JP2011175304A - Voice interactive device and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice interactive device and method capable of responding at suitable timing. <P>SOLUTION: At first, a word string is extracted from input voice (S30). A speaking speed of the input voice is calculated (S40). Then, by comparing an extracted word string with an appearance probability list for storing the word string which is predicted to follow a present input word string (hereunder, a following predicted word string) and an appearance probability corresponding to the following predicted word string, the following predicted word string of the highest appearance probability in the following predicted word strings which are predicted to follow the extracted word string is extracted (S50). By using the calculated speaking speed, time which requires for the following predicted word string to be input (hereunder, following input time) is calculated (S60). Later, the appearance probability of the extracted following predicted word string is regarded as a confidence factor, and it is applied to the extracted following predicted word string (S70). When the applied confidence factor is a response determination value or more, output timing prediction is fixed at the following input timing (S80). <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a voice interactive apparatus and method for performing a response corresponding to a voice uttered by a user.

  2. Description of the Related Art Conventionally, in a voice dialogue apparatus, there are known ones that reduce anxiety of voice input by notifying humans that voice is being input, and those that facilitate input when there is a gap during dialogue . As one of them, by making humanized artificial agents interact and whisper during conversations, by clarifying the intentions such as “listening to the story” or “continuing the story” to humans In addition, a technique for smoothly promoting a dialogue between a human and an artificial agent has also been proposed (see, for example, Patent Document 1, Patent Document 2, Patent Document 3, Patent Document 4, Patent Document 5, and Patent Document 6). .)

  For example, the speech dialogue system disclosed in Patent Document 1 determines response timing and semantic processing timing based on speech recognition results, pitch time-series information, line-of-sight time-series information, dependency information, and the like. When it is determined that it is a response timing and not a semantic processing timing, a keyword that is in conflict or uttered is issued.

  The conflict robot disclosed in Patent Document 2 is configured to read corresponding conflict registration data (such as “That's right”) when the recognition result matches the voice registration table.

  Patent Document 3, Patent Document 4, Patent Document 5, and Patent Document 6 disclose a technique for outputting a conflict when a predetermined pose or keyword is detected.

JP-A-2005-196134 JP 2003-88686 A JP-A-7-191687 JP 7-219961 A JP-A-8-211986 JP 2004-86001 A

  However, in the techniques described in Patent Documents 1 to 6 described above, the determination is performed using the feature amount immediately before the timing of performing the reconciliation or whispering, such as language information such as a pose or a keyword. Then, since it takes time to perform a process of making a check and a call after making the determination, it is difficult to add a check and a call at an appropriate timing.

  And, in the case of such inadequate timing, it may give a bad impression that the talk is interrupted (refer to the document "Realization of language recognition and prosodic information for conflict recognition / generation function in spoken dialogue system") , Miyake et al., 2005 Acoustical Society of Japan Autumn Meeting, 1-P-20, pp.191-192) "Evaluation", Takeuchi et al., IPSJ 64th National Convention, Vol.2, pp.101-102), there is a possibility of disrupting the rhythm of dialogue.

  Moreover, in human-to-human dialogue, there are many conflicts that overlap utterances (for example, the literature “Corpus-based temporal analysis and discussion of conflicts”, Haruka Nakazato, Artificial Intelligence Society study material, SIG-SLUD-A003-7 ( (See 3/2)). However, with the method of determining the timing for performing a match or a call by the expression of the pause or the end of the utterance, it is not possible to realize a conflict that overlaps the utterance.

  The present invention has been made in view of these problems, and an object thereof is to provide a voice interactive apparatus and method capable of responding at an appropriate timing.

  In the spoken dialogue apparatus according to any one of claims 1 to 5, which is made to achieve the above object, the input means inputs the voice uttered by the user, and the prediction means based on the voice input to the input means. Thus, timing prediction information indicating a prediction result of response timing for performing a response corresponding to the input voice is acquired. Further, the timing determination unit determines whether or not the response timing has been reached based on the timing prediction information acquired by the prediction unit. The response means makes a response when it is determined by the timing determination means that the response timing has come.

  According to the voice interaction apparatus configured as described above, since the response timing can be predicted in advance by the prediction means, it is possible to secure a processing time for causing the response means to start a response. That is, it is possible to suppress the occurrence of a situation in which the response timing is reached before the response means responds, and an excellent effect is achieved in that a response can be made at an appropriate timing.

