JP2018017776A - Voice interactive device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice interactive device capable of speaking in consideration of an individual difference in real time.SOLUTION: The voice interactive device has a speech reception part for identifying speech of a person and a speech part for speaking. The speech part speaks when a silent time which is a time during which the speech reception part does not receive any speech after the speech part speaks exceeds a predetermined standby time. The voice interactive device also has: a storage part for storing a response time starting from a speech by the speech part through a response by the person until reception of the speech by the speech reception part; an estimation part for calculating an average and a dispersion of the response time stored by the storage part, creating a Gamma distribution concerning the response time and the speech probability by using the calculated average and dispersion and estimating a silence time which allows determination that the person does not have intention to speak by using the created Gamma distribution; and a speech control part for updating the standby time in the speech part, setting the silent time estimated by the estimation part as a new standby time.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a voice interaction apparatus that performs a conversation with a person.

In recent years, a speech dialogue apparatus such as a dialogue robot for carrying out a dialogue with a person has been put into practical use. However, in a situation where the voice interaction device interacts with a person, both the person and the voice interaction device may be silent, and if such a silence occurs frequently, the conversation is interrupted and an awkward atmosphere is created.
Therefore, recently, a technique has also been proposed in which speech is spoken from the voice interaction device side when the silent time exceeds a predetermined waiting time (for example, Patent Document 1).

JP 2010-152119 A

However, as in Patent Document 1, when the silent time exceeds a predetermined waiting time, in the technology of speaking from the voice interactive apparatus side, the waiting time becomes a predetermined time and waits in consideration of individual differences. There was a problem that the time could not be changed.
In addition, when a person's utterance tendency changes with the passage of time (for example, getting used to a voice interaction device to shorten the response time), the waiting time cannot be changed in response to the change in the utterance tendency. There was also.

  The present invention has been made in view of the above, and an object of the present invention is to provide a voice interactive apparatus capable of uttering in consideration of individual differences in real time.

A spoken dialogue apparatus according to one aspect of the present invention includes:
An utterance receiving unit for recognizing a person's utterance, and an uttering unit for uttering, wherein the utterance unit determines a silent time which is a time when the utterance receiving unit does not accept an utterance after the utterance unit utters A spoken dialogue device that utters when a specified waiting time is exceeded,
A storage unit that stores a response time from when the utterance unit utters until a person responds and the utterance reception unit receives the utterance;
The average and variance of the response times stored in the storage unit are calculated, a gamma distribution relating to the response time and the utterance probability is created using the calculated average and variance, and a person speaks using the created gamma distribution. An estimation unit for estimating a silent time that can be determined to be absent,
An utterance control unit that updates the waiting time in the utterance unit with the silent time estimated by the estimation unit as a new waiting time.

  According to the aspect of the present invention described above, there is an effect that it is possible to provide a voice interactive apparatus capable of speaking in real time in consideration of individual differences.

It is a block diagram which shows the structural example of the voice interactive apparatus which concerns on embodiment. It is a figure explaining the operation example of the measurement part which concerns on embodiment. It is a figure explaining the operation example of the estimation part which concerns on embodiment. It is a flowchart explaining the example of an operation | movement after the utterance part uttered in the speech dialogue apparatus which concerns on embodiment.

Embodiments of the present invention will be described below.
<Outline of the embodiment>
First, an outline of the voice interactive apparatus according to the present embodiment will be described.
In many cases, a speech dialogue apparatus, such as a dialogue robot, is designed so that when a person speaks, the voice dialogue apparatus responds to the utterance and speaks. Therefore, when a person speaks, the silent time does not occur, or even if it occurs, it is very short time, so it is considered that the possibility that the dialogue is interrupted is low.

  On the other hand, when a voice dialogue device utters, the person speaks in response to the utterance, so the silent time (that is, after the voice dialogue device utters, the person's utterance is spoken) There is a possibility that the time (which was not accepted by the interactive device) will be longer. Therefore, if the silent time that has elapsed since the voice dialogue device uttered exceeds the waiting time, it is determined that there is no intention to speak to the person, and the dialogue is interrupted by speaking from the voice dialogue device side. Can be avoided.

  However, there is an individual difference in the silent time when it can be determined that a person is not willing to speak, and in real time according to changes in the utterance tendency (for example, the response time is shortened after becoming accustomed to a voice interaction device). It will change.

  Therefore, in the