JP4228442B2 - Voice response device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To inform a user of the reliability of an answer sent back by an answering system to a question inputted by the user and also to adaptively vary the detailedness of the answer. SOLUTION: An input analysis block 2 makes a retrieval block 5 to retrieve data according to information inputted from an input block 1, an answer text generation block 4 generates an answer sentence from the retrieved data, and a voice synthesis block 3 outputs a synthesized voice of the answer sentence. The answer text generation block 4 generates the answer sentence on which the reliability of the answer is reflected, varies the abstraction degree of the expression of the corresponding answer sentence according to the accuracy of the retrieved data, and also varies the abstraction degree of the answer sentence at a request for accuracy estimated from input information and a history of a history management block 9.

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a system that responds to, for example, a question input from a user or the like, and more particularly, to a voice response device that responds by natural language speech.
[0002]
[Prior art]
Conventionally, for example, there is a system that responds to a question or the like input from a user or the like. As an example of the response system, a voice response device that responds to a question from a user or the like with a natural language voice is known.
[0003]
There are also speech synthesis systems and speech synthesis methods that change the response speech depending on the content of the response and the source of the information.
[0004]
[Problems to be solved by the invention]
By the way, for example, when a response is returned from a response system to a question input by a user or the like, there is a case where it is desired to inform the user of the reliability of the response. However, many conventional voice response devices have a constant tone and sound quality. In this case, the user cannot know the reliability unless he / she listens to the details of the response.
[0005]
For example, even if a system that changes the response voice according to the content of the response and the source of the information is used, it is not sufficient to convey the reliability of the response. For example, if there is an anxiety in understanding the user's question, it can be said that the reliability of the response is low. On the other hand, even if the details of the information used for the response are uncertain, it is considered that a sufficient response can be made if the user's request does not require details.
[0006]
Furthermore, adaptively changing the detail level of the response is also an important point in the voice response. That is, for example, if it is explained in detail more than necessary with the audio output, it cannot be said that the response is appropriate. As one method of response in such a case, as disclosed in Japanese Patent Laid-Open No. 8-137698, only the explanation of the part that seems necessary is reflected, reflecting the form and history of the question from the user. A method can be considered in which the others are performed and others are omitted. As another method, the degree of detail may be changed depending on the level of abstraction when data is converted into a response sentence. For example, when notifying the date of a certain event, whether to notify the “month and season” of when, or to the specific “date and time”.
[0007]
However, in the conventional voice response system, when data is converted into a response sentence, data items are often put in a prepared response sentence template, and data is converted into a response sentence in a one-to-one correspondence. Therefore, it cannot cope with abstraction for a simple response.
[0008]
Therefore, the present invention has been made in view of such a situation, and when returning a response from a response system to a question input from a user or the like, it is possible to inform the user about the reliability of the response. Another object of the present invention is to provide a voice response device that can adaptively change the level of detail of a response.
[0010]
[Means for Solving the Problems]
The voice response device of the present invention is As a question Based on input information and input means to input information Respond to questions data of It has a search means for performing a search, a response sentence generation means for generating a response sentence from the searched data, and a synthesized sound conversion output means for converting the response sentence into a synthesized sound and outputting it. When generating a response sentence based on the processed data, The expected accuracy of the response from the question entered Of the retrieved data The result of comparing accuracy Depending on the response Predetermined Abstraction Word or ending expression according to By changing the above, the above-described problems are solved.
[0012]
That is, in the present invention, when a response is returned to a question input by a user or the like, the tone of the response sentence, voice, etc. are changed according to, for example, the type of data, certainty, detail, amount, etc. Thus, it is possible to inform the user about the reliability of the response, and it is possible to adaptively change the level of detail of the response, and to output a response sentence in a concise manner according to the point instructed by the user. I have to.
[0013]
DETAILED DESCRIPTION OF THE INVENTION
A preferred embodiment of the present invention will be described with reference to the drawings.
[0014]
FIG. 1 shows a configuration example of a voice response system according to an embodiment to which the voice response device of the present invention is applied.
