JP2007328283A - Interaction system, program and interactive method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an interactive technology capable of attaining more natural interaction. <P>SOLUTION: The interactive system comprises: a nodding means for outputting voice of nodding, every time a period of voice by utterance of an interactive party, is input (Step 21, 24 and 25); an extracting means for extracting a word or a phrase, in which appearance frequency in the input voice reaches a prescribed value or more (Step 21 to 23); a searching means for searching a word or a phrase related to the word or the phrase extracted by the extracting means (Step 26); and an outputting means for outputting the voice using the word or the phrase which the searching means has searched and obtained (Step 28). <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a dialogue apparatus, a program, and a dialogue method that extract a word or phrase included in a dialogue partner's utterance and output a voice as an utterance of contents related thereto.

  In recent years, development of an interactive agent device that uses voice input and enables anyone to receive various information and services with peace of mind has been progressing. FIG. 4 shows the configuration of a conventional interactive agent device. This apparatus includes an input unit 41 that converts voice uttered by a user into a voice signal, a voice dialogue system 40 that generates a voice signal for output based on a voice signal from the input unit 41, and a voice signal from the voice dialogue system 40. Is output to a voice.

  The voice dialogue system 40 performs voice recognition based on a voice signal from the input unit 41 and extracts text data, text data from the voice recognition engine 42 and text data based on predetermined scenario data. A dialogue processing engine 43 to be generated, a storage unit 44 that stores scenario data used by the dialogue processing engine 43, and a voice synthesis engine 45 that generates a voice signal based on text data from the dialogue processing engine 43 and supplies the voice signal to the output unit 46. Is provided.

  Here, the scenario data is a combination of process item data and transition definition data, and determines the operation of the dialog processing engine 43. The process item data describes the processing contents for each predetermined process item, and the transition definition data defines a transition that can occur between processes of each process item. In the case of dialogue processing, it can be said that the scenario data defines the content of the corresponding utterance to be output for words and phrases included in the input speech based on the utterance of the dialogue partner.

  In this configuration, the voice uttered by the user is converted into a voice signal at the input unit 41 and supplied to the voice recognition engine 42. The voice recognition engine 42 converts the supplied voice signal into text data and supplies it to the dialogue processing engine 43. The dialogue processing engine 43 creates text data corresponding to the speech to be output according to the scenario data based on the supplied text data, and supplies the text data to the speech synthesis engine 45. The voice synthesis engine 45 converts the supplied text data into a voice signal and supplies it to the output unit 46. The output unit 46 converts the supplied audio signal into sound and outputs the sound. In this way, the agent device functions as a user's dialogue partner.

  In addition, as an example of dialogue technology with an electronic computer using voice input / output functions, a database used for speech recognition is created in advance for each topic scene, and input is performed by disassembling input text from the user during dialogue. Identify scenes that contain words, and based on the identified scenes, proceed with the conversation by limiting the content of the conversation, and if there is a change in the conversation status with the user, build a new database according to the change, There has been proposed a system in which information from the new database is provided to a user (see, for example, Patent Document 1).

  On the other hand, as an example of information retrieval technology using voice input, an information provision request including a keyword of inputted desired information is transmitted to an information provision server, and an information provision screen transferred from the server is received in response to the request. An information providing support system has been proposed (see, for example, Patent Document 2). According to this, various information and services can be provided very easily and easily without requiring a troublesome operation.

  A spoken dialogue apparatus has also been proposed in which a paraphrase of a searched official name is generated, and a speech recognition dictionary including the paraphrase is generated as a speech recognition dictionary used in the next speech recognition process. (For example, refer to Patent Document 3). According to this, even when the user does not know the official name registered in the database correctly, the official name can be searched and presented.

JP 2005-004501 A JP 2004-38252 A JP 2005-338274 A

  However, according to the interactive agent device of FIG. 4 described above, since the utterance is performed based on the utterance text prepared in advance, it is necessary to prepare the utterance text in advance. Also, in order to increase the text that can be spoken, it is necessary to accumulate a huge amount of data. Even in the case of the above-described dialogue technique with an electronic computer, it is necessary to prepare a database in which scene-specific words are accumulated.

  In addition, according to each of the above-mentioned techniques, when a word that is not recognized is spoken, a cold reply such as “Please speak again” is often given. Also, when providing information from a Web server, the user's voice is recognized, and the information providing screen of the Web server is simply displayed accordingly. Is not done. Therefore, the current interactive device has a resistance to the elderly and the like, and has a problem in terms of operability, so that it has not yet spread.

