JP5378907B2 - Spoken dialogue apparatus and spoken dialogue program - Google Patents

Description

  The present invention relates to a voice dialog for searching for information by voice, and relates to a voice dialog device and a voice dialog program having high search efficiency.

  Conventionally, a technique for recognizing a voice generated by a user and performing information retrieval based on the recognition result is known and used for facility retrieval in a car navigation device.

As an example of using such a technology, for example, if a speech recognition keyword dictionary is expanded based on a keyword registered in a facility specifying information database, and the keyword can be recognized by speech, a search is performed using this keyword as a keyword. A facility search device (for example, see Patent Document 1) and a navigation device (for example, see Patent Document 2) have been proposed. Here, the facility identification information is a road name related to the facility, a town name where the facility exists, a building name of the facility, and a service content of the facility in the facility search device of Patent Document 1, and a facility name in the navigation device of Patent Document 2. , Large genre, small genre, address, position, business day, feature, user evaluation, and the like.
JP 2006-139203 A JP 2007-163226 A

  However, in the facility search device of Patent Literature 1 and the navigation device of Patent Literature 2, the user cannot know in advance all the search keywords that the device can accept. For this reason, if the search keyword which the user came up with is not included in the attribute information of all the registered facilities, the speech recognition result of the user utterance will be erroneously recognized or unrecognized. In such a case, the user does not know whether the same keyword is recognized again or is an invalid keyword in the first place, and as a result of repeating voice input many times, search efficiency decreases. is there.

  In addition, if a keyword that could be input as a search keyword in the past has been deleted from the facility information due to reasons such as closing the facility when the facility information is updated, the search keyword that the system can accept is changed. I can't know that. For this reason, there is a problem that the user can input voices and repeats voice input many times.

  The present invention has been made to solve the above-described problems, and an object thereof is to provide a spoken dialogue apparatus and a spoken dialogue program with high search efficiency by reducing misrecognition and unrecognition of user utterances. To do.

In order to achieve the above object, a spoken dialogue apparatus according to claim 1 is a voice command list comprising a pair of a word that is a voice command and a classification indicating that it is a voice command, and a search keyword in a search target database. A search keyword list comprising a pair of a word and a classification indicating that it is a search keyword, and a pair of a word that is not a search target word and not a voice command or the search keyword, and a classification that indicates a search non-search word A speech recognition dictionary storage means for storing a speech recognition dictionary having a non-target word list comprising: a speech recognition dictionary stored in the speech recognition dictionary storage means for speech recognition of speech data input by a user; a speech recognition means for extracting included word of each and their classification, the classification of each word extracted by the voice recognition unit, the sound Command, the search keyword and determine which of the non-target word, the classification of each word which is the extraction does not contain a voice command, and, when containing the search keyword and a search excluded words, the while searching the search target database classification based on the word is the search keyword, a dialogue control means for generating a response indicating that the word the classification is Ru Oh search covered word is excluded words, the interactive Presenting means for presenting a search result retrieved by the control means and a response generated by the dialogue control means.

According to the first aspect of the present invention, when a word that is not a search target is input, the fact that the word is not a search target is presented, and the user is informed that the search cannot be performed using the word spoken by the user. it can. When a search keyword is input, the search target database can be searched based on the search keyword and the search result can be presented.
According to a second aspect of the present invention, in the voice interaction device according to the first aspect, the voice recognition dictionary storage means is optionally used when performing the search with the voice command list, the search keyword list, and the like. Classification indicating that the word is a non-search word and the non-search word obtained by deleting the voice command and the search keyword from an option word list including a plurality of option words that are highly likely keywords A speech recognition dictionary having the non-target word list consisting of a pair of
The voice dialogue system according to claim 3, wherein, in the voice dialogue system according to claim 2, wherein the voice command dictionary that stores word is a voice command, and options dictionary which stores the option word element item to be searched The speech command list is generated by associating a word stored in the voice command dictionary with a classification indicating a voice command using the search target database storing words indicating information of The search keyword list is generated by associating a word stored in the target database with a classification indicating a search keyword, stored in the option word dictionary, and stored in the voice command dictionary and the search target database. By associating a word that has not been searched with a classification indicating that the word is not a search target word, It generates Dorisuto, the voice command list thus generated further includes a speech recognition dictionary generating means for generating said speech recognition dictionary by registering the search keyword list and the non-target word list.

According to the third aspect of the present invention, it is possible to generate a voice recognition dictionary based on the voice command dictionary, the search target database, and the option word dictionary.