[0015]
First, as the voice response system configuration according to the first embodiment, an information search system that searches for information corresponding to a question from a user, generates a voice from the searched information, and outputs the information is taken as an example.
[0016]
In FIG. 1, an input block 1 is an input conversion means for receiving an input from a user and converting the input into a numerical value or a symbol or converting it into a text. Here, the input from the user is, for example, a keyword, a command, a question, etc., and these inputs include, for example, an input by a user operating a button, a keyboard, a touch panel, etc., or an input by voice to a microphone. Conceivable.
[0017]
The input analysis block 2 receives the processing result of the input block 1, interprets the user input, decides what kind of information is used as a keyword to search for data, and informs the search block 5 of the decision result. . In addition, the input analysis block 2 determines how much the data obtained as a result of the search in the search block 5 is to be output as voice, and informs the response text generation block 4 of the determination result.
[0018]
The history management block 9 has a role to compensate for the lack of information for making a decision in the input analysis block 2. It manages the history of the user's past input and the response output made to it. The history management block 9 sends an answer to an inquiry about insufficient information from the input analysis block 2.
[0019]
The search block 5 searches for desired data from the information stored in the database block 6 based on the information from the input analysis block 2. Note that the data obtained as a result of the search includes information that the user does not want to know or information more detailed than necessary, but the search result data is sent to the response text generation block 4 as it is.
[0020]
The database block 6 is a place where information to be searched is recorded. The information stored in the database block 6 includes information that the user edits and adds later, such as a schedule and a telephone book, in addition to information prepared on the system side such as a map and a dictionary. In addition, information acquired via a network, such as traffic information, news, and weather forecasts, is also included.
[0021]
The response text generation block 4 generates response text based on the data received from the search block 5. That is, in the response text generation block 4, for example, based on information from the input analysis block 2, a data item not requested by the user is cut or an abstraction level of the expression is determined. The response text generation block 4 draws words from the text generation data storage block 8 and connects them to generate text, and determines the quality and size of the voice when the text is output as speech.
[0022]
In the text generation data storage block 8, a word corresponding to each element of the data and a ending set for expressing the characteristics of the output information are prepared. There is a one-to-many correspondence between data and words. For example, the expressions “October 5”, “early October”, and “this autumn” all correspond to one data. Which one is selected depends on the accuracy of the data and the abstraction level of the expression determined by the speech text generation block 4. In addition, the ending is used properly so that the features of the output information can be conveyed to the user sensuously. For example, the ending is "~" when the response reliability is high, and "~ kana" when the response is low.
[0023]
The speech synthesis block 3 receives the text and the output method from the response text generation block 4. The speech synthesis block 3 acquires necessary data from the speech synthesis data storage block 7 based on them, and combines them to generate synthesized speech data. The synthesized text data of the response sentence generated by the speech synthesis block 3 is finally sent as a voice signal to the speaker, and the synthesized voice of the response sentence is output from the speaker.
[0024]
The speech synthesis data storage block 7 is provided with a data set such as phonemes for converting text into synthesized speech. The information prepared in the speech synthesis data storage block 7 is properly used so that the characteristics of the output information can be conveyed to the user sensuously. That is, in the system of the present embodiment, for example, an adult voice is used when the response reliability is high, and a child voice is used when the response is low.
[0025]
Here, factors that determine the reliability of the response include, for example, confidence in the interpretation of the user input, the origin of information used for the response, and the accuracy that the user expects from the response.
[0026]
First, an example in which confidence in the interpretation of the user input is an element that determines the reliability of the response will be specifically described.