  An object of the present invention is to provide a dialogue technique that can realize a more natural dialogue in view of the problems of the prior art.

  In order to achieve the above object, the dialogue apparatus according to the first invention includes an extraction means for extracting a word or phrase whose appearance frequency in the utterance of the conversation partner is a predetermined value or more, and the word or phrase extracted by the extraction means. And generating means for generating a text indicating the content of the utterance related to the voice, and output means for outputting voice based on the text generated by the generating means.

  Here, as the interactive device, for example, a device implemented in a navigation device, a portable terminal, a personal computer, or the like is applicable. For example, a word or phrase that has appeared more than a predetermined value corresponds to a word or phrase that has appeared more than a predetermined number of times since the start of dialogue. As text indicating the content of the utterance related to the extracted word or phrase, for example, text using “Bo-chan” or “Sanshiro” related to the extracted “Natsume Soseki”, for example, “Natsume Soseki is Sanshiro "". Moreover, the text matched with the word or phrase by the above-mentioned scenario data may be sufficient.

  In this configuration, a word or phrase whose appearance frequency in the input speech is greater than or equal to a predetermined value is considered to represent an item that the user has some interest or interest in. Therefore, extracting such a word or phrase and uttering a content related thereto is a response in accordance with the user's interest or interest. Therefore, natural dialogue can be realized.

  According to a second aspect of the present invention, there is provided the interactive apparatus according to the first aspect, further comprising changing means for changing the predetermined value in accordance with a given instruction.

  According to a third aspect of the present invention, there is provided a dialogue apparatus according to the first or second aspect, further comprising search means for searching for a word or phrase related to the word or phrase extracted by the extraction means, wherein the generation means includes the related utterance. A text using a word or a phrase obtained by the search by the search means is generated as the text indicating the content of the text.

  According to a fourth aspect of the present invention, in the third aspect of the present invention, the search unit searches the related word or phrase based on a database on the network using the word or phrase extracted by the extraction unit as a keyword. It is what is performed.

  In the third or fourth invention, the dialogue apparatus according to the fifth invention is any one of the registration means for registering the word or phrase extracted by the extraction means, and when there are a plurality of the related words or phrases. And determining means for determining whether to generate the text using a word or phrase by referring to the word or phrase registered by the registration means.

  A dialogue apparatus according to a sixth invention is characterized in that, in any one of the first to fifth inventions, there is a dialogue means for outputting a voice indicating that a dialogue is made for each utterance by a dialogue partner. For example, an utterance that is followed by a silence state for a predetermined time or longer, or an utterance that ends with a predetermined ending, for example, “... Dai”, “... Ine”, etc. Applicable.

  A program according to a seventh invention causes a computer to function as each means in any one of the first to sixth interactive devices.

  In the dialogue method according to the eighth invention, the dialogue device extracts a word or a phrase whose appearance frequency in the dialogue partner's utterance is equal to or higher than a predetermined value, and the dialogue device extracts the word or phrase extracted in the extraction step, or It comprises a generation step for generating text indicating the content of an utterance related to a phrase, and an output step for outputting speech based on the text generated in the generation step.

  According to the present invention, when it is possible to extract a word or phrase whose appearance frequency in the utterance of the conversation partner is equal to or higher than a predetermined value, it is possible to output an utterance having contents related to the extracted word or phrase. Therefore, it is possible to appropriately carry out a dialogue by uttering a content in line with the user's interest or interest.

  In addition, since the predetermined value can be changed according to a given instruction, the extraction frequency of the word or phrase is adjusted, and the utterance of the content corresponding to the extracted word or phrase from the interactive device side is adjusted. The frequency can be adjusted.

  Further, as the text indicating the content of the utterance related to the extracted word or phrase, the word or phrase related to the extracted word or phrase is searched, and the text using the obtained word or phrase is generated. It is possible to conduct a dialogue by uttering a content that meets the user's interest or interest.

  In addition, because the extracted word or phrase is used as a keyword and the related word or phrase is searched based on the database on the network, the related word or phrase is used without the need to provide a large database on the interactive device. Dialogue can be performed by uttering content.

  Also, register the extracted word or phrase and refer to the registered word or phrase for which word or phrase to use to create text when there are multiple related words or phrases Therefore, information that the user does not know is given using words or phrases that have not been used so far, and the dialogue content is guided in a direction to expand the dialogue content, It is possible to use a word or phrase to guide the user to a conversation that deepens the field of interest.