  In the spoken dialogue apparatus according to any one of claims 1 to 5, the predicting means can respond by interrupting at least one character before the end of the utterance by the user or during the utterance by the user. Timing prediction information is acquired at least one character before.

  According to the spoken dialogue apparatus configured in this manner, it is possible to secure a time during which at least one character is uttered from when the response means predicts the response timing to when the response timing is reached.

  Further, in the spoken dialogue apparatus according to any one of claims 1 to 5, the prediction means, specifically, as described in claim 6, a prediction model in which response timing is determined in advance; Timing prediction information may be acquired by comparing the feature amount of the prediction model with the feature amount of the input speech for the speech input by the input means.

  Further, in the spoken dialogue apparatus according to claim 6, as described in claim 7, the feature amount includes a syntactic feature amount indicating a syntactic feature and a prosodic feature regarding the utterance by the user. It is preferable that at least one of the prosodic feature values to be shown.

  The syntactic feature value and the prosodic feature value can be obtained sequentially from the input voice. According to the spoken dialogue apparatus configured in this way, The response timing can always be predicted at the time of input.

  Further, in the spoken dialogue apparatus according to claim 7, as described in claim 8, the syntactic feature amount is at least one of a preset keyword, word string, morpheme string, part of speech string, and phoneme string. The prosodic feature value is information including at least one of utterance length, basic frequency time-series information, pitch time-series information, power time-series information, and speech speed time-series information. It may be.

  In the spoken dialogue apparatus according to claim 4, the timing prediction information includes at least the number of words, the number of morphemes, the number of parts of speech, and the number of phonemes that continue from when the prediction unit acquires the timing prediction information until the response timing is reached. In the spoken dialogue apparatus according to claim 5, at least one of a word string, a morpheme string, a part-of-speech string, and a phoneme string that continues from when the prediction unit acquires timing prediction information until the response timing is reached. It is.

  The spoken dialogue apparatus according to claim 1, further comprising: a speech speed calculation unit that calculates a current speech speed based on the voice input to the input unit, and the prediction unit acquires the timing prediction information. Obtain at least one of the number of words, the number of morphemes, the number of parts of speech, the number of phonemes, the word sequence, the morpheme sequence, the part of speech sequence, and the phoneme sequence that follow until the response timing is reached, and these are calculated by the speech speed calculation means Based on the talk speed, the response timing arrival time until the response timing is calculated is calculated, and the response timing arrival time is used as timing prediction information.

According to the voice interaction apparatus configured as described above, the response timing can be adjusted according to the speech speed.
In the spoken dialogue apparatus according to claim 2, the predicting unit predicts a time that elapses after the predicting unit acquires the timing prediction information until the response timing is reached, and uses the predicted time as timing prediction information. In the spoken dialogue apparatus according to claim 3, the number of frames that continue until the response timing comes after the prediction means acquires the timing prediction information is used, and the predicted number of frames is used as the timing prediction information.

  Further, in the voice interactive apparatus according to any one of claims 1 to 8, the response timing is a timing at which a response is made overlapping with a user's utterance as described in claim 9. Good.

According to the voice dialogue apparatus configured as described above, it is possible to bring a person close to a state in which dialogue is being performed, and the dialogue can be advanced more smoothly.
Further, in the voice interaction method according to any one of claims 10 to 14, in the input step, first, the voice uttered by the user is input, and in the subsequent prediction step, the input voice is based on the voice input in the input step. Timing prediction information indicating a prediction result of response timing for performing a response corresponding to is acquired. Further, in the timing determination step, it is determined whether or not a response timing has been reached based on the timing prediction information acquired in the prediction step. In the response step, when it is determined that the response timing is reached in the timing determination step, a response is made.

  The voice interaction method is a method executed by the voice interaction device according to claims 1 to 5, and by executing the method, the voice interaction device according to claims 1 to 5. Similar effects can be obtained.

  Furthermore, in the spoken dialogue method according to any one of claims 10 to 14, the prediction step can respond by interrupting at least one character before the end of the utterance by the user or during the utterance by the user. Timing prediction information is acquired at least one character before.

  The voice interaction method is a method executed by the voice interaction device according to claims 1 to 5, and by executing the method, the voice interaction device according to claims 1 to 5. Similar effects can be obtained.