[0027]
For example, when a user inputs a question “What is the time from Tokyo to Shizuoka?”, The question is about the means of transportation such as whether to use a car or train to travel from Tokyo to Shizuoka. Is missing. Here, it is assumed that the input analysis block 2 manages the history of user input stored in the history management block 9, and it is found from the history of the history management block 9 that, for example, the user often moves by car. . In that case, the system according to the present embodiment does not ask the user about the contents of the moving means missing in the user's question, but can check the moving time in the vehicle and return a response for the time being. However, in this case, since it is unclear whether or not it is correct to return the travel time by car in the response, the reliability of the response is reduced accordingly. That is, the confidence in the interpretation of the user input indicates, for example, how certain the interpretation is when the user input is interpreted as described above, and the reliability of the response is determined by the confidence in the interpretation of the user input. Is determined.
[0028]
Next, an example in which the source of information used for the response is an element that determines the reliability of the response will be described in detail.
[0029]
For example, when calculating the travel time by car from Tokyo to Shizuoka, for example, the method of calculating the travel time by car using a road map or the travel time by car using traffic information There are methods. When the road map is used, for example, the road distance between two points of the starting point and the destination is calculated, and the travel time is obtained by dividing the road distance by the expected speed. In addition, when using traffic information, for example, check the time required between each measurement point such as traffic volume and traffic jam, which is provided between two points of departure and destination, and move by summing them. Ask for time. As described above, when there are two sources of information used for response, when using a road map and when using traffic information, the above traffic information that can take into account real-time traffic volume and traffic information etc. When used, the response is more reliable. Thus, the reliability of the response is determined by the source of the information used for the response.
[0030]
Next, an example in which the accuracy expected by the user for the response is an element that determines the reliability of the response will be described in detail.
[0031]
For example, it is assumed that a calculation result of 2.1 hours as a travel time is obtained by calculating the travel time by car from Tokyo to Shizuoka from a road map. Here, for example, if there is only accuracy in units of one hour, it can be considered that the response “about 2 hours” is reliable for the user's question “What is the time from Tokyo to Shizuoka?”. However, the response “2 hours 6 minutes” to the user's question “How long does it take from Tokyo to Shizuoka?” Is considered to be unreliable. That is, even if the responses are from the same data, the reliability of the responses changes depending on the comparison with the user's expectation. Thus, the reliability of the response is determined by the accuracy that the user expects for the response.
[0032]
On the other hand, factors that determine the level of abstraction of the data representation include, for example, the accuracy of the data itself and the accuracy that the user expects from the response.
[0033]
First, an example in which the accuracy of the data itself is an element that determines the level of abstraction of the data representation will be specifically described.
[0034]
For example, it is assumed that a calculation result of 2.1 hours as a travel time is obtained by calculating the travel time by car from Tokyo to Shizuoka from a road map. In such a case, as a way of expressing the response to the user, 2.1 hours that is the calculation result of the travel time can be expressed as “2 hours 6 minutes”, “2 hours a little”, Alternatively, the expression can be changed according to an assumed error so as to express “about 2 hours”. In this way, the abstraction level of the data representation of the response is determined according to the accuracy of the data itself.
[0035]
Next, an example in which the accuracy expected by the user for the response is an element that determines the abstraction level of the data expression will be described in detail.
[0036]
For example, in response to a question “What are your plans for tomorrow?” From the user, a response “tennis in the morning” is made. On the other hand, in response to the question “What is your plan for tomorrow in the morning?” From the user, the answer is “tennis from 9 o'clock to 11 o'clock”. In this way, the degree of abstraction of the data representation is determined by the accuracy that the user expects for the response. Note that the accuracy expected by the user for the response can be estimated not only from the shape of the question input from the user but also from the past history. For example, for a user who always asks for an accurate time, when asked “What are your plans for tomorrow?”, Answer “tennis from 9 o'clock to 11 o'clock”.
[0037]
Hereinafter, the operation of the system according to the first embodiment will be specifically described with reference to FIGS. 1 and 2.
[0038]
Here, it is assumed that when a user inputs a question together with a keyword, a response by voice output is returned from the system of the present embodiment.
[0039]
First, as an initial setting, a mode is set in which the user operates the input block 1 to obtain information on the road.
[0040]
Next, as step S1, the user inputs a question "What is the time from Tokyo to Shizuoka?" Here, it is assumed that a voice input is assumed as an interface for inputting a question, and the user is directed toward the microphone of the system. Input speech from the microphone is subjected to speech recognition processing in the input block 1 and converted into text data.