The voice interaction device according to claim 4 is the voice interaction device according to any one of claims 1 to 3 , wherein the dialogue control means is configured such that the extracted word classification includes a voice command. The process corresponding to the voice command is executed.

According to the fourth aspect of the present invention, when a voice command is input, processing corresponding to the command can be executed.

Audio interactive program according to claim 5 causes the computer to function as each means constituting the voice interaction equipment according to any one of claims 1 to 4.

  According to the fifth aspect of the present invention, when a word that is not a search target is input, the fact that the word is not a search target is presented, and the user is notified that the search cannot be performed using the word spoken by the user. it can.

According to a sixth aspect of the present invention, there is provided a voice interaction program comprising: a voice command list comprising a pair of a word that is a voice command and a classification that indicates a voice command; a word that is a search keyword in a search target database; A search keyword list consisting of a pair with a classification indicating that it is present, and a non-target word consisting of a pair of a word that is not a search target word and not a voice command or the search keyword and a classification indicating that it is a non-search target word A voice recognition dictionary storage means for storing a voice recognition dictionary having a list, each of words included in the voice recognition dictionary stored in the voice recognition dictionary storage means for voice recognition of voice data input by a user, and voice recognition means to extract the classification, the classification of each word extracted by the voice recognition unit, said voice command , The search keyword and determine which of the non-target word, the classification of each word which is the extraction does not contain a voice command, and, when containing the search keyword and a search excluded words, the classification together but searching the retrieval target database based on word is the search keyword, dialogue control means for generating a response indicating that the word the classification is Ru Oh search covered word is pair Zogai word, and the interaction It is made to function as a presentation means for presenting a search result retrieved by the control means and a response generated by the dialogue control means.

  According to the sixth aspect of the present invention, when a word not to be searched is input, it is indicated that the word is not to be searched, and the user is informed that the word cannot be searched with the word spoken by himself / herself. it can.

  As described above, according to the present invention, it is possible to reduce the misrecognition and unrecognition of the user utterance and improve the convenience of the voice interactive apparatus.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the present embodiment, dialogue control in the case where the voice dialogue apparatus according to the present invention is used for facility search in a vehicle car navigation system (hereinafter referred to as “navigation”) having a voice recognition function will be described.

  FIG. 1 is a block diagram showing a configuration of a voice interaction apparatus according to an embodiment of the present invention. As shown in the figure, the voice interaction apparatus includes a voice data input unit 11, a voice command dictionary 12, an option word dictionary 13, a search target database 14, a voice recognition dictionary generation unit 15, and a voice recognition dictionary 16. , A voice recognition unit 17, a dialogue control unit 18, an information search unit 19, and a presentation unit 20.

  The voice data input unit 11 includes a microphone, and accepts user voice data input to execute a voice command or search a search target database.

  The voice command dictionary 12 is a dictionary that stores the names of commands that can be operated by voice, and is used for registering voice commands in the voice recognition dictionary 16. FIG. 2 shows a configuration example of the voice command dictionary 12.

  The option word dictionary 13 stores keywords that are likely to be used optionally by the user when performing an information search. For example, we conducted a questionnaire survey to many general users, such as “What keyword is used as an option in addition to area and genre when searching for facilities?” And extracted keywords related to facility attributes from the results. It is good to store in the option word dictionary. The option word dictionary 13 is used to supplement the voice recognition dictionary 16 so that the user can utter a keyword even if the keyword is not included in the search target database 14 in the voice command dictionary 12. FIG. 3 shows a configuration example of the option word dictionary 13.

  The search target database 14 is a database that stores various pieces of information about element items to be searched. In this embodiment, since the voice interactive apparatus is applied to facility search for navigation, the element item is a facility. Accordingly, the search target database 14 stores a plurality of keywords related to the “attribute” of the facility, in addition to basic information such as the “name”, “area”, and “genre” of the facility. For example, keywords related to the attributes of the facility may be extracted based on the contents of the self-PR collected from the manager of the facility through a questionnaire. Further, keywords may be extracted based on information on questionnaires to third parties, reputations on the Internet, and word-of-mouth, or users may be able to register free keywords. FIG. 4 shows a configuration example of the search target database 14.

  The voice recognition dictionary generation unit 15 generates a voice recognition dictionary 16 based on the voice command dictionary 12, the option word dictionary 13, and the search target database 14. The voice recognition dictionary generation unit 15 stores the classification information of “voice command”, “search effective vocabulary”, and “search invalid vocabulary” together with each keyword.