  In addition, since the voice to the effect of competing is output during the search by the search means, even if the search is in progress, the conversation is interrupted without making the other party feel that the conversation is interrupted. Can appeal.

  FIG. 1 is a block diagram showing a configuration of an agent device according to an embodiment of the present invention. This agent device can be implemented as part of a navigation device, for example. It may be implemented as a device dedicated to dialogue or by software on a personal computer. Further, the present invention may be implemented by displaying a character representing an agent and having an interface for communicating with the character.

  As shown in the figure, this agent device includes an input unit 11 that converts a voice uttered by a user into a voice signal, and a voice recognition engine 12 that performs voice recognition and extracts sentence data based on a voice signal from the input unit 11. , A word is extracted from sentence data from the speech recognition engine 12, and based on the word given from the dialogue engine control unit 13, the dialogue engine control unit 13 that performs various processes and outputs based on the word, A Web data search unit 14 acquired from a Web database 15 on a predetermined Web server, an utterance text generation unit 16 that generates text used for utterance based on information acquired by the Web data search unit 14, a dialog engine control unit 13, A speech synthesis engine 17 that generates a speech signal based on text data from the dialog text creation unit 16. Based on the audio signal from the speech engine 17, and a word list storage unit 19 for storing the words output unit 18 for outputting voice, and interactive engine control unit 13 it is extracted.

  The agent device functions as an interactive agent by its entire configuration, but mainly the dialog engine control unit 13 and the utterance text creation unit 16 play the main roles. The dialog engine control unit 13 has a function that combines the functions of the dialog processing engine 43 and the scenario data storage unit 44 in the conventional example of FIG. A dialog system is configured by the functions of the dialog processing engine and the three engines of the speech recognition engine 12 and the speech synthesis engine 17.

  The operation of the dialogue system is determined by scenario data. That is, the dialog processing engine realizes a dialog function by controlling the speech recognition engine 12 and the speech synthesis engine 17 based on the scenario data.

  The input unit 11 includes means for accepting voice input such as a microphone. The input speech signal is converted into text by the speech recognition engine 12 and supplied to the dialog engine control unit 13. As a text conversion method, a dictation method for extracting sentence data is used. The dialogue engine control unit 13 extracts words from the text data supplied by the speech recognition engine 12 based on a predetermined logic, and provides the web data search unit 14 with the words.

  The Web data search unit 14 has a function of accessing the Web database 15, transmitting a search request including a given word, and receiving a search result returned from the Web database 15 in response thereto. The Web database 15 has a function of performing an information search using a word included in the search request as a keyword and providing a search result to the client. For example, in response to a search request including the word “Natsume Soseki”, a search result “Bo-chan, Sanshiro” can be returned. The web database 15 may be an existing website or a dedicated website. However, it is better to use a dedicated website for smooth conversation.

  The utterance text creation unit 16 creates an utterance text based on the information acquired by the Web data search unit 14 and supplies the utterance text to the speech synthesis engine 17. The speech synthesis engine 17 converts the speech text supplied from the speech text creation unit 16 into a speech signal and supplies the speech signal to the output unit 18. The output unit 18 converts a given audio signal into a sound wave by a speaker and outputs the sound as a sound.

  The dialog engine control unit 13 selects a character string that has been repeatedly used a predetermined number of times from each sentence extracted by the speech recognition engine 12 and extracts it as a word frequently used by the user in the current dialog. The number of extracted words changes according to the predetermined number of times. Each time one sentence is input, the agent device outputs a message to the effect that the agent makes a conflict and turns it around to the listener. During that time, the extracted word is registered in the word list of the storage unit 19 even once. For registered words, the number of times of extraction is recorded. That is, when the extracted word is already registered, the number of extractions is counted up. The word list is used by the utterance text creation unit 16 to determine which word should be used to create the utterance text when there are a plurality of words as information acquired by the Web data search unit 14.

  In addition, you may make it delete from the word list about the word for which fixed time passed since registration. When the number of registrations in the word list exceeds a predetermined value, the oldest one may be deleted. In addition, the user himself / herself can register the word list in advance so that his / her preference can be reflected in the determination of the word to be used when creating the utterance text.