[0041]
The input analysis block 2 analyzes the text data sent from the input block 1 and determines a corresponding search process. That is, the input analysis block 2 has a list of keywords for determining processing, and first searches for a word that matches the keyword.
[0042]
Here, from the question “What is the time from Tokyo to Shizuoka?”, Find three words that match the keyword: “Tokyo”, “Shizuoka”, and “Time”. At this time, although it can be understood from the place name such as “Tokyo” and “Shizuoka” and the keyword “time”, the transportation means (movement means) cannot be understood. Therefore, the input analysis block 2 makes an inquiry to the history management block 9. If it is found from this inquiry that, for example, a car has been selected as a transportation means in the past, the input analysis block 2 assumes that the transportation means is a car in step S10 based on this. As a result, although the user does not specify the transportation means (movement means), it is assumed that there is a high possibility of a car from the history and that the user input is confident.
[0043]
Next, the input analysis block 2 instructs the search block 5 to “acquire traffic information on the Tomei Expressway” as step S2, and also instructs the response text generation block 4 “from the searched information. Calculate and output the required time between Tokyo and Shizuoka. Further, the input analysis block 2 determines that it is sufficient to inform about the time as a response from the form of the user's question, and expresses the minute unit of time to the response text generation block 4 in an abstract manner. To instruct.
[0044]
In response to an instruction from the input analysis block 2, the search block 5 accesses the database block 6 and obtains traffic information. At this time, the search block 5 examines information stored locally in the database block 6 which is a storage unit of the system. If the information in the database block 6 is old or not, for example, the latest information is obtained via the network. Get information. It is assumed that the traffic information is obtained from where on the network and the operation of the system for obtaining the information is determined in advance. The traffic information obtained by the search in the search block 5 is sent to the response text generation block 4.
[0045]
The response text generation block 4 at this time is informed from the input analysis block 2 that “the time required between Tokyo and Shizuoka is calculated and output”. First, as a step S 3, the response text generation block 4 is obtained by searching the search block 5. For example, the time required for each section of “Tokyo-Atsugi”, “Atsugi-Gotemba”, “Gotemba-Shizuoka” is picked up as necessary data, and the total time (215 minutes in this example) is calculated as step S4. To do.
[0046]
Next, in the response text generation block 4, a response text is generated. First, the expression of time is determined. In the case of the above example, according to the instruction from the input analysis block 5, the minute unit is determined as an abstract expression. In determining this abstract expression, the response text generation block 4 refers to the text generation data storage block 8 and uses, for example, “(don't say)”, “a little”, “half” as abstract expression words. For example, the expression “half” is selected from the four “weak”, and the time of 215 minutes is changed to the expression “three and a half hours” in step S5. In addition, the calculated time is based on traffic information and is highly accurate. It is determined that there is a text, and an instruction regarding text synthesis and voice synthesis for outputting a voice with a large and clear tone of an adult voice is output to the voice synthesis block 3.
[0047]
In the speech synthesis block 3, a text “3 hours and a half” and an instruction regarding speech synthesis are received from the response text generation block 4. As a result, the voice synthesis block 3 generates voice output data of a large and clear tone of the adult voice “3 hours and a half” as step S11, and outputs it as a response from the speaker.
[0048]
By the way, the information source for examining “time from Tokyo to Shizuoka” is not limited to traffic information. Although it is inaccurate, it is possible to estimate the required time by examining the distance from the road map.
[0049]
In this case, the input analysis block 2 informs the search block 5 at step S6, such as “Get the map information of the Tomei Expressway”.
[0050]
The search block 5 at this time accesses the database block 6 and obtains map information. Information obtained by the search block 5 searching the database block 6 is sent to the response text generation block 4.