  The speech recognition dictionary 16 is a dictionary that is referred to by the speech recognition unit 17 during speech recognition. A plurality of keywords are included, and each keyword is assigned a classification of “voice command”, “search effective vocabulary”, and “search invalid vocabulary”.

  The voice recognition unit 17 recognizes the voice data input by the voice data input unit 11 using the voice recognition dictionary 16. The recognized words are output with the classification of “voice command”, “search effective vocabulary”, and “search invalid vocabulary”.

  The dialogue control unit 18 performs dialogue processing with the user based on the keyword recognized by the voice recognition unit 17. Specifically, the dialogue control unit 18 controls the information search unit 19 to generate a response for indicating a recognition result by the voice recognition unit 17 or to search the search target database 14.

  The information search unit 19 searches the search target database 14 using the keyword recognized by the voice recognition unit 17 as a search condition, and returns the search result to the dialogue control unit 18.

  The presentation unit 20 includes a speaker and a display, and presents response contents to the user by voice or text. Specifically, the presentation unit 20 presents the response generated by the dialogue control unit 18 and the search result by the information search unit 19.

  The voice interaction apparatus configured as described above generates a voice recognition dictionary 16, recognizes voice data input by the user based on the generated voice recognition dictionary 16, and performs processing such as information search in the search target database 14. I do. FIG. 5 is a flowchart showing the flow of the operation of the voice interaction apparatus when the voice recognition dictionary 16 is generated.

  First, in step 100, the speech recognition dictionary generation unit 15 initializes the speech recognition dictionary 16.

  In step 102, the voice recognition dictionary generation unit 15 creates a voice command vocabulary list by assigning a classification of “voice command” to each keyword stored in the voice command dictionary 12 shown in FIG.

  In step 104, the voice recognition dictionary generation unit 15 registers the created voice command vocabulary list in the voice recognition dictionary 16. Here, the voice command may be registered in the voice recognition dictionary 16 by registering all the keywords stored in the voice command dictionary 12 or by using the screen currently displayed to the user or the content of the response message. In addition, it may be registered after narrowing down to only keywords that the user can speak next. By registering after narrowing down keywords, the accuracy of speech recognition can be improved.

  In step 106, the speech recognition dictionary generation unit 15 extracts keywords that serve as search conditions from the search target database 14. When the keyword recognition is performed, the speech recognition dictionary generation unit 15 uses “area: prefecture”, “area: city” when using the search target database 14 storing the area, genre, and attribute information shown in FIG. 4 as fields. , “Genre”, and “attribute information (1) to (5)” fields are extracted without overlapping each other.

  In step 108, the speech recognition dictionary generation unit 15 assigns a “search effective vocabulary” classification to each keyword extracted from the search target database 14 and creates a search effective vocabulary list. At this time, the keyword “around the current location” may be added to the search effective vocabulary list so that the facility search around the current location can be performed. FIG. 6 shows a configuration example of the search effective vocabulary list.

  In step 110, the speech recognition dictionary generation unit 15 registers the created search effective vocabulary list in the speech recognition dictionary 16.

  In step 112, the speech recognition dictionary generation unit 15 deletes the keywords included in the above-described speech command vocabulary list or search effective vocabulary list from the keywords in the option word dictionary 13 shown in FIG. A search invalid vocabulary list is created by assigning a classification of “vocabulary”. For example, when the option word dictionary 13 shown in FIG. 3 is used, the keywords “cheap”, “can banquet”, and “unknown” included in the search effective vocabulary list of FIG. There are keywords such as “There is parking”, “There is a parking lot”, and “Delicious”. The speech recognition dictionary generator 15 creates a search invalid vocabulary list by assigning classifications to these keywords.

  FIG. 7 shows a generation example of the speech recognition dictionary 16 finally obtained in this way.

  Next, information retrieval by voice dialogue will be described. FIG. 8 is a flowchart showing the flow of the operation of the voice interaction apparatus during information retrieval.

  First, in step 200, the speech recognition dictionary generation unit 15 generates the speech recognition dictionary 16 as described above. In the present embodiment, the speech recognition dictionary 16 is generated every time information search is performed, but the timing for generating the speech recognition dictionary 16 is not limited to this. For example, it may be executed at the timing when any one of the voice command dictionary 12, the option word dictionary 13, and the search target database 14 is updated. In addition, when applied to a system having a function of limiting the search target range in the search target database 14 based on the vehicle position or the like, the speech recognition dictionary 16 is generated in accordance with the timing when the search target range is changed. May be.

  In step 202, the voice data input unit 11 accepts voice data issued by the user for information retrieval by navigation.