  Further, in the voice interaction method according to any one of claims 10 to 14, the prediction step, specifically, as described in claim 15, a prediction model in which response timing is determined in advance, Timing prediction information may be acquired by comparing the feature amount of the prediction model with the feature amount of the input speech for the speech input in the input step.

  Furthermore, in the spoken dialogue method according to claim 15, as described in claim 16, the feature amount includes a syntactic feature amount indicating a syntactic feature and a prosodic feature regarding the utterance by the user. It is preferable that at least one of the prosodic feature values to be shown.

  This voice interaction method is a method executed by the voice interaction device according to claim 7, and the same effect as that of the voice interaction device according to claim 7 can be obtained by executing the method. .

  Further, in the spoken dialogue method according to claim 16, as described in claim 17, the syntactic feature quantity is at least one of a preset keyword, word string, morpheme string, part of speech string, and phoneme string. The prosodic feature value is information including at least one of utterance length, basic frequency time-series information, pitch time-series information, power time-series information, and speech speed time-series information. It may be.

  In the voice interaction method according to claim 13, the timing prediction information includes at least one of a word number, a morpheme number, a part-of-speech number, and a phoneme number that continues until the response timing comes after the timing prediction information is acquired by the prediction step. 15. The spoken dialogue method according to claim 14, wherein at least one of a word string, a morpheme string, a part-of-speech string, and a phoneme string following the timing prediction information obtained by the prediction step until the response timing is reached. It is.

  The voice interaction method according to claim 10 further comprises a speech speed calculation step for calculating a current speech speed based on the voice input in the input step, and the prediction step acquires timing prediction information by the prediction means step. And at least one of the number of words, the number of morphemes, the number of parts of speech, the number of phonemes, the word sequence, the morpheme sequence, the part of speech sequence, and the phoneme sequence that are acquired until the response timing is reached, and these are calculated by the speech speed calculation step Based on the spoken speed, the response timing arrival time until the response timing is calculated is calculated, and this response timing arrival time is used as timing prediction information.

  This voice interaction method is a method executed by the voice interaction device according to claim 1, and the same effect as the voice interaction device according to claim 1 can be obtained by executing the method. .

  In the spoken dialogue method according to claim 11, the prediction step predicts a time that elapses from when the timing prediction information is acquired by the prediction step until the response timing is reached, and the predicted time is used as the timing prediction information. In the voice interaction method according to the twelfth aspect, after predicting the timing prediction information in the prediction step, the number of frames that continue until the response timing is predicted, and the predicted number of frames is used as the timing prediction information.

  Further, in the voice interaction method according to any one of claims 10 to 17, the response timing is a timing at which a response is made overlapping with a user's utterance as described in claim 18. Good.

  This voice interaction method is a method executed by the voice interaction device according to claim 9, and the same effect as that of the voice interaction device according to claim 9 can be obtained by executing the method. .

1 is a block diagram showing a configuration of a voice interaction device 1. FIG. It is a functional block diagram which shows the outline | summary of the process which the control part 4 performs. It is a flowchart which shows a voice interaction process. It is a figure explaining the method of output timing prediction. FIG. 6 is a diagram showing the contents of a volume parameter list 21. It is a figure explaining the conventional output timing determination method and the output timing determination method of the voice interactive apparatus.

Embodiments of the present invention will be described below with reference to the drawings.
FIG. 1 is a block diagram showing the configuration of the voice interaction apparatus 1 of this embodiment.
As shown in FIG. 1, the voice interaction apparatus 1 includes a voice input unit 2 that inputs voice spoken by a user, a voice output unit 3 that outputs voice, and various processes according to input from the voice input unit 2. And a control unit 4 that controls the audio output unit 3.

Among these, the voice input unit 2 outputs an electric signal (voice signal) based on the inputted voice to the control unit 4 when the user inputs (speaks) voice.
The control unit 4 is mainly composed of a well-known microcomputer comprising a CPU, ROM, RAM, I / O and a bus line connecting these components, and is based on a program stored in the ROM and RAM. Perform various processes.