[0051]
The response text generation block 4 at this time is informed from the input analysis block 2 that “the required time between Tokyo and Shizuoka is output”, and the map obtained by the search of the search block 5 as step S7. For example, the distance of each section of “Tokyo-Atsugi”, “Atsugi-Gotemba”, and “Gotemba-Shizuoka” is picked up as necessary data from the information, and the total distance (210 km in this example) is calculated as step S8. Furthermore, in the response text generation block 4, the moving speed of the vehicle is set to 100 km / h on average, and the required time of 2.1 hours is estimated from the total distance and the average speed.
[0052]
Next, in the response text generation block 4, a response text is generated. Also in this case, first, the expression of time is determined. In the case of the above example, according to the instruction from the input analysis block 5, the minute unit is determined as an abstract expression. In determining this abstract expression, the response text generation block 4 refers to the text generation data storage block 8 and, as the word of the abstract expression, “(Do not say anything)” “A little” “ For example, the expression “(Do not say anything)” is selected from the four of “half” and “weak”, and the above 2.1 hours is changed to “2 hours” in step S9. In addition, the calculated time is assumed to be appropriate for the approximate time, and the ending of the expression is made to be an all-out type “is”. However, when using map information, the response is less reliable than when using the traffic information described above, so instructions for text-to-speech synthesis for voice output of an adult voice but a low voice Is output to the speech synthesis block 3.
[0053]
In the speech synthesis block 3, the text “2 hours” and an instruction regarding speech synthesis are received from the response text generation block 4. As a result, the voice synthesis block 3 generates voice output data of a small voice, although it is an adult voice “two hours”, in step S12, and outputs it as a response from the speaker.
[0054]
In addition, when calculating the time from the road map as described above, if the question from the user is not "What is the time from Tokyo to Shizuoka?" But "How many minutes from Tokyo to Shizuoka?" Response is possible.
[0055]
In this case, the input analysis block 2 determines that the user desires a response of expression up to the minute unit.
[0056]
Thereby, in the response text generation block 4, the 2.1 hours obtained by the above calculation is expressed as “2 hours 6 minutes” based on this. However, if the unit of the calculation result is unreliable, the “6 minutes” portion of the response is likely to be inaccurate (lie). For this reason, in the response text generation block 4, since the reliability of the response is not sufficient, “kana” expressing ambiguity is used for the ending. Then, in the response text generation block 4, for example, a text for voice output with a child's voice and an instruction regarding voice synthesis are output to the voice synthesis block 3.
[0057]
In the speech synthesis block 3, the text “2 hours 6 minutes” and an instruction regarding speech synthesis are received from the response text generation block 4. As a result, the voice synthesis block 3 generates voice output data of a child's voice “2 hours 6 minutes” and outputs it as a response from the speaker.
[0058]
Furthermore, for example, if the policy is not to represent data that cannot guarantee accuracy, when calculating the time from the road map as described above, if the question from the user was "How many minutes from Tokyo to Shizuoka?" A response such as
[0059]
In the case of this example, the input analysis block 2 determines that the user desires a response of expression up to the minute unit.
[0060]
Accordingly, the response text generation block 4 tries to express 2.1 hours obtained by the above calculation as “2 hours and 6 minutes” based on this. However, here, assuming that the minute unit of the calculation result is unreliable, the “6 minutes” portion of the response is likely to be inaccurate (lie). Therefore, in the response text generation block 4, the level of abstraction of the expression is increased and the expression “2 hours” is determined without saying a specific minute unit. Further, the response text generation block 4 adds “about” to the end of the word to reflect that it cannot be answered more specifically. However, in this case, since the user's request is not satisfied as a response, the response text generation block 4 outputs to the speech synthesis block 3 an instruction relating to text and speech synthesis, for example, for voice output by a child's voice. .
[0061]
In the speech synthesis block 3, the text “It is about 2 hours” and an instruction related to speech synthesis are received from the response text generation block 4. As a result, the voice synthesis block 3 generates voice output data of a child's voice “about 2 hours” and outputs it as a response from the speaker.
