without a scenario between a driver and an operator. In the figure, the utterances “Let's go today”, “Wanted”, “Strictly” and “Budget” uttered by the operator are characteristic utterances. Then, “Koenji” and “Japanese food” uttered by the driver and operator before and after the characteristic utterance are important keywords related to the driver's intention. Accordingly, “Koenji” and “Japanese food” are sent as search keywords from the speech recognition processing unit 7 to the automatic search processing unit 8, and the automatic search processing unit 8 executes an automatic POI search for “Koenji” and “Japanese food”. It will be.

  In the above, the procedure S51 of the telephone 5 and the voice recognition processing unit 7 constitutes utterance information acquisition means for acquiring utterance information during a call between the driver and the operator. The procedure S52 constitutes characteristic utterance determining means for determining whether or not characteristic utterance is included in the utterance information acquired by the utterance information acquiring means. The steps S53 to S55 constitute keyword extracting means for extracting a keyword from utterance contents before and after the characteristic utterance when the characteristic utterance determining means determines that the characteristic utterance is included. The procedure S56 and the automatic search processing unit 8 constitute search means for performing information search based on the keyword extracted by the keyword extraction means.

  As described above, in the present embodiment, characteristic utterances are extracted from the utterance information at the time of a call between the driver and the operator so as to easily identify the position of the keyword related to the driver's intention before and after the utterance. Since the keywords are extracted from the utterance information before and after the keyword, erroneous recognition of the keywords used for the automatic POI search can be suppressed when a natural call without a scenario between the driver and the operator is performed.

  In addition, since the POI search using a keyword related to the driver's intention is automatically executed, the operator does not have to input the search word one by one or operate the search instruction button. Accordingly, it is possible to improve the operator's work efficiency. As a result, the response time of the search result to the driver is shortened, so that the stress received by the driver can be reduced.

  FIG. 5 is a block diagram showing a schematic configuration of the second embodiment of the automatic search system according to the present invention. In the figure, the same or equivalent elements as those in the first embodiment are denoted by the same reference numerals, and the description thereof is omitted.

  In the figure, the automatic search system 1 of this embodiment includes a controller unit 20 instead of the controller unit 4. The controller unit 20 includes a dictionary switching processing unit 21 in addition to the voice recognition processing unit 7 and the automatic search processing unit 8 described above.

  The dictionary switching processing unit 21 switches the speech recognition dictionary 9 to be used with a flow of a call between the driver and the operator or an operation of the operator terminal 6 by the operator as a trigger. As the speech recognition dictionary 9, for example, a dictionary in which all words are registered, a genre word, a dictionary in which words related to areas and requested words are registered, a dictionary in which words related to meals are registered, words related to menus, and related words are registered. A plurality of types are prepared such as a dictionary or the like (see FIG. 8).

  FIG. 6 is a flowchart showing details of the procedure of the speech recognition dictionary switching process executed by the dictionary switching processing unit 21. In the figure, first, utterance information of a driver and an operator is acquired from the telephone 5 (step S61).

  Subsequently, it is determined whether or not the search instruction button of the operator terminal 6 has been pressed by the operator (step S62). When it is determined that the search instruction button has been pressed, the voice recognition dictionary 9 suitable for the search keyword input to the operator terminal 6 is switched (step S63).

  When a word related to a genre word or area that is essential as a search keyword appears during a call, the operator may input the word related to the genre word or area to the operator terminal 6 and further press a search instruction button on the operator terminal 6. Therefore, the operation of the operator terminal 6 by the operator plays an important role as a trigger for switching the voice recognition dictionary 9. At this time, for example, when the genre word is a restaurant, the language is switched to the voice recognition dictionary 9 in which words related to the meal menu are registered.

  When it is determined that the search instruction button has not been pressed, it is determined whether or not a characteristic utterance prepared in advance is included in the utterance information acquired in step S61 (step S64).

  As described above, the characteristic utterance is a word that can be expected to acquire a keyword related to the driver's intention before and after the utterance, and examples thereof include those shown in FIG. Characteristic utterances are divided into dialogue phases. The dialogue phase is a group of calls made by dividing a call between a driver and an operator according to progress or changes in topics. As the dialogue phase, there are a visit, a condition inquiry (outline), and a condition inquiry (detail).

  When it is determined that the acquired utterance information does not include a characteristic utterance, the process returns to step S61. On the other hand, when it is determined that the acquired utterance information includes a characteristic utterance, the dialogue phase of the characteristic utterance is determined (step S65). Then, the speech recognition dictionary 9 suitable for the determined dialogue phase is switched (step S66).

  For example, if the characteristic utterance is “Would you like to do it today?”, The dialogue phase will be a patronage, indicating that it has shifted to the dialogue phase intended to acquire the main keywords of the search. Therefore, the voice recognition dictionary 9 is switched by using the transition to the dialogue phase as a trigger. When the dialogue phase is a visit, the dictionary can be narrowed down to genre words and areas, so that the speech recognition dictionary 9 in which words related to genre words and areas are registered is switched.