  In step 204, the voice recognition unit 17 recognizes voice data input by the user using the voice recognition dictionary 16. The speech recognition unit 17 searches for the same keyword item in the speech recognition dictionary 16 as the keyword obtained as a result of speech recognition, and outputs the result together with the corresponding classification. For example, when the recognition result is “Izakaya”, using a format such as “Izakaya (search effective vocabulary)” together with the classification “search effective vocabulary” assigned to the same keyword in the speech recognition dictionary of FIG. Output. When a plurality of keywords are recognized in one utterance of the user, it is output as a result list.

  The voice recognition by the voice recognition unit 17 may be performed by any known method, for example, by the following method. First, a plurality of phoneme list candidates are generated by referring to an acoustic model from time-series data of speech feature values of speech data, and phonemes that can be expressed by combinations of words registered in the speech recognition dictionary 16 from the phoneme list candidates. Extract list candidates. Next, the acoustic likelihood is calculated from the time series data of the speech feature amount and the phoneme list, and the top N items are extracted in descending order of likelihood. Then, the phoneme list is replaced with a recognition word with classification in the corresponding speech recognition dictionary 16 to obtain a recognition result candidate list. Only candidates that are grammatically established are extracted from this recognition result candidate list, and the highest candidate among the extracted candidates is output as a recognition result. If there is no grammatical candidate, no recognition result is given.

  In step 206, the dialogue control unit 18 determines whether or not the keyword classification recognized by the voice recognition unit 17 is a voice command. If the command is a voice command, the process proceeds to step 208. If the voice command is not, the process proceeds to step 214. move on.

  In step 208, the dialogue control unit 18 executes processing corresponding to the voice command recognized by the voice recognition unit 17, and the presentation unit 20 presents the execution result.

  In step 210, the dialogue control unit 18 determines whether or not the voice command is a command for terminating the search. For example, in the case of a voice command that shifts to an operation at the next stage of search such as “Set as destination”, the dialogue control unit 18 ends the information search. If the voice command is a voice command that continues the search operation such as “return”, the process proceeds to step 212.

  In step 212, the dialogue control unit 18 generates a message for prompting the next input, and after the presentation unit 20 presents the message, the process returns to step 202.

  In step 214, the dialogue control unit 18 determines whether or not the keyword recognized by the voice recognition unit 17 includes one or more search effective vocabularies. If the keyword includes one or more search effective vocabularies, the process proceeds to step 216. If not included, the process proceeds to step 222.

  In step 216, the dialogue control unit 18 generates a message so that it can be identified whether the search effective vocabulary or the search invalid vocabulary for each recognized keyword, and the presentation unit 20 presents the message.

  In step 218, the information search unit 19 searches the search target database 14 using a keyword that is a search effective vocabulary among the recognized keywords as a search condition.

  In step 220, the presentation unit 20 presents the search result by the information search unit 19. Further, after the dialogue control unit 18 generates a message for prompting the next input and the presentation unit 20 presents the message, the process returns to step 202.

  Information retrieval by the information retrieval unit 19 is performed as follows. For example, the speech recognition unit 17 determines from the user's utterance “restaurant with lunch near the current location”, “around the current location (search effective vocabulary)”, “with lunch (search invalid vocabulary)”, “restaurant (search effective vocabulary). ) "Is recognized. In this case, the dialogue control unit 18 generates a message indicating whether the keyword is valid or invalid, such as ““ There is lunch ”is a keyword that is invalid for search. Then, the presentation unit 20 presents this. Thereafter, from the result of searching the search target database 14 under the two conditions “current location” and “restaurant”, the dialogue control unit 18 generates and presents a message such as “5 found”, for example. Part 20 presents this. At the same time, the dialogue control unit 18 generates a message for prompting the next input, such as “The nearest restaurant is XX. Do you want to set the destination? The presentation unit 20 presents this.

  In step 222, the dialogue control unit 18 generates a message indicating that the recognized keyword is a search invalid vocabulary, and a message for prompting another input such as “Please say again”. After the presentation, the process returns to step 202.

  As described above, the voice interaction apparatus according to the present embodiment uses the option word dictionary 13 for keywords not included in the voice command dictionary 12 or the search target database 14 when generating the voice recognition dictionary 16. And categorize whether the search keyword is valid or invalid. Thereby, it is possible to cope with voice recognition of a keyword that is highly likely to be used when the user searches for information. In addition, it is possible to immediately notify the user whether or not the recognized keyword is valid for the search, and the user does not repeatedly input the invalid keyword by voice many times, and as a result, information retrieval is efficiently performed. be able to.