[0062]
Next, as a voice response system configuration according to the second exemplary embodiment of the present invention, a scheduler that manages a user's schedule is taken as an example. Hereinafter, the operation of the system according to the second embodiment will be specifically described with reference to FIGS. 1 and 3.
[0063]
First, as an initial setting, a mode is set in which the user operates the input block 1 to obtain personal information such as a schedule and a telephone number.
[0064]
Next, as step S31, the user inputs a question such as "What is your plan for tomorrow?" Here, it is assumed that a voice input is assumed as an interface for inputting a question, and the user is directed toward the microphone of the system. Input speech from the microphone is subjected to speech recognition processing in the input block 1 and converted into text data.
[0065]
The input analysis block 2 analyzes the text data sent from the input block 1 and determines a corresponding search process. That is, the input analysis block 2 has a list of keywords for determining processing, and first searches for a word that matches the keyword.
[0066]
Here, from the question “What is the schedule for tomorrow?”, Two words “tomorrow” and “schedule” are found as words that match the keyword. Then, the input analysis block 2 determines to search the schedule from the keyword “schedule”, and determines to search for tomorrow from the keyword “tomorrow”.
[0067]
Therefore, the input analysis block 2 instructs the search block 5 to “obtain tomorrow's schedule data” as step S32, and the response text generation block 4 “notifies all tomorrow's schedule”. " However, in this case, the input analysis block 2 determines that the user wants to know the outline from the form of the user's question, and does not say the detailed time and details of the schedule.
[0068]
The search block 5 that has received the search instruction from the input analysis block 2 accesses the database block 6 and retrieves tomorrow's schedule as step S33. It is assumed that information extracted at this time is “9: 00-11: 0 tennis at Shinagawa” and “14: 00-shopping at Shinjuku with friends”. Information obtained by the search in the search block 5 is sent to the response text generation block 4.
[0069]
Next, the response text generation block 4 generates response text. At this time, the time is first determined for the expression, but since the input analysis block 2 is informed that “the outline of tomorrow's schedule is output” as described above, the response text generation block 4 Decides to increase the level of abstraction of the expression. For example, referring to the text generation data storage block 8, there are, for example, three categories of “morning”, “daytime”, and “night”, and the response text generation block 4 is “9: 00-11: It is determined that “00” corresponds to “morning” and “14: 00- (after 14:00)” corresponds to “daytime”. Also, for the contents of the schedule, omit the place and partner, and say only the business. That is, the response text generation block 4 generates response texts such as “AM tennis” and “PM shopping” in step S34.
[0070]
Next, the response text generation block 4 determines an instruction relating to the sentence ending and speech synthesis. As in the second embodiment, in the case of the scheduler, there is no other mistake in interpreting the user input, and the response data is also entered by the user himself. Reliability is sufficient. Therefore, in step S35, the response text generation block 4 uses the utterance type “Da” as the ending expression, and the voice synthesis block gives instructions for text and voice synthesis for voice output in a large and clear tone of an adult voice. 3 is output.
[0071]
The speech synthesis block 3 receives from the response text generation block 4 a text “sound in the morning, shopping in the afternoon” and an instruction regarding speech synthesis. As a result, the speech synthesis block 3 generates voice output data with a large and clear tone in a voice of an adult saying “Tennis in the morning and shopping in the afternoon” as step S39, and outputs it as a response from the speaker.
[0072]
However, when the user asks "What is your plan for tomorrow?"
[0073]
For example, in the input analysis block 2, suppose that it is found from the information in the history management block 9 that “if you give an outline to the user, you will often be asked again for a detailed date and time”. In this case, in the input analysis block 2, the history is prioritized over the expression of the current input, and the response text generation block 4 is instructed to “make each tomorrow's schedule more detailed”.
[0074]
In this case, in the response text generation block 4, an instruction regarding text and speech synthesis for performing voice output in a loud and clear tone of an adult voice, such as “tennis from 9 am to 11 am, shopping from 2 pm”. Is output to the speech synthesis block 3.