  FIG. 2 is a flowchart showing dialogue processing in the apparatus of FIG. FIG. 3 shows an example of the conversation contents in this process. When the dialogue process is started, first, in step 21, a single voice input is accepted. At this time, for example, as shown in FIG. 3, “I read“ Botchan ”from“ Natsume Soseki ”. ”Can be accepted. For example, the voice input that is followed by a silence state for a predetermined time or longer, or a voice that ends with a predetermined ending, for example, “... Dayo”, “... Ine”, etc. Applicable.

  Next, in step 22, the speech recognition engine 12 extracts text data based on the speech signal output from the input unit 11 in response to speech input. As a result, I read the text data corresponding to the voice input, for example, “Bokchan” of “Natsume Soseki”. Can be obtained.

  Next, in step 23, the dialogue engine control unit 13 determines whether or not the sentence data extracted by the speech recognition engine 12 includes a word that has appeared more than a predetermined number of times after the dialogue processing is started. This determination is performed by incrementing the number of appearances for each word that appears in the extracted sentence data each time the process of step 22 is performed after the start of the dialogue process. Can be determined by checking whether or not has exceeded a predetermined number of times.

  If it is determined that there is no word that has appeared more than a predetermined number of times, in step 24, a voice output indicating that a match has been made is performed, and the process returns to step 21. The output indicating that a match is made is text data indicating that a match is made, for example, “Fun. Is supplied to the speech synthesis engine 17. The supplied text data is output as a voice that strikes a conflict through the speech synthesis engine 17 and the output unit 18. When returning to step 21, according to the example in FIG. 3, “Natsume Soseki is interesting. Can be obtained.

  If it is determined in step 23 that a word that has appeared a predetermined number of times or more is present, in step 25, a voice output indicating that a conflict is made is performed in the same manner as in step 24. In the example of FIG. 3, “Natsume Soseki” is illustrated as a word that appears more than a predetermined number of times. That is, “Natsume Soseki” is extracted as a frequent word that appears more than a predetermined number of times. In addition, in the example of FIG. Is output.

  Next, in step 26, the Web data search unit 14 sends a search request using the frequent word as a keyword to the Web database 15 and receives the search result from the Web database 15. In the example of FIG. 3, the Web data search unit 14 sends a search request with “Natsume Soseki” as a keyword that appears more than a predetermined number of times in the dialogue process to the Web database 15 and obtains “Bo-chan” and “Sanshiro” as search results. is doing. Such acquisition of information from the Web database 15 can be performed in parallel while turning to the listener by performing a voice output to the effect of making a conflict in step 25.

  In step 26, registration to the word list in the storage unit 19 is also performed. That is, it is determined in step 23 that it has appeared a predetermined number of times or more, and the extracted frequent word is registered in the word list, and the extracted number is set to “1”. If the word has already been registered, the number of extractions is counted up.

  Next, in step 27, the utterance text creation unit 16 creates a text to be used for the utterance based on the word acquired by the Web data search unit 14. At this time, if there are a plurality of acquired words, text is created using words that are not registered in the word list of the storage unit 19. When all the words are registered, the word with the smallest number of extractions is used. Since it is considered that the user has little knowledge about words that appear less frequently in dialogue, it is more meaningful to create utterance texts using such words as providing new knowledge to the user and It is easy to spread.

  For example, as shown in FIG. 3, if the acquired words are “Bo-chan” and “Sanshiro”, it is checked whether “Bo-chan” and “Sanshiro” are registered in the word list. In the case of the example in FIG. 3, since “Bo-chan” is registered in the word list and “San-Shiro” is not registered, the utterance text creation unit 16 uses “San-Shiro”, and ““ Natsume Soseki ” Create an utterance text saying "There is Sanshiro".

  This creation can be performed using a format prepared in advance, such as “There is XXX if XXX”. The extracted frequent word is applied to the “xxx” portion, and the word acquired by the search is applied to the “xxx” portion. Therefore, by applying “Natsume Soseki” to “XXX” and “Sanshiro” to “XX”, the utterance text “Natsume Soseki has“ Sanshiro ”in addition to that.” Can be created.

  Contrary to the above, the utterance text may be created using words with a large number of extractions. In this case, it is possible to create an utterance text according to the user's interest. In addition, a method for selecting a word to be used when creating these utterance texts may be switched as appropriate.

  Next, in step 28, voice output is performed based on the created utterance text. That is, the speech synthesis engine 17 converts the speech text created by the speech text creation unit 16 into a speech signal and supplies the speech signal to the output unit 18. The output unit 18 converts the sound signal into sound and outputs the sound.