  In addition, this invention is not limited to the above-mentioned embodiment, It is applicable also to what changed the design within the range described in the claim. For example, it can be applied to music search.

It is a block diagram which shows the structure of the voice interactive apparatus which concerns on embodiment of this invention. It is a figure which shows the structural example of a voice command dictionary. It is a figure which shows the structural example of an option word dictionary. It is a figure which shows the structural example of a search object database. It is a flowchart which shows the flow of an effect | action at the time of the speech recognition dictionary production | generation of the speech dialogue apparatus which concerns on this Embodiment. It is a figure which shows the structural example of a search effective vocabulary list. It is a figure which shows the structural example of a speech recognition dictionary. It is a flowchart which shows the flow of an effect | action at the time of the information search of the voice interactive apparatus which concerns on this Embodiment.

DESCRIPTION OF SYMBOLS 11 Voice data input part 12 Voice command dictionary 13 Option word dictionary 14 Search object database 15 Voice recognition dictionary production | generation part 16 Voice recognition dictionary 17 Voice recognition part 18 Dialog control part 19 Information search part 20 Presentation part

Claims (6)

A search keyword list consisting of a pair of a voice command list consisting of a word that is a voice command and a classification indicating that it is a voice command, and a pair indicating a word that is a search keyword in the search target database and a classification indicating that it is a search keyword And a speech recognition dictionary storing a speech recognition dictionary having a word list that is not a search target word but a word that is not a search target word and a classification indicating that it is a non-search target word Dictionary storage means;
Speech recognition means for speech recognition of speech data input by a user, and extracting each of words included in the speech recognition dictionary stored in the speech recognition dictionary storage means and their classification;
It is determined whether the classification of each word extracted by the voice recognition means is the voice command, the search keyword, or the non-target word, the classification of each extracted word does not include a voice command, and search keyword and a search in cases involving non-target word, with searching the retrieval target database based on word the classification is a search keyword, is exempt word Ah Ru words the classification is by not searched word A dialogue control means for generating a response indicating that
Presenting means for presenting a search result retrieved by the dialog control means and a response generated by the dialog control means;
Spoken dialogue device with The voice recognition dictionary storage means includes the voice command list, the search keyword list, and the voice command from an option word list including a plurality of option words that are keywords that are likely to be used optionally when performing a search. And a speech recognition dictionary having a non-search word list formed by pairs of words that are non-search words obtained by deleting the search keyword and classifications indicating non-search words. The voice interactive apparatus according to claim 1. Using a voice command dictionary that stores word is a voice command, and options dictionary which stores the option word, the search target database, which stores the word indicating information element item to be searched, the voice command The speech command list is generated by associating words stored in the dictionary with classifications indicating voice commands, and the words stored in the search target database are associated with classifications indicating search keywords. To generate a search keyword list and associate a word that is stored in the option word dictionary and not stored in the voice command dictionary or the search target database with a classification that indicates a word that is not a search target. The non-target word list is generated, the generated voice command list, the search key Dorisuto and voice dialogue system of the further claims 2, further comprising a speech recognition dictionary generating means for generating a speech recognition dictionary by registering the non-target word list. The dialogue control unit, when the extracted word classification includes voice command, voice dialogue system according to any one of claims 1 to 3 for executing a process corresponding to the voice command. Computer, voice interaction program for functioning as each unit constituting the voice interaction equipment according to any one of claims 1 to 4. Computer
A search keyword list consisting of a pair of a voice command list consisting of a word that is a voice command and a classification indicating that it is a voice command, and a pair indicating a word that is a search keyword in the search target database and a classification indicating that it is a search keyword And a speech recognition dictionary storing a speech recognition dictionary having a word list that is not a search target word but a word that is not a search target word and a classification indicating that it is a non-search target word Dictionary storage means,
The voice data input by the user to recognize speech, the speech recognition means to extract each and that classification of words included in the stored speech recognition dictionaries in the speech recognition dictionary storing means,
It is determined whether the classification of each word extracted by the voice recognition means is the voice command, the search keyword, or the non-target word, the classification of each extracted word does not include a voice command, and search when a keyword and the search outside word, as well as searching for the search target database based on word the classification is a search keyword, the classification is the Ah Ru terms in the search target outside word pair Zogai Interaction control means for generating a response indicating a word; and
Presenting means for presenting a search result retrieved by the dialog control means and a response generated by the dialog control means;
Spoken dialogue program to function as.

Images