  In parallel with the processing of steps S65 and S66, the speech recognition dictionary 9 suitable for words (words) appearing before and after the characteristic utterance is switched (step S67). Since words appearing before and after the characteristic utterance are often important words related to the driver's intention, they are handled as triggers for switching the speech recognition dictionary 9.

  FIG. 8 shows an example of switching the voice recognition dictionary in a call between a driver and an operator. In the figure, first, a speech recognition dictionary in which all words are registered is used.

  When the operator says, “What is your business today?”, The dialogue phase (inquiry) of the characteristic utterance is used as a trigger in the speech recognition dictionary where genre words, area words, and required words are registered. Can be switched. Then, in response to the characteristic utterances uttered by the operator, when the driver says "I want to eat in Shinjuku", the word "meal" is used as a trigger, and here is a genre word (related to the restaurant where you eat) Word) is switched to the registered speech recognition dictionary.

  After that, when an important word related to the driver's intention called “ramen shop” appears in the call between the driver and the operator, the operator inputs “Shinjuku ramen shop” to the operator terminal 6 as a search word, and searches for the operator terminal 6. Press the instruction button. Then, the operation of the search instruction button is used as a trigger to switch to the speech recognition dictionary in which words related to the ramen menu and related words are registered.

  Thereafter, when an important word related to the driver's intention called “Tonkotsu Ramen” appears in the call between the driver and the operator, for example, the automatic search processing unit 8 executes an automatic POI search using “Shinjuku Tonkotsu Ramen” as a search keyword. It will be.

  In the above, the communication device 5 and the procedure S61 of the automatic search processing unit 8 constitute speech information acquisition means for acquiring speech information during a call between the driver and the operator. The procedure S64 constitutes characteristic utterance determining means for determining whether or not characteristic utterance is included in the utterance information acquired by the utterance information acquiring means. The above steps S62, S63, S66, and S67 constitute a dictionary switching means for switching a voice recognition dictionary used for recognition of voices of drivers and operators. The procedure S65 constitutes a dialogue phase discrimination means for discriminating the dialogue phase of the characteristic utterance.

  By the way, as the vocabulary registered in the speech recognition dictionary 9 increases, the speech recognition rate tends to decrease. For this reason, conventionally, for example, there is a method of preparing a speech recognition dictionary for each category and sequentially switching the speech recognition dictionary when erroneous recognition is performed. However, in this case, if the erroneous recognition continues, the voice recognition time becomes long.

  On the other hand, in this embodiment, the flow of the call between the driver and the operator and the operation of the search instruction button on the operator terminal 8 by the operator are used as triggers to switch to the appropriate speech recognition dictionary 9 with a small registered vocabulary at an appropriate timing. As a result, the accuracy of voice recognition increases. Thereby, the misrecognition of the keyword used for an automatic POI search can be further suppressed. Moreover, the time until the keyword is recognized can be shortened.

  The present invention is not limited to the above embodiment. For example, the voice recognition system of the present invention can be applied to systems other than the automatic search system described above as long as voice recognition is performed during a call between a driver and an operator.

  DESCRIPTION OF SYMBOLS 1 ... Automatic search system, 4 ... Controller unit, 5 ... Call machine (speech information acquisition means, 6 ... Operator terminal (search operation means), 7 ... Voice recognition process part (Speech information acquisition means, characteristic utterance judgment means, keyword Extraction means, search means), 8 ... automatic search processing section (search means), 9 ... voice recognition dictionary, 20 ... controller unit, 21 ... dictionary switching processing section (utterance information acquisition means, characteristic utterance determination means, dictionary switching means) , Dialog phase discrimination means).

Claims (6)

Utterance information acquisition means for acquiring utterance information during a call between a driver and an operator;
Characteristic utterance determination means for determining whether or not characteristic utterance is included in the utterance information acquired by the utterance information acquisition means;
Voice recognition comprising: keyword extraction means for extracting a keyword from utterance contents before and after the characteristic utterance when the characteristic utterance determination means determines that the characteristic utterance is included. system.   The speech recognition system according to claim 1, further comprising dictionary switching means for switching a speech recognition dictionary used for recognition of speech of the driver and the operator. A dialogue phase discrimination means for discriminating a dialogue phase of the characteristic utterance;
3. The speech recognition system according to claim 2, wherein the dictionary switching means includes means for switching the speech recognition dictionary in accordance with the dialog phase determined by the dialog phase determining means.   4. The speech recognition system according to claim 2, wherein the dictionary switching means includes means for switching the speech recognition dictionary according to words appearing before and after the characteristic utterance. The operator further comprises search operation means for performing an operation related to information search,
The dictionary switching means includes means for switching the speech recognition dictionary according to the search word when a search word is input by the search operation means and an instruction to start search processing is given. The speech recognition system according to any one of claims 2 to 4. Utterance information acquisition means for acquiring utterance information during a call between a driver and an operator;
Characteristic utterance determination means for determining whether or not characteristic utterance is included in the utterance information acquired by the utterance information acquisition means;
Keyword extracting means for extracting a keyword from utterance contents before and after the characteristic utterance when the characteristic utterance determining means determines that the characteristic utterance is included;
An automatic search system comprising: search means for performing information search based on the keyword extracted by the keyword extraction means.

Images