[0075]
When the speech synthesis block 3 receives the instruction about the text and the speech synthesis from the response text generation block 4, the tone is large and clear with an adult voice saying “Tennis from 9 am to 11 am, shopping from 2 pm”. Is output as a response from the speaker.
[0076]
On the other hand, for example, consider a case where, after the user designates a mode for obtaining personal information by operating the input block 1, for example, a question “What are your plans for tomorrow in the morning?” Is entered as step S 36. Also in this case, as described above, it is assumed that a voice input is assumed as an interface for inputting a question, and the user is directed toward the microphone of the system. Input speech from the microphone is subjected to speech recognition processing in the input block 1 and converted into text data.
[0077]
In this case, the input analysis block 2 analyzes the text data sent from the input block 1 and determines a corresponding search process. The input analysis block 2 has a list of keywords for determining processing, and first searches for a word that matches the keyword.
[0078]
Here, from the question “What is your plan for tomorrow in the morning?”, Three words “Tomorrow”, “Morning” and “Schedule” are found as words that match the keyword. Then, the input analysis block 2 determines to “search for tomorrow's schedule” from the keyword combination of “tomorrow” and “schedule”. At this time, the input analysis block 2 instructs the search block 5 to “acquire tomorrow's schedule data” in step S32, and the response text generation block 4 “tomorrow's morning schedule”. Only tell me ". In the case of this example, the input analysis block 2 determines that the user wants to know the detailed time from the form of the user's question, and says the detailed time of the schedule and then the detailed content.
[0079]
The search block 5 that has received the search instruction from the input analysis block 2 accesses the database block 6 and retrieves tomorrow's schedule as step S33. Information obtained by the search in the search block 5 is sent to the response text generation block 4.
[0080]
Next, the response text generation block 4 generates response text. Since the response text generation block 4 at this time is informed from the input analysis block 2 that “only the schedule in the morning is output”, in the response text generation block 4, “14: 00−” is set as step S 37. The schedule is excluded from the output because it is not in the morning, but the time, place, and requirements are all left for output. Further, in the response text generation block 4, in order to inform the time as much as possible, “9: 00-11: 000” is expressed as “from 9 o'clock to 11”.
[0081]
Next, the response text generation block 4 determines an instruction relating to the sentence ending and speech synthesis. In this example, if there is no other mistake in interpreting the user input and the response data is correct by the user himself / herself, the reliability of the response is sufficient. Therefore, in the response text generation block 4, as the step S38, the ending expression is “Dai”, and “It is tennis in Shinagawa from 9:00 to 11:00” with a large and clear tone of adult voice. Text for voice output and instructions regarding voice synthesis are output to the voice synthesis block 3.
[0082]
When the voice synthesis block 3 receives an instruction regarding text and voice synthesis from the response text generation block 4, in step S 40, a voice with a large and clear tone is heard in an adult voice, “It is tennis in Shinagawa from 9:00 to 11:00.” Output data is generated and output as a response from the speaker.
[0083]
Next, FIG. 4 shows a hardware configuration for realizing the voice response system of the present invention.
[0084]
In FIG. 4, an input unit 11 corresponds to the input block 1 of FIG. 1, and includes interfaces such as buttons, a keyboard, a touch panel, and a microphone. A user inputs a question and a keyword from the input unit 11.
[0085]
The CPU 12 executes control of each part of the system and various program processes. The CPU 12 is responsible for various processes in the input analysis block 2, the speech synthesis block 3, the response text generation block 4 and the like in FIG.
[0086]
The ROM 13 stores fixed data and fixed programs. Various data for the text generation and speech synthesis described above and programs for processing are also stored in the ROM 13.
[0087]
The RAM 14 temporarily stores necessary data during program processing by the CPU 12. Information input from the user, text generation, speech synthesis, information obtained by the communication unit 16, and the like are temporarily stored in the RAM 14.
[0088]
The auxiliary storage unit 15 includes a flash memory and an EEPROM, and stores data that is to be kept at all times even during program processing, with data to be added or rewritten. Personal information of the history management block 9 in FIG. 1, data obtained by communication, other data used for processing of the CPU, and the like are stored in the auxiliary storage unit 15.
[0089]
The communication unit 16 is used when the system acquires or sends information via a network. For example, communication such as a telephone, the Internet, infrared rays, and radio is performed.
[0090]
The speaker unit 17 outputs a voice such as a voice synthesized response.
[0091]
The display unit 18 includes a display device such as a liquid crystal panel, and outputs a so-called GUI (Graphical User Interface) screen and displays various information such as a table and a list.
[0092]
In the system according to the embodiment of the present invention, the auxiliary storage unit 15, the communication unit 16, and the display unit 18 are not necessarily required.
[0093]
As described above, according to the embodiment of the present invention, the tone, sound quality, and the like when outputting a response to a question from the user are changed according to the reliability of the response. It is possible to listen to the audio output while knowing to some extent whether or not a response matching the request is returned. Also, according to the present embodiment, even when only listening to audio output, the system status becomes easy to understand, and the next countermeasure (for example, listening properly, entering a deficiency condition, asking another question, etc.) is quickly performed. It is possible. Furthermore, according to the present embodiment, the degree of accuracy can be expressed or the output can be simplified by changing the abstraction level of the data expression in the response output.
[0094]
【The invention's effect】
As is clear from the above description, the voice response device of the present invention performs a data search based on the input information, generates a response sentence from the searched data, and outputs a response as a synthesized sound. To generate a response sentence that reflects reliability, or to change the abstraction level of the expression of the response sentence according to the accuracy of the retrieved data, or to request the accuracy estimated from the input information and history By changing the level of abstraction of the response sentence accordingly, it is possible to inform the user about the reliability of the response when returning a response from the response system to the question entered by the user, etc. It is also possible to adaptively change the level of detail of the response.
[0095]
In other words, according to the present invention, by changing the tone, sound quality, etc. when outputting the response sentence in accordance with the reliability of the response, the user knows to some extent whether the response that meets his request is returned. On the other hand, it is possible to hear the audio output, and even when only the audio output is heard, the state of the system becomes easy to understand, and the next countermeasure is accelerated. Furthermore, according to the present invention, the degree of accuracy can be expressed or the output can be simplified by changing the abstraction level of the data expression in the response output.
[Brief description of the drawings]
FIG. 1 is a functional block diagram showing a schematic configuration of a voice response system according to an embodiment of the present invention.
FIG. 2 is a flowchart showing a flow of one operation example of the system according to the first embodiment;
FIG. 3 is a flowchart showing a flow of an operation example of the system according to the second embodiment;
FIG. 4 is a block circuit diagram showing a hardware configuration for realizing the voice response system of the present invention.
[Explanation of symbols]
1 input block, 2 input analysis block, 3 speech synthesis block, 4 response text generation block, 5 search block, 6 database block, 7 speech synthesis data storage block, 8 text generation data storage block, 9 history management block, 11 Input unit, 12 CPU, 13 ROM, 14 RAM, 15 auxiliary storage unit, 16 communication unit, 17 speaker unit, 18 display unit

Claims (3)

An input means for inputting information as a question ;
Search means for searching for data serving as a response to the question based on the input information;
Response sentence generation means for generating a response sentence from the retrieved data;
A synthesized sound conversion output means for converting the response sentence into a synthesized sound and outputting the synthesized sound;
In the response sentence generation unit, when generating a response sentence based on the data the search, depending on the result of comparison between accuracy of precision and data the retrieval expected for the response from the question that the input the A voice response device that changes a word or ending expression according to a predetermined abstraction level of a response sentence expression. 2. The voice response device according to claim 1, wherein the synthesized sound conversion output means changes the type or volume of the voice according to the comparison result . In the response sentence generating means, when generating a response sentence based on the searched data, the expression of the word or ending according to a predetermined abstraction level of the expression of the response sentence is changed according to the comparison result ,
The voice response device according to claim 1, wherein the synthesized sound conversion output means changes a type or a volume of the voice according to the comparison result .

Images