JP2006003743A - Speech dialogue method and device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech dialogue method and its device with which discrimination is made to discriminate the validity/invalidity of input information at the time of re-opening when the dialogue of a speech dialogue system is interrupted and then, it is re-opened. <P>SOLUTION: The speech dialogue device consists of an interactive history generating device 11 which generates a dialogue history of input information being inputted before the interrupted point of time of the dialogue and which consists of the control program position of the dialogue scenario that is being executed at that point of time and time information at that point of time, and of a dialogue history obtaining device 4 which takes out the dialogue history obtained by the dialogue history generating device 11 and judges whether the time information within the input information is still valid or not while referring to the time information in the input information within the taken out dialogue history to the time information at the point in time of dialogue reopening. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a voice interaction method and apparatus in a case where a dialogue process with a user is paused and then the dialogue process is restored (resumed).

  In a voice dialogue system that obtains a voice response output required for information processing by a computer in response to a voice input to the computer by a user, conventionally, the voice response on the computer side (system side) starts with a user input operation. By prompting the user for voice input or prompting for selection in the process, the dialog is sequentially advanced to obtain information required by the user. In this case, in general, such a series of dialogue between the user and the computer (system) is either progressed to the end of the dialogue when the dialogue starts, or it is necessary to cut off halfway and start over from the beginning. The control program that is interrupted in the middle of the dialog (or pauses the system) and resumes from the interrupted part of the dialog is not found in a system that is actually operating.

On the other hand, as a technical proposal, in Patent Document 1, a position where a conversation is being executed is marked and saved during a voice conversation, and then the conversation is not restarted from the beginning by inputting the saved mark. In addition, a voice dialogue system is disclosed in which a dialogue on the way can be omitted by jumping to the marked position. At this time, the information input in the dialog is also saved and can be recalled, so that it is possible to save the input contents and the trouble of re-input.
In Patent Document 2 and Patent Document 3, a dialog system that manages information on what the current topic is (the focus of the topic) as a stack is disclosed, and the system was about to advance to that point in the middle of a certain voice conversation When a new question related to a topic different from the topic is input as a user's utterance, focus information including the goal of answering the new question as a new topic is generated, and the focus information is managed. The stack area is stacked on top of the focus information of the topic at the top (push), and when the new topic is completed and the goal is achieved, the focus information of the completed topic is removed (popped) from the stack. ) By making it possible to refer to the focus information of the original topic again, it is possible to respond while maintaining the consistency of the system utterance about the temporary suspension and return of a topic That.

According to Patent Documents 1 to 3 described above, when a dialog being executed in the voice dialog system is temporarily interrupted, the interrupt position is marked and stored together with the input information, and stored when the interrupted dialog is restored. It becomes possible to resume the information.
For example, when this is embodied as a weather forecast in which the user's desired date and area are determined, the user wants to know the weather of “March 23” and “Eastern Kanagawa” on March 22, for example. It is assumed that the user (User) (supplied as U) and the system (System) (assumed as S) have exchanged (interactive) input voice and output voice.
Now, when the output from the system S is confirmed as “March 23 weather in eastern Kanagawa”, the user U inputs “suspend”, and then on March 24, the dialog “Resume” "Is input as the output sound from the system S, and the process restarts from the confirmation of" the weather on March 23 in eastern Kanagawa "at the time of the interruption. That is, the output voice at the time of the interruption such as “Would you like the weather on March 23 in eastern Kanagawa Prefecture?” Is repeated and spoken from the system. However, since March 23 is already in the past and the weather on March 23 is meaningless for the user, the input voice of the user U is naturally “No”, and therefore the output from the system As a voice, an inquiry for correction of information such as "How should we fix March 23 in eastern Kanagawa Prefecture?" Then, the user U inputs a date-designated utterance such as another date “March 25”.
JP 2003-223187 A JP 2001-142870 A JP 2001-356797 A

As described in the weather forecast of the specific example above, in the normal weather forecast, the purpose of the dialog system is to know the future date / time forecast. After the interruption on March 24 in the above dialog example, The date information “March 23” in the input information before the interruption is already invalid (not valid).
However, in the conventional interactive system, the input information before the interruption is repeated as it is after the restart, and there is date information that is no longer valid as the input information after the interruption, that is, the presence of invalid input information is detected. Can not do it. For this reason, there is a problem in that the user himself / herself confirms that there is date information that is not valid in the input information before the interruption, which imposes a burden.

  Also, in the dialogue system according to the prior art, it is impossible to recognize which information of “region” or “date” is invalid in the input information before the interruption (in the specific example, it is a date). Therefore, the dialogue system presents both “region” and “date” and asks for correction of information. For this reason, in order to recognize the correction of this information, it is necessary to make it possible to recognize both the “region” and “date” corrections. This means that a speech recognition grammar and dictionary for recognizing a wide range of utterances must be used, and there arises a problem that the recognition accuracy decreases as the speech recognition target becomes wider.

Furthermore, in the prior art, as shown in the specific example, since the dialogue at the time of interruption is repeated as it is, the word “Now” is used as it is, and the expression of the system response sentence completely considers the time interval from interruption to resumption. The user's response is a sudden and unnatural system.
Considering the above, it is said that the validity / invalidity of the time information in the input information in which the validity / invalidity of the input information is noticeable at the time of resumption is not determined at the time of the interruption or the subsequent resumption during the conversation There is a problem, and due to the fact that this is not discriminated, the above-mentioned derivative problem is envied.

  The present invention has been invented to solve the above-described problem, and a voice interaction method and apparatus for determining whether input information at the time of resumption is valid or not with respect to interruption or subsequent resumption during a conversation of the voice interaction system. The purpose is to provide.

 The present invention which achieves the above-mentioned object is the voice information processing the voice input and outputting the response voice. When the interruption of the dialogue is inputted, the input information inputted up to the point of interruption of the dialogue and being executed at that time Dialog history generation process for generating a dialog history consisting of the control program position of the dialog scenario and the time information at that time, and when dialog resumption is input, the dialog history obtained in the dialog history generation process is extracted. A dialogue history acquisition process for determining whether the time information in the input information is still valid or invalid by comparing the time information in the input information in the extracted dialogue history with the time information at the time of restarting the dialogue. It is characterized by.

  According to the present invention, it is possible to automatically determine the validity or invalidity of time information such as date and time according to the purpose of a dialogue scenario when a suspended dialogue is resumed in a voice dialogue system. Become.

Now, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 shows an overall schematic block configuration of a voice interaction apparatus according to an embodiment of the present invention. In this spoken dialogue apparatus, first, the dialogue control device 1 performs dialogue based on the input from the utterance understanding device 3 and the dialogue history acquisition device 4 as input devices in accordance with the contents of the dialogue scenario that is advanced by the execution of the dialogue control program 2. The control calculation of the entire apparatus is performed while updating the control memory 5. The object to be controlled by the dialogue control device 1 is an input device. The speech recognition device 6 recognizes speech by the recognition dictionary 7 and the recognition grammar 8 after obtaining the user's speech input, and the speech recognition device 6. An utterance understanding device 3 that recognizes words and phrases and obtains an utterance understanding result, an audio output device 9 that outputs response speech generated by the system, input information to be stored in the dialogue history storage device 10 when a dialogue scenario is interrupted, and The dialog history generating device 11 for extracting the dialog control program position from the dialog control memory 5, the input information from the dialog history generating device 11 stored in the dialog history storage device 10 and the dialog history information such as the dialog control program position are extracted and described later. Has a dialog history acquisition device 4 that performs a determination based on the progress flag or calculation of the interruption elapsed time amount information. The dialog history generation device 11 and the dialog history acquisition device 4 are connected to a clock 12 for obtaining current time information at that time.

In the configuration of FIG. 1, first, the speech recognition device 6 is based on a control signal from the dialogue control device 1 and a recognition dictionary 7 storing words in the sentence to be recognized and information on their pronunciation, and in the sentence to be recognized. The recognition grammar 8 representing how words are arranged is used to recognize a user's voice and output a word string.
Based on the control signal from the dialogue control device 1, the utterance understanding device 3 extracts a word meaningful to the voice dialogue device from the word string output from the voice recognition device 6 and outputs it as the utterance understanding result or the voice dialogue. Phrases that are meaningful to the device are identified and output as utterance comprehension results. The utterance understanding result that is the output information of the utterance understanding device 3 becomes input information from the user to the dialogue control device 1. If the speech recognition device 6 recognizes only an isolated word as the recognition result, the recognition result may be used as it is as input information to the dialog control device 1, and in this case, the speech understanding device 3 is used. May be omitted.

The voice output device 9 uses a voice synthesizer (not shown) or the like based on a control signal from the dialogue control device 1 based on a character string instructed from the dialogue control device 1 or information indicating the meaning (concept) of the response. Synthesize speech and output as system response. In this case, by reproducing the voice information recorded or synthesized in advance instructed from the dialogue control device 1, it is output as a system response to the user.
As the voice dialogue system, the dialogue control device 1 that captures the input information from the voice recognition device 6 and the utterance understanding device 3 creates a response sentence by calculation while interacting with the dialogue control memory 5 according to the dialogue scenario. Then, a response voice is output from the voice output device 9.

  Here, when interruption and resumption occur during the dialogue, processing by software is mainly performed in the dialogue history generation device 11 and the dialogue history acquisition device 4. That is, the dialogue history generation device 11 is based on the control signal from the dialogue control device 1 and the input information stored in the dialogue control memory 5 and the dialogue control program position (dialog scenario) which is the execution position in the dialogue scenario. Position) is taken out, current time information taken out from the timepiece 12 is used as time information, and these are combined and stored in the dialogue history storage device 10 as dialogue history information.

FIG. 5 exemplifies processing software in the dialog history generation device 11. As a flowchart, the input information and the dialog scenario position stored in the dialog control memory 5 are acquired (step 51), and the current time is displayed by the clock 12. Time information is obtained and time information is obtained (step 52), and each piece of input information, dialogue scenario position and time information is generated as dialogue history information (step 53).
The dialogue history storage device 10 is a storage device that stores at least dialogue history information from the dialogue history generation device 11, and maintains the memory without being erased even when the dialogue control device 1 is interrupted or powered off. This dialogue history storage device 10 is a non-volatile memory such as a flash memory, a hard disk drive, or a relational database connected via a network other than the voice dialogue device main body on which the dialogue control device 1 is operating. It can be realized using an apparatus or the like.

  Further, the dialog history acquisition device 4 acquires corresponding dialog history information from a plurality of dialog history information in the dialog history storage device 10 based on a control signal from the dialog control device 1, and from the dialog history information. The input information and dialogue scenario position and time information are taken out, and the taken time is acquired by the clock 12 as time information, and the elapsed time amount from the time information in the dialogue history information to the time information at the time taken out is calculated. Get information on the amount of elapsed time for interruption. Here, in addition to the time information at the time of dialog history information generation and the time information at the time of dialog history information retrieval, the time information of the input information itself due to the presence of time information indicating the time such as date and time in the input information itself. Is compared to the time information at the time of retrieving the conversation history information. For the purpose of using the dialog system, the time information of the input information itself is invalid (invalid), and the time information is valid (false) A progress flag indicating (valid) is calculated for each piece of input information. The dialog history acquisition device 4 controls dialog control using the input information, the dialog scenario position, the time information, the progress flag, the time information of the dialog history information extraction time, and the interruption elapsed time amount information in the dialog history information as the dialog history acquisition information. Output to device 1. The processing software of the dialog history acquisition device 4 is illustrated in the flowchart shown in FIG. 6, but the description of FIG. 6 will be described together with the operation described later. Further, the process of calculating the progress flag for the input information is illustrated in FIG. 7, and it is determined that the required time level and the same time information from the dialog control device 1 are valid according to the purpose of use of the dialog system. Flag indicating whether or not to determine invalidity is instructed, and the time information in the input information and the time information at the time of extracting the conversation history information are calculated. Further, the clock 12 returns current time information indicating the time when the request command is received to the request source when there is a request command.

The dialog control memory 5 stores the input information and the execution position of the dialog control program (dialog scenario position) during the execution of the dialog control device 1. Further, in the dialog control memory 5, when the dialog control apparatus 1 receives the dialog history acquisition information from the dialog history acquisition apparatus 4, each of the input information, the progress flag, the dialog scenario position, and the interruption elapsed time information are stored. Saved.
The dialogue control program is based on the contents of the dialogue control memory 5 and the utterance understanding result output from the utterance understanding device 3, and the speech recognition device 6, the utterance understanding device 3, the voice output device 9, the dialogue history generation device 11, and the dialogue history acquisition. Information on control and instructions from the dialog control device 1 to the device 4 and the dialog control memory 5 is described, and is read and executed by the dialog control device 1.

Next, the overall operation of the voice interaction system in this embodiment will be described with reference to FIG. In this embodiment, the voice input of the user and the response output of the system are input, processed, and output in accordance with the dialog control program. In this operation description, the operation of interrupting and resuming (returning) the dialog is described. Put on the main focus. Of course, in a state where no interruption or resumption occurs, a dialogue for obtaining a voice output with normal voice input information is performed.
Further, as a premise of this operation description, identification of a plurality of users is performed by ID information such as a telephone number in the case of a telephone and an address and a password in the case of a network, for example. Therefore, the matching of the dialogue at the time of interruption or resumption is performed, for example, by checking the coincidence of the ID information. Furthermore, when the dialog is interrupted or resumed, the time information in the input information is compared with the time information at the time of restart to determine whether the time information in the input information is valid or invalid according to the purpose of use of the dialog system. The time information is not limited to the time information at the time of interruption, and it is also possible to determine whether the time information in the input information at any time before the interruption is valid or invalid. However, in the description of the operation here, first, a case where time information is included in the input information at the time of interruption will be described.

  FIG. 2 shows a general schematic flowchart of an interaction scenario. When there is an initial utterance from the user and there is a match of the user ID, a start response from the system is output as shown in FIG. 2 (step 21). Thereafter, it is determined whether or not the utterance from the user is a resumption (return) utterance (step 22), and in the case of resumption, resumption processing as described later is performed (step 23). In the case of utterance from the beginning of the dialogue scenario instead of restarting in step 22, or after the restart processing in step 23, processing for each unit scenario is performed in accordance with the normal dialogue scenario (step 24), then all It is determined whether or not the unit scenario has been processed (step 25), and if all the scenarios have not been processed yet, processing for shifting to the next unit scenario (step 26) is performed.

  Next, the voice dialogue system will be described more specifically with the weather forecast in mind. FIG. 3 is a specific flowchart of a dialogue scenario that is obtained as input information by voice from a user and specifies a region and a date. Here, the region acquisition process includes the restart determination, and the unit scenario process (step 24) in FIG. 2 is area acquisition 1, date acquisition 1, input confirmation 1, and re-input 1. In FIG. 3, a processing unit indicated by a rectangle is a unit scenario process, and is recorded as a dialog control program position (dialog scenario position) in the dialog control memory 5. However, “interrupt process 1” and “interrupt process 2” are not recorded in the dialog control memory 5 as dialog scenario positions.

In FIG. 3, first asking for the area as the area acquisition 1 (step 31), asking the date while repeating the area as the date acquisition 1 after voice input for the area (step 32), after inputting the voice for the date, the input confirmation 1 And confirming the region and date while repeating (step 33). If the input is denied as a result of the confirmation, the correction content is asked as re-input 1 (step 34), and the process returns to the input confirmation 1 with the correction utterance. When the input confirmation 1 is affirmed, the corresponding weather forecast is output as the result output 1 (step 35).
“Interrupt process 1” in FIG. 3 instructs the user to interrupt the dialog at the position of the dialog scenario in each unit scenario process of “region acquisition 1” “date acquisition 1” “input confirmation 1” “re-input 1”. When an utterance is input, it is called as indicated by the broken line in FIG. In this case, the content that saves the current input information and informs that the dialogue is temporarily interrupted is output from the voice output device 9 (step 36).

In “interruption process 2” in FIG. 3, the dialog history generation device 11 is instructed to generate dialog history information at the time when the corresponding dialog scenario position is executed. That is, the input information in the dialog control memory 5 at this time, the dialog scenario position (dialog control program position), and the time information from the clock 12 are generated (step 37).
In FIG. 3, “Resume processing 1” is an utterance understanding result indicating that the previous dialogue is instructed by the user's utterance in the “region acquisition 1” stage (step 31) immediately after the start of the voice dialogue device. Is called (indicated by a chain line in FIG. 3). At this time, the dialogue control device 1 instructs the dialogue history acquisition device 4 to output the dialogue history acquisition information in the previous dialogue, and as a result, information in the dialogue history acquisition information output from the dialogue history acquisition device 4. Is taken into the dialogue control memory 5.

  Dialog history information is stored in the dialog history storage device 10 from the dialog history generation device 11 by the “interruption process 2”, and dialog history acquisition information is stored in the dialog history acquisition device 4 from the dialog history storage device 10 in the “resume process 1”. Is obtained. In the dialog history acquisition device 4, when an instruction to acquire the dialog history acquisition information is received from the dialog control device 1, the dialog history information from the dialog control device 1 is obtained from the dialog history information in the dialog history storage device 10 as shown in FIG. The dialogue history information corresponding to the instruction is searched and acquired (step 61). The search can be performed using time information by the clock 12 included in each dialogue history information as an index. In this case, since the dialog history information at the time of interruption is searched here, the latest information may be selected as the time information.

  Next, in the dialog history acquisition device 4, a request command is transmitted to the clock 12 for the time information in the acquired dialog history information to obtain time information indicated by the current time (step 62). Then, interruption elapsed time amount information, which is the amount of passage between the time information at the time of interruption in the dialogue history information and the time information at the current time obtained at the time of acquisition, is obtained (step 63). At this time, the interrupted elapsed time amount information not only obtains a mere elapsed amount value by subtracting a numerical value for each time unit such as how many days, how many hours, but also the difference in calendar date, month, year. calculate. For example, when the time information in the dialogue history information represents “15:00 pm on March 26, 2004”, the current time obtained is “1:05 pm on April 10, 2004”. If there is, information that indicates “year unit: no change, month unit: January change, day unit: 15 day change, time unit: 360 hours, minute unit: 21600 minutes” is generated. In other words, information on changes in the calendar such as how many times crossing the midnight that is the boundary of the date, how many times it is crossing the end of the month that is the boundary of the month, and how many times it crosses the end of the year that is the boundary of the year Granted at each time level.

  Further, in the dialog history acquisition device 4, when the input information itself in the acquired dialog history information includes time information indicating date and time (step 64), the time information in the input information and the current information acquired from the clock 12 are displayed. Compared with the time information of the time, it is determined with the progress flag true or false whether the time information in the input information is valid or invalid for the purpose of use of the voice interactive device (step 65). The determination in this case is performed by the method shown in FIG. 7, and when the time information in the input information is invalid, the true value of the progress flag is given to the input information. Further, when the time information of the input information is valid, a false value is assigned as the progress flag. If there is time information in the input information and the progress flag is true or false and there is no time information in the input information, the dialog history acquisition information from the dialog history acquisition device 4 is the progress flag and the elapsed time information of interruption. Are added and output (step 66).

  Here, a method for determining the progress flag will be described with reference to FIG. A flowchart shown in FIG. 7 is used to determine whether the time information in the input information is valid or invalid. In FIG. 7, the storage time information is time information in the input information and is indicated by U1, and the current time information is time information of the current time in the dialog information acquisition device 4 and is indicated by U2. In addition, the minimum time unit used in the conditional branch indicates the minimum level of the time unit required for the purpose of use and is displayed as Up, and the same time validity determination flag enables or disables the same time. This flag is indicated by Flg, and true is true and valid. These minimum use time unit and the same time validity determination flag are designated in advance in the dialogue scenario according to the purpose of use of the voice dialogue device.

  For example, in the dialogue scenario in FIG. 3, the date of “date acquisition” in the input information represents the day for the minimum unit of use. It is assumed that the time information in the “date” input information in the dialog history is “March 23, 2004” when the dialog is executed. At this time, when the information up to the date of the current time of the time information acquired by the dialog history acquisition device 4 from the clock 12 based on the processing in the “resume processing 1” is “March 24, 2004”, the weather forecast It is unnatural to use a past date for the purpose of inquiring. The progress flag for the input information “date” which has become before yesterday should be determined to be a true value (has passed). If the date of the current time when the dialogue is restored and the date of the stored input information are both “March 23, 2004”, it is natural to check the weather forecast for the day, so the progress flag is Should be a false value. That is, it should be determined that the date that is the minimum unit of time information handled by the voice interactive apparatus has not yet passed between the same dates.

In such a case, at the time of determination of the elapsed flag in FIG. 7, the above-described determination is performed by instructing execution with the “day” same time validity determination flag being true (valid) for each minimum use time unit. By providing such a mechanism, the determination as to whether the time information in the input information in the dialogue history acquisition device 4 is valid or invalid is not simply based on whether a certain date or time has passed, It is characterized in that the determination is made according to the purpose of use of the voice interactive device (dialog system).
Here, the flowchart in FIG. 7 will be described. In step 71, in the comparison between the storage time information U1 (March 23, 2004) and the current time information U2 (March 24, 2004), this information starts from the year that is the maximum usage time level, and the month, day Set the hour, minute level. Next, in step 72, it is determined whether the current time information U2 is larger than the storage time information U1 (time has passed). If there is no difference at the same time level, the process proceeds to step 73, and this time level is the minimum used level. Judge whether or not the time unit. When the year and month are the same in the current weather forecast, Steps 71, 72, and 73 are repeated, the time level of “day” is reached, Step 73 is satisfied, the process proceeds to Step 74, and the progress flag becomes true (invalid). If step 72 is not satisfied even at the time level of the day, it becomes the minimum usable time unit at step 73, so the routine proceeds to step 75, where it is determined whether or not the simultaneous validity determination flag is true. The flag becomes false and the storage time information becomes valid. That is, the weather forecast for the same day is valid. If the simultaneous validity determination flag is false at step 75, the process proceeds to step 77, where the stored time information and the current time information are all compared, and the result is determined as true or false.

  Here is another example of determining the progress flag. The purpose of this voice dialogue device is a reservation system for performance tickets such as movies and plays, and the time information in the input information of “reservation desired performance start time” stored in the dialogue history information is “March 2004” If the current time information obtained from the clock 12 is “18:10 PM on March 23, 2004” at “18:00 PM on the 23rd”, it is not possible to make a reservation for a performance that has already started. Therefore, in step 74, the progress flag should be set to a true value. That is, it is determined that it has already passed even if the dialog resumption process is performed just after 18:00. In this case, at the time of determination of the elapsed flag in FIG. 7, the above determination can be made by instructing execution as “time” in the minimum use time unit and the same time validity determination flag in step 75 as false.

As described above, through the processing in the dialogue history acquisition device 4 shown in FIGS. 6 and 7, the dialogue history acquisition device 4 obtains the dialogue history including the dialogue scenario position, input information, elapsed flag, and interruption elapsed time amount information. Information is output and passed to the dialog control device 1. At this time, these values are written in the dialogue control memory 5. However, the information related to the dialog scenario position is stored at a position different from the position where the currently executed dialog scenario position is stored for reference in subsequent processing.
The processing in the dialog history acquisition device 4 shown in FIG. 6 and FIG. 7 is related to the restart processing after the interruption processing at the time of generation in the dialog history generation device 11. The restart process will be described.

In “Resume process 1” in FIG. 4, the dialog history acquisition information is fetched from the dialog history acquisition device 4 into the dialog control memory 5 (step 41). In “Resume process 2”, the interruption is the time during which the dialog was interrupted. The audio output device 9 is instructed to reproduce a response expressing the interruption time interval based on the elapsed time amount information. This expression is selected based on the information on the elapsed time of interruption in the dialog control memory 5 in the dialog scenario or in the dialog control memory 5 in light of the relationship diagram in the first table of the language expression shown in FIG. (Step 42).
As an example of use in FIG. 8, when “2 days” has passed in “day unit” when the conversation is resumed after 2 days, the expression “Ototo” is used in Japanese because it corresponds to item number 5 in FIG. 8. In addition, when the conversation is resumed after one month and “January” has elapsed in “Month unit”, the expression “Last month” (No. 10 in FIG. 8) is used in Japanese. By using such an idiomatic expression, not only the purpose of notifying the interruption time can be achieved, but also a natural and simple response that is not mechanical to the user can be output. In this way, it is possible to construct a dialog system that performs a natural and simple response that expresses the time interval from dialog to return using conventional expressions when the dialog returns.

  Next, when "Region" and "Date" are included in the input information in the dialog history acquisition information, a voice output is output so that a response including an expression that informs the content as the content of the dialog in this input information is output. Instruct the device 1. However, even if the stored input information includes the input information for “date”, the “date” information is read out when the progress flag indicates a true value (invalid) for the input information. No response. To determine the response output, a second table of saved information is provided in the dialogue scenario or in the dialogue control memory 5 as shown in FIG. 9, and the saved information in this second table is interrupted by the dialogue control memory 5. The elapsed time information, input information, and elapsed flag are selected as conditions (step 42).

In “Resume processing 3” in FIG. 4, the information on the dialogue scenario position obtained from the dialogue history acquisition device 4 is referred to determine whether it is “region acquisition 1”. (Step 43).
In “Resume processing 4” in FIG. 4, the information on the dialogue scenario position obtained from the dialogue history acquisition device 4 is referred to determine whether it is “date acquisition 1”. (Step 44).
Further, in “Resume process 5” in FIG. 4, the information on the dialogue scenario position obtained from the dialogue history acquisition device 4 is referred to determine whether it is “re-input 1”. Jump to input 1 "(step 45).

Furthermore, in “Resume process 6” in FIG. 4, the information on the dialogue scenario position obtained from the dialogue history acquisition device 4 is referred to, and it is determined whether it is “input confirmation 1”, and “date” It is determined whether the progress flag for is a false value (valid), and if both are satisfied, the process jumps to “input confirmation 1” (step 46).
In “date acquisition 2” in FIG. 4, a response process is performed for the user to input again the “date” whose progress flag is true and invalid (step 47). First, the voice output device 9 is instructed to output content that prompts the user to re-enter the date, and the grammar and dictionary for recognizing the utterance related to the date are used in addition to a command such as interruption to the voice recognition device 6. Instruct to perform recognition and wait for input from the user. When voice input is performed and date information is output from the utterance understanding device 3, the information is recorded in the dialog control memory 5 as “date” input information.

If the progress flag for the input information “Date” is true during the “Resume Process 6” process, it is necessary to input “Date”. When jumping to “”, the process of “re-input 1” is performed due to the denial from the user. However, in the re-input 1 process, the speech recognition apparatus uses a grammar and a dictionary for allowing utterance when re-inputting the area in addition to the date. Therefore, there was a possibility of misrecognizing the date utterance as the utterance related to the regional input.
However, as described here, only re-input of the date is accepted through a series of branch processing from “resume process 6” to “date acquisition 2” using the information of the progress flag determined at the time of obtaining the conversation history information. By setting the grammar and dictionary and driving the speech recognition device 6, it is possible to recognize with high accuracy. That is, it is possible to construct a dialogue system with higher recognition accuracy by using only the information regarding the date and time invalidated in the determination of validity as the target of speech recognition. As a result, it is possible to construct a voice dialogue system with less user's cognitive burden.

In this case, since the user does not determine the validity of the “date” information but the system automatically detects it, the burden on the user can be reduced.
In “Region acquisition 1” in FIG. 4, the voice output device 9 is instructed to output the content for prompting the user to input the region, and the voice recognition device 6 is instructed regarding the region in addition to the interruption and restart instructions. Instructs the user to perform recognition using a grammar and dictionary for recognizing the utterance, and waits for input from the user. When voice input is performed and regional information is output from the utterance understanding device 3, it is recorded as “regional information” in the user input information in the dialogue control memory 5.

In “date acquisition 1” in FIG. 4, the voice output device 9 is instructed to output the content that prompts the user to input the next date, and the speech recognition device 6 is uttered about the date in addition to a command such as interruption. Instructs recognition using a grammar and dictionary for recognition and waits for input from the user. When voice input is performed and date information is output from the utterance understanding device 3, the information is recorded in the dialog control memory 5 as “date” input information.
In “input confirmation 1” in FIG. 4, the input information of “region” and “date” in the dialogue control memory 5 is output as a repetition, and it is urged to affirm or deny whether or not this information can be notified of the weather. Instructs the voice output device 9 to output the contents, and instructs the voice recognition device 6 to perform recognition using a grammar and dictionary for recognizing an utterance related to affirmation or denial in addition to a command such as interruption. Wait for input from the user. If the input is performed by voice and the output from the utterance understanding device 3 is input information indicating affirmation, “result output 1” is executed. If the input information indicates negative, “re-input 1” is executed.

In “re-input 1” in FIG. 4, in response to a negative input from the user in “input confirmation 1”, the voice output device 9 is instructed to output to prompt the user to input correction contents, and the voice recognition device 6 is instructed to perform recognition using a grammar and a dictionary for recognizing an utterance related to a region and a date in addition to a command such as interruption, and waits for input from a user.
In “result output 1” in FIG. 3, the weather forecast corresponding to “region” and “date” recorded in the dialog control memory 5 in response to a positive input from the user in “input confirmation 1”. The audio output device 9 is instructed to output the contents of.

With the control flowchart described above, it is possible to perform a dialogue such as the following dialogue example.
Dialogue example (In the following, U is a user and S is a system)
(Suppose that the following utterances 1 to 6 were made on March 22)
Utterance 1: S: “Please tell me the area you want to know” (Reply / Recognition by Area Acquisition 1 :)
Speak 2: U: “Eastern Kanagawa”
Speak 3: S: “When is the date you want to know about eastern Kanagawa?”
(Response recognition by date acquisition 1 :)
Speaking 4: U: “It ’s March 23”
Speaking 5: S: “Then, is it OK for the weather on March 23 in eastern Kanagawa Prefecture?” (Response / recognition by input confirmation 1 :)
Utterance 6: U: “I will suspend this dialogue”
Utterance 7: S: “Now save and suspend this conversation” (response by suspend process 1: after this, process by suspend process 2 is performed)
(The following utterances 8 to 14 are assumed to have been performed on March 24)
Utterance 8: S: “Please tell me the region you want to know” (Response / recognition by region acquisition 1 :)
Speaking 9: U: “We will return to the previous dialogue”
Speaking 10: S: “Otoi, you were checking the weather in eastern Kanagawa Prefecture. Please tell me the date you want to check again.” (Responding to Resume 2: After Resume 1: (Resume process 3 :, Resume process 4 :, Resume process 5 :, Resume process 6: Response of date acquisition 2: is performed after each determination process)
Utterance 11: U: “The day I want to investigate is March 25”
Speaking 12: S: “Are you sure the weather on March 25 in eastern Kanagawa?”
(Response / recognition by input confirmation 1 :)
Utterance 13: U: “Yes”,
Utterance 14: S: “The weather on March 25 in eastern Kanagawa is clear” (Processing by Result Output 1 :)
(End of dialogue example)
As described above, the embodiment has been described with the input information having the time information at the time of interruption in mind, but the present invention can determine whether the input information at an arbitrary time before the interruption is valid or invalid at the time of resumption. In this case, based on the input information temporarily stored together with the dialog scenario position in the dialog control memory 5 for each unit scenario process, it is generated as dialog history information along with interruption in the dialog history generating device 11 and stored in the dialog history storage device 10. When restarting, the scenario position is selected from the dialogue history information, the input information is taken into the dialogue history acquisition device, and is output to the dialogue control device 1. Then, the time information in the corresponding input information is compared with the time information at the time of restart to determine whether the progress flag is valid or invalid. Also, the interruption elapsed time amount information at the time of interruption and at the time of resumption is calculated, and language expression or storage information is selected from the tables of FIGS. When performing the valid or invalid process for the input information at an arbitrary time, in step 37 of FIG. 2, in the interruption process 2, the dialog history generating device 11 generates dialog history information including the previous arbitrary time. In step 61 of FIG. 6, the dialog history information at an arbitrary point in the dialog history storage device is retrieved and acquired.

Also, in the explanation of the embodiments so far, as the concrete explanation, the weather reservation or the ticket reservation for the performance has been described. For example, as the ticket reservation has more options such as the type of seat, the number of seats, etc. Considering that the number of times increases, the more the number of choices becomes, the more effective and invalid the determination of the time information in the input information becomes.
In this embodiment, the speech synthesizer is used for output, but a method such as printing text on the screen may be used.

  In the present embodiment, the dialog history information is stored after receiving a request for interruption from the user, but may be stored at any time during the dialog.

It is a block block diagram which shows embodiment of this invention. It is a flowchart which shows the outline of the dialogue scenario which concerns on this invention. It is a flowchart of a dialogue scenario. FIG. 4 is a flowchart of a restart process following FIG. 3. It is a processing flowchart of a dialogue history generation device. It is a processing flowchart of a dialogue history acquisition device. It is a process flowchart of the determination method of progress flag. It is a related figure of the language expression which shows the interruption elapsed time amount. It is a related figure of the preservation | save information in a restart process.

Claims (11)

In a voice interaction method that processes voice input and outputs response voice,
When a dialog interruption is input, a dialog history is generated that includes the input information input up to the point of interruption of the dialog, the control program position of the dialog scenario being executed at that time, and the time information at that time. Processing,
When a dialog resume is input, the dialog history obtained in the dialog history generation process is extracted, and the input is made by referring to the time information in the input information in the extracted dialog history against the time information at the time of restarting the dialog. A dialogue history acquisition process for determining whether the time information in the information is still valid or invalid. The voice interaction method according to claim 1,
Further, the dialog history acquisition process to which an operation for obtaining an elapsed time from interruption to resumption from time information at the time of interruption and time information at the time of resumption is added,
A response using a natural language expression corresponding to the elapsed time from the interruption point to the resumption point obtained in this dialog history acquisition process is output, and resumption corresponding to the time information in the input information in the case of the above invalid determination And a dialogue control process for outputting a response based on the time information at the time. In a voice interaction method that processes voice input and outputs response voice,
When a dialog interruption is input, a dialog that generates a dialog history consisting of the input information input up to any point in the dialog, the control program position of the dialog scenario currently being executed, and the time information at that time History generation processing,
When a dialog resume is input, the dialog history obtained in the dialog history generation process is extracted, and the input is made by referring to the time information in the input information in the extracted dialog history against the time information at the time of restarting the dialog. A dialogue history acquisition process for determining whether the time information in the information is still valid or invalid. The voice interaction method according to claim 3.
Further, the dialog history acquisition process to which an operation for obtaining an elapsed time from an arbitrary time point to the restart time point is added from the time information at the arbitrary time point and the time information at the restart time point;
Outputs a response using natural language expression in accordance with the elapsed time from the above arbitrary time point to the restart point obtained in this dialog history acquisition process, and corresponds to the time information in the input information in the case of the above invalid determination And a dialogue control process for outputting a response based on the time information at the time of resumption.   5. The voice dialogue method according to claim 1, wherein whether the time information is still valid or invalid is decided according to a purpose of use of the dialogue history.   6. The spoken dialogue method according to claim 1, wherein the natural language expression is extracted from a table of idiomatic expressions according to elapsed time.   7. The spoken dialogue method according to claim 1, wherein in the natural language expression, stored information corresponding to a determination result of time information is extracted from a table. In a voice interactive device that processes voice input and outputs response voice,
A dialog history generating device for generating a dialog history including input information input up to the point of interruption of the dialog, a control program position of a dialog scenario being executed at that time, and time information at that time;
Take out the dialogue history obtained by this dialogue history generation device, and check whether the time information in the input information is still valid by checking the time information in the input information in the extracted dialogue history against the time information at the time of restarting the dialogue A voice dialog device, comprising: a dialog history acquisition device that determines whether it is invalid. The voice interaction apparatus according to claim 8, wherein
Furthermore, the dialog history acquisition device to which an operation for obtaining an elapsed time from interruption to resumption from the time information of the time information at the time of interruption and the time information at the time of resumption is added,
A response using a natural language expression corresponding to the elapsed time from the interruption point to the restart point obtained by the dialog history acquisition device is output, and the restart corresponding to the time information in the input information in the case of the above invalid determination And a dialogue control device for outputting a response based on the time information at the time. In a voice interactive device that processes voice input and outputs response voice,
A dialog history generation device for generating a dialog history including input information input up to an arbitrary point in time during the dialog, a control program position of the dialog scenario being executed at that time, and time information at that time;
Take out the dialogue history obtained by this dialogue history generation device, and check whether the time information in the input information is still valid by checking the time information in the input information in the extracted dialogue history against the time information at the time of restarting the dialogue A voice dialog device, comprising: a dialog history acquisition device that determines whether it is invalid. The voice interaction apparatus according to claim 10, wherein
Further, the dialog history acquisition device to which an operation for obtaining an elapsed time from an arbitrary time point to the restart time point is added from the time information at the arbitrary time point and the time information at the restart time point;
Outputs a response using natural language expression corresponding to the elapsed time from the arbitrary time point to the restart time obtained by the dialog history acquisition device, and corresponds to the time information in the input information in the case of the invalid determination And a dialogue control device for outputting a response based on time information at the time of resumption.

Images