  According to the present embodiment, by competing with the user's utterance (step 24), the user turns to the user's listener, extracts a word that appears more than a predetermined number of times in the user's utterance, and acquires a word related to the word Even when a search is being performed (step 25), even if the user just wants to speak, the user can turn to the listener, and at the same time, the voice is interrupted by the search, so the flow of the conversation Even in a scene where there is a possibility of interruption, it is possible to appeal the impression that the user is listening seriously. In other words, because the answer is returned, the user can continue the conversation in a natural manner, and the utterance according to the topic is returned at an appropriate time, so that a more humane conversation can be performed. .

  At that time, if the user wants the agent device to talk to the user, the word extraction frequency can be set to a large value by setting a predetermined number of times as a reference for word extraction based on the number of appearances. And the utterance frequency from the agent device can be suppressed. Conversely, when it is desired to increase the frequency of utterances other than the conflict from the agent device, the word extraction frequency may be increased. Thereby, the user can perform a dialogue adapted to his / her pace.

  Even when new information from the Web database 15 is provided, the provided information is not simply displayed as a Web page, but is communicated by voice including words as provided information. As a thing, it can be inserted into the conversation in a natural way. Therefore, provision information can be easily transmitted to elderly people and the like.

  In addition, this invention is not limited to the above-mentioned embodiment, It can deform | transform and implement suitably. For example, in the above description, a word is extracted from the input voice and given to the Web data search unit 14, but in addition to this, a phrase may also be extracted and given.

  In the above description, the Web data search unit 14 requests the Web database 15 to search related words. However, a dedicated database is provided on the apparatus, and the related words are searched using this database. You may do it. At that time, if the search cannot be performed in the dedicated database, a search request may be made to the Web database 15. In this case, the search result obtained from the Web database 15 may be reflected in the dedicated database.

  Moreover, in the above-mentioned, the web data search part 14 acquires the related word about the word extracted from the input audio | voice by the dialog engine control part 13, and the utterance text preparation part 16 produces the text for speech containing the acquired word. However, instead of this, the dialog engine control unit 13 creates an utterance text according to the scenario data based on the extracted word as in the past, and the utterance because the scenario data does not correspond. Only for extracted words for which text cannot be created, the utterance text may be created in the utterance text creation unit 16 according to the embodiment described above.

It is a block diagram which shows the structure of the agent apparatus which concerns on one Embodiment of this invention. It is a flowchart which shows the dialogue process in the apparatus of FIG. It is a figure which shows an example of the dialog content in the process of FIG. It is a block diagram which shows the structure of the conventional interactive agent apparatus.

Explanation of symbols

11, 41: Input unit, 12, 42: Speech recognition engine, 13: Dialogue engine control unit, 14: Web data search unit, 15: Web database, 16: Utterance text creation unit, 17, 45: Speech synthesis engine, 18 , 46: output unit, 40: voice dialogue system, 44: scenario data.

Claims (8)

An extraction means for extracting a word or phrase whose appearance frequency in the utterance of the conversation partner is a predetermined value or more;
Generating means for generating text indicating the content of the utterance related to the word or phrase extracted by the extracting means;
An interactive apparatus comprising: output means for outputting voice based on the text generated by the generating means.   The interactive apparatus according to claim 1, further comprising a changing unit that changes the predetermined value in accordance with a given instruction. Search means for searching for a word or phrase related to the word or phrase extracted by the extraction means,
The said generation means produces | generates the text using the word or phrase obtained by the said search means as a text which shows the content of the said related utterance, It is characterized by the above-mentioned. Interactive device.   4. The dialogue according to claim 3, wherein the search means searches for the related word or phrase based on a database on a network using the word or phrase extracted by the extraction means as a keyword. apparatus. Registration means for registering the word or phrase extracted by the extraction means;
A determining unit that determines which word or phrase is used to generate the text with reference to the word or phrase registered by the registering unit when there are a plurality of related words or phrases. The interactive apparatus according to claim 3 or 4, characterized by comprising:   The dialogue apparatus according to any one of claims 3 to 5, further comprising a conflicting unit that outputs a voice indicating that a conflict has been made during the search by the search unit.   A program for causing a computer to function as each means in the interactive apparatus according to claim 1. An extraction step in which the dialogue device extracts words or phrases whose appearance frequency in the utterance of the dialogue partner is equal to or higher than a predetermined value;
A generating step in which the interactive device generates text indicating the content of the utterance related to the word or phrase extracted in the extracting step;
And an output step of outputting voice based on the text generated in the generation step.

Images