Here, FIG. 2 is a functional block diagram showing an outline of processing executed by the control unit 4.
As illustrated in FIG. 2, the control unit 4 includes a speech recognition unit 11 that performs recognition processing of speech input by the speech input unit 2, and a response (e.g. A response generation unit 12 that generates a response to a voice input such as “How is the weather tomorrow?” And a response that is generated by the response generation unit 12 as a voice output unit 3, an output timing prediction unit 14 that predicts a timing for outputting a response such as a conflict based on a recognition result by the speech recognition unit 11, and an output timing predicted by the output timing prediction unit 14. The output timing control unit 15 that causes the audio output unit 3 to output to the unit 13 and the response by the audio output unit 3 are changed to the output unit 13 based on the prediction result in the output timing prediction unit 14 It includes a response change unit 16 to the prediction model used to predict the output timing prediction unit 14 (e.g., models created in advance using the learning data of the corpus or the like) and a model storage unit 17 for storing.

  Among these, the model storage unit 17 includes, as the prediction model, a word string predicted to follow the current input word string (hereinafter also referred to as a subsequent prediction word string) and an appearance probability corresponding to the subsequent prediction word string. The appearance probability list 17a to be stored is stored.

In the voice interaction apparatus 1 configured as described above, the control unit 4 executes a voice interaction process for performing a conversation based on the input voice.
Here, the procedure of the voice dialogue process executed by the control unit 4 of the voice dialogue apparatus 1 will be described with reference to FIGS. FIG. 3 is a flowchart showing a voice dialogue process, and FIG. 4 is a diagram for explaining a method for predicting output timing.

This voice interaction process is a process repeatedly executed while the control unit 4 is activated (powered on).
When the voice dialogue processing is executed, the control unit 4 first determines whether or not a voice is input to the voice input unit 2 in S10. If no voice is input here (S10), the voice dialogue process is temporarily terminated. On the other hand, when voice is input (S10), voice recognition is performed on the voice input to the voice input unit 2 in S20.

  Thereafter, in S30, based on the voice recognition result in S20, a word string is extracted from the voice input to the voice input unit 2. Further, in S40, the speech speed of the voice input to the voice input unit 2 is calculated based on the voice recognition result in S20.

  In S50, the word string extracted in S30 is compared with the appearance probability list 17a, and the word string extracted in S30 is predicted to follow the word string extracted in S30 (subsequent predicted word string). Extract high succession prediction word string. Further, in S60, using the speech speed calculated in S40, the time required to input the subsequent predicted word string extracted in S50 (hereinafter also referred to as subsequent input time) is calculated. After that, in S70, the appearance probability of the subsequent prediction word string extracted in S50 is assigned as the certainty to the subsequent prediction word string extracted in S50.

  Thereafter, in S80, it is determined whether or not the certainty given to the subsequent predicted word string extracted in S50 is equal to or higher than a preset response determination value (for example, “0.1” in the present embodiment). . That is, it is determined whether the output timing prediction can be confirmed. Here, when the certainty factor is less than the response determination value, it is determined that the output timing prediction cannot be determined (S80), the process proceeds to S10, and the above-described processing is repeated. On the other hand, if the certainty factor is greater than or equal to the response determination value, it is determined that the output timing prediction can be confirmed (S80), and the process proceeds to S90.

  Here, a specific example of the processing of S50 to S80 will be described with reference to FIG. First, as shown in FIG. 4, when the word string “very” is input, it is “happy” (appearance probability is 0.015) from the appearance probability list 17 a as a subsequent predicted word string “very”. ), “It was fun” (appearance probability is 0.013), “I want something” (appearance probability is 0.002), etc., and the one with the highest appearance probability, for example, “I ’m happy” A subsequent predicted word string (appearance probability is 0.015) is extracted (S50). Then, a certainty factor of “0.015” equal to the appearance probability is given to the subsequent predicted word string “I'm happy” (S70). However, at this time, since the given certainty is less than the response determination value (0.1), the output timing prediction is not finalized (S80).

  After that, when the word string “interesting” is input, “ta <pause>” (appearance probability is 0.235), “ta” as the subsequent predicted word string “very interesting” from the appearance probability list 17a. ”(Appearance probability is 0.186),“ I think it is ”(appearance probability is 0.008), etc., among them, the one with the highest appearance probability, such as“ ta <pause> ”(appearance A subsequent predicted word string having a probability of 0.235) is extracted (S50). Then, a certainty factor “0.235” equal to the appearance probability is given to the subsequent predicted word string “ta <pause>” (S70). At this time, since the given certainty is greater than or equal to the response determination value (0.1), the output timing prediction is confirmed (S80).

  Returning to FIG. 3, when the process proceeds to S <b> 90, a process of changing the response method according to the certainty given in S <b> 70 is performed. Specifically, the volume parameter is set so that the volume of the response is increased in proportion to the certainty given in S70.

  In S100, it is determined whether or not the subsequent input time calculated in S60 has elapsed since the output timing prediction was determined in S80. That is, it is determined whether or not the output timing has come. If the subsequent input time has not elapsed (S100), the process of S100 is repeated. On the other hand, when the subsequent input time has elapsed, it is determined that the output timing has come (S100), and in S110, the response generated by the response generator 12 is set to a volume corresponding to the volume parameter set in S90. The voice output unit 3 is made to output, and the voice dialogue processing is once ended.

  According to the spoken dialogue apparatus 1 configured as described above, since the output timing can be predicted in advance by the processing of S50 to S80, the processing time for starting output of the response by the processing of S110 (hereinafter referred to as response processing). Time). That is, it is possible to suppress the occurrence of a situation in which the output timing is reached before starting the output of the response, and an excellent effect is obtained that the response can be made at an appropriate timing.

  Also, the output timing can be the timing at which a response is made overlapping with the user's utterance, so that it can be brought closer to the state in which humans are engaged in dialogue, and the dialogue proceeds more smoothly be able to.

  Specifically, conventionally, as shown in FIG. 6A, for example, in the utterance “It was very interesting <pause>”, at the time HT1 when the final particle “yo” was uttered or at the time HT2 of the pose, It is determined whether or not it is the end of the sentence, and a response is output at the time point OT1 and OT2 when the response processing time SJ1 has elapsed from this determination. For this reason, it is difficult to output a response immediately after the end of the utterance or to output a response overlapping the end of the sentence.

  On the other hand, as shown in FIG. 6B, the voice interactive apparatus 1 predicts the time point when the utterance ends, for example, at the time point HT3 of “very interesting” in FIG. 6B, before the utterance ends. For this reason, at the time point OT3 when the response processing time SJ1 has elapsed from this prediction, the utterance is not yet finished. Thereby, a response can be output immediately after the utterance is completed, or a response can be output by overlapping the end of the sentence.

  In the process of S50, the word string extracted in S30 is compared with the appearance probability list 17a, and the subsequent predicted word string is extracted at least one character before the end of the utterance by the user. For this reason, it is possible to secure a response processing time for a time during which at least one character is uttered.

  Further, since the word string extracted in the process of S30 can be obtained sequentially from the input voice, it is always output at the time of voice input by sequentially comparing with the appearance probability list 17a in the process of S50. Timing can be predicted.

  In the process of S60, the time (subsequent input time) required for inputting the subsequent predicted word string extracted in S50 is calculated using the speech speed calculated in S40. For this reason, output timing can be adjusted according to speech speed.

  Moreover, in the process of S70, when the appearance probability of the subsequent prediction word string extracted in S50 is given as the certainty to the subsequent prediction word string extracted in S50, and the certainty is equal to or higher than the response determination value in the process of S80. To confirm the output timing prediction. For this reason, since a reliable subsequent prediction word sequence can be extracted based on the certainty factor, a response can be performed at a more appropriate timing.

  Further, in the process of S90, a process of changing the response method according to the certainty factor is performed. Specifically, the volume parameter is set so as to increase the response volume in proportion to the certainty factor. For this reason, when the certainty factor is low, the effect of acting on the user can be reduced, and the rhythm of the dialogue can be prevented from being lost.

  In the embodiment described above, the voice input unit 2 is the input means and input step in the present invention, the processes in S50 to S80 are the predictor and prediction step in the present invention, and the process in S100 is the timing determination means and timing determination step in the present invention. , S110 is the response means and response step in the present invention, the subsequent prediction word string is the timing prediction information in the present invention, and the appearance probability list 17a is the prediction model in the present invention.

As mentioned above, although one Embodiment of this invention was described, this invention is not limited to the said embodiment, As long as it belongs to the technical scope of this invention, a various form can be taken.
For example, in the above-described embodiment, the voice interaction device 1 has shown a response such as a conflict by outputting a voice from the voice output unit 3, but is not limited to this, and a visual response is made. It may be. For example, an LED may be provided to respond by turning on the LED, or a display may be provided to display a whispering action, or a robot may be whispered.

  In the above embodiment, the word string is extracted in S30, and the output timing is predicted using the extracted information. However, the present invention is not limited to this, and the syntactic feature quantity different from the word string is used. For example, at least one of a predetermined keyword, morpheme string, part-of-speech string, and phoneme string may be extracted, utterance length, basic frequency time-series information, pitch time-series information, power At least one of prosodic feature quantities such as time-series information and speech speed time-series information may be extracted.

  In S50 of the above embodiment, the output timing is predicted by comparing the word string and the appearance probability list 17a. However, the present invention is not limited to this, and the distance from the model created from the time series data May be predicted by a method of measuring (for example, template matching) or another N-gram model (for example, word N-gram, part-of-speech N-gram, phoneme N-gram).

  In S50 of the above embodiment, the following word string is predicted. However, the present invention is not limited to this, and at least one of the following morpheme string, part-of-speech string, and phoneme string may be predicted. In addition, at least one of the following number of words, number of morphemes, number of parts of speech, and number of phonemes may be predicted.

  Alternatively, the time until the output timing is reached in S50 may be directly predicted, or the number of frames until the output timing is reached may be predicted. In this case, the process (S60) for calculating the subsequent input time is not necessary.

  In S60 of the above embodiment, the time required to input the subsequent word string (following input time) is calculated using the speech speed. However, the present invention is not limited to this. For example, the number of words is calculated. When predicted, the time taken to input the predicted number of words may be calculated from the speech speed. Further, the subsequent input time may be calculated from the predicted number of words, number of phonemes, word string, and the like, and a predetermined time per word and time taken to input the word string.

  Moreover, in S70 of the said embodiment, although the appearance probability of the subsequent prediction word sequence extracted by S50 is made into the certainty degree, it replaces with this, or with this, the matching rate with a model, the certainty degree of the recognition result by the speech recognition part 11 In addition, the certainty factor in S70 may be calculated using at least one piece of information of a time constant set so that the new prediction result has higher prediction reliability than the old prediction result.

  In S80 of the above embodiment, the output timing prediction is determined based on whether or not the certainty level is greater than or equal to the response determination value. However, the present invention is not limited to this. It may be performed by comparing the degree of certainty added to this candidate from a plurality of candidates whose output timing is predicted within the number, etc., or the magnitude comparison between the response determination value and the certainty degree You may make it carry out by the combination of these.

  In S90 of the above embodiment, the response volume is increased in proportion to the certainty factor. However, the present invention is not limited to this, and the certainty factor and the volume parameter corresponding to the subsequent predicted word string are stored. A volume parameter list 21 (see FIG. 5) may be provided in advance, and the volume of the response may be set by referring to the volume parameter list 21.

  In S90 of the above embodiment, the response method is changed according to the certainty factor. However, the method is not limited to this, and the response is made according to the predicted word string, morpheme string, part of speech string, phoneme string, and the like. You may make it change the way of.

  In addition, the response in the above embodiment is by voice, but the form of response may be whispering, blinking, or the like, and a typical function of the companion is “indicating understanding of utterance content” It may be a message or action having functions such as “indicating what is being heard”, “clarification of turn taking”, “indicating feelings and consent / denial”, and “promoting utterance”. For example, turn on the LED, change the brightness of the display, change the tilt of the object, change the speed of movement, change the number of operations, change the color and brightness, change the voice of the response voice It is conceivable to change the response message or change the CG animation.

  DESCRIPTION OF SYMBOLS 1 ... Voice interactive apparatus, 2 ... Voice input part, 3 ... Voice output part, 4 ... Control part, 11 ... Voice recognition part, 12 ... Response generation part, 13 ... Output part, 14 ... Output timing prediction part, 15 ... Output Timing control unit, 16 ... response changing unit, 17 ... model storage unit, 17a ... appearance probability list, 21 ... volume parameter list

Claims (18)

An input means for inputting voice uttered by the user;
Prediction means for acquiring timing prediction information indicating a prediction result of response timing for performing a response corresponding to the input voice based on the voice input to the input means;
Timing determination means for determining whether or not the response timing has been reached based on the timing prediction information acquired by the prediction means;
Response means for performing the response when it is determined by the timing determination means that the response timing has been reached, and
The prediction means includes
The timing prediction information is acquired at least one character before the end of the utterance by the user, or at least one character before the user can interrupt and perform the response.
A speech speed calculating means for calculating a current speech speed based on the voice input to the input means;
The prediction means includes
Acquire at least one of the number of words, the number of morphemes, the number of parts of speech, the number of phonemes, the word string, the morpheme string, the part of speech string, and the phoneme string after the prediction means acquires the timing prediction information and before the response timing is reached. Then, based on these and the speech speed calculated by the speech speed calculating means, a response timing arrival time until the response timing is reached is calculated, and the response timing arrival time is used as the timing prediction information. A featured voice dialogue device. An input means for inputting voice uttered by the user;
Prediction means for acquiring timing prediction information indicating a prediction result of response timing for performing a response corresponding to the input voice based on the voice input to the input means;
Timing determination means for determining whether or not the response timing has been reached based on the timing prediction information acquired by the prediction means;
Response means for performing the response when it is determined by the timing determination means that the response timing has been reached, and
The prediction means includes
The timing prediction information is acquired at least one character before the end of the utterance by the user, or at least one character before the user can interrupt and perform the response.
A spoken dialogue apparatus characterized by predicting a time that elapses from when the prediction means acquires the timing prediction information until the response timing is reached, and using the predicted time as the timing prediction information. An input means for inputting voice uttered by the user;
Prediction means for acquiring timing prediction information indicating a prediction result of response timing for performing a response corresponding to the input voice based on the voice input to the input means;
Timing determination means for determining whether or not the response timing has been reached based on the timing prediction information acquired by the prediction means;
Response means for performing the response when it is determined by the timing determination means that the response timing has been reached, and
The prediction means includes
The timing prediction information is acquired at least one character before the end of the utterance by the user, or at least one character before the user can interrupt and perform the response.
A spoken dialogue apparatus characterized by predicting the number of frames that continue from when the prediction means acquires the timing prediction information until the response timing is reached, and using the predicted number of frames as the timing prediction information. An input means for inputting voice uttered by the user;
Prediction means for acquiring timing prediction information indicating a prediction result of response timing for performing a response corresponding to the input voice based on the voice input to the input means;
Timing determination means for determining whether or not the response timing has been reached based on the timing prediction information acquired by the prediction means;
Response means for performing the response when it is determined by the timing determination means that the response timing has been reached, and
The prediction means includes
The timing prediction information is acquired at least one character before the end of the utterance by the user, or at least one character before the user can interrupt and perform the response.
The timing prediction information is
The spoken dialogue apparatus according to claim 1, wherein the prediction means is at least one of the number of words, the number of morphemes, the number of parts of speech, and the number of phonemes that follow from the acquisition of the timing prediction information to the response timing. An input means for inputting voice uttered by the user;
Prediction means for acquiring timing prediction information indicating a prediction result of response timing for performing a response corresponding to the input voice based on the voice input to the input means;
Timing determination means for determining whether or not the response timing has been reached based on the timing prediction information acquired by the prediction means;
Response means for performing the response when it is determined by the timing determination means that the response timing has been reached, and
The prediction means includes
The timing prediction information is acquired at least one character before the end of the utterance by the user, or at least one character before the user can interrupt and perform the response.
The timing prediction information is
A spoken dialogue apparatus, comprising: at least one of a word string, a morpheme string, a part-of-speech string, and a phoneme string that continues from when the prediction unit acquires the timing prediction information until the response timing is reached. The prediction means includes
The timing prediction information is obtained by comparing a feature amount of the prediction model with a feature amount of the input speech for the prediction model in which the response timing is determined in advance and the speech input by the input unit. The voice dialogue apparatus according to claim 1, wherein the voice dialogue apparatus is obtained. The feature amount is
The spoken dialogue apparatus according to claim 6, which is at least one of a syntactic feature indicating a syntactic feature and a prosodic feature indicating a prosodic feature regarding an utterance by a user. The syntactic feature is
Information including at least one of a preset keyword, word string, morpheme string, part of speech string, phoneme string,
The prosodic feature amount is:
The voice according to claim 7, wherein the audio includes at least one of utterance length, basic frequency time-series information, pitch time-series information, power time-series information, and speech speed time-series information. Interactive device. The response timing is
The voice interactive apparatus according to any one of claims 1 to 8, wherein the response is made at the timing when the response is made in duplicate with the user's utterance. An input step for inputting voice uttered by the user;
A prediction step of acquiring timing prediction information indicating a prediction result of a response timing for performing a response corresponding to the input voice based on the voice input in the input step;
A timing determination step for determining whether or not the response timing has been reached based on the timing prediction information acquired in the prediction step;
When it is determined that the response timing is reached by the timing determination step, the response step performs the response, and
The prediction step includes
The timing prediction information is acquired at least one character before the end of the utterance by the user, or at least one character before the user can interrupt and perform the response.
A speech speed calculating step of calculating a current speech speed based on the voice input in the input step,
The prediction step includes
Acquire at least one of the number of words, the number of morphemes, the number of parts of speech, the number of phonemes, the word string, the morpheme string, the part of speech string, and the phoneme string after the timing prediction information is acquired by the prediction step until the response timing is reached. Then, based on these and the speech speed calculated in the speech speed calculation step, a response timing arrival time until the response timing is reached is calculated, and the response timing arrival time is used as the timing prediction information. A featured voice interaction method. An input step for inputting voice uttered by the user;
A prediction step of acquiring timing prediction information indicating a prediction result of a response timing for performing a response corresponding to the input voice based on the voice input in the input step;
A timing determination step for determining whether or not the response timing has been reached based on the timing prediction information acquired in the prediction step;
When it is determined that the response timing is reached by the timing determination step, the response step performs the response, and
The prediction step includes
The timing prediction information is acquired at least one character before the end of the utterance by the user, or at least one character before the user can interrupt and perform the response.
A voice interaction method characterized by predicting a time that elapses after the timing prediction information is acquired by the prediction step until the response timing is reached, and the predicted time is used as the timing prediction information. An input step for inputting voice uttered by the user;
A prediction step of acquiring timing prediction information indicating a prediction result of a response timing for performing a response corresponding to the input voice based on the voice input in the input step;
A timing determination step for determining whether or not the response timing has been reached based on the timing prediction information acquired in the prediction step;
When it is determined that the response timing is reached by the timing determination step, the response step performs the response, and
The prediction step includes
The timing prediction information is acquired at least one character before the end of the utterance by the user, or at least one character before the user can interrupt and perform the response.
A voice interaction method characterized by predicting the number of frames that continue from when the timing prediction information is acquired by the prediction step until the response timing is reached, and using the predicted number of frames as the timing prediction information. An input step for inputting voice uttered by the user;
A prediction step of acquiring timing prediction information indicating a prediction result of a response timing for performing a response corresponding to the input voice based on the voice input in the input step;
A timing determination step for determining whether or not the response timing has been reached based on the timing prediction information acquired in the prediction step;
When it is determined that the response timing is reached by the timing determination step, the response step performs the response, and
The prediction step includes
The timing prediction information is acquired at least one character before the end of the utterance by the user, or at least one character before the user can interrupt and perform the response.
A spoken dialogue method, comprising: at least one of the number of words, the number of morphemes, the number of parts of speech, and the number of phonemes that continue from when the timing prediction information is acquired by the prediction step until the response timing is reached. An input step for inputting voice uttered by the user;
A prediction step of acquiring timing prediction information indicating a prediction result of a response timing for performing a response corresponding to the input voice based on the voice input in the input step;
A timing determination step for determining whether or not the response timing has been reached based on the timing prediction information acquired in the prediction step;
When it is determined that the response timing is reached by the timing determination step, the response step performs the response, and
The prediction step includes
The timing prediction information is acquired at least one character before the end of the utterance by the user, or at least one character before the user can interrupt and perform the response.
The timing prediction information is
A spoken dialogue method, comprising: at least one of a word string, a morpheme string, a part-of-speech string, and a phoneme string after the timing prediction information is acquired by the prediction step and before the response timing is reached. The prediction step includes
For the prediction model in which the response timing is determined in advance and the speech input in the input step, the timing prediction information is obtained by comparing the feature amount of the prediction model with the feature amount of the input speech. The voice dialogue method according to claim 10, wherein the voice dialogue method is obtained. The feature amount is
The spoken dialogue method according to claim 15, wherein at least one of a syntactic feature indicating a syntactic feature and a prosodic feature indicating a prosodic feature of an utterance by a user is provided. The syntactic feature is
Information including at least one of a preset keyword, word string, morpheme string, part of speech string, phoneme string,
The prosodic feature amount is:
The speech according to claim 16, characterized in that it is information including at least one of speech length, basic frequency time-series information, pitch time-series information, power time-series information, and speech speed time-series information. How to interact. The response timing is
The voice interaction method according to any one of claims 10 to 17, wherein the response is made at the timing when the response is made in duplicate with the user's utterance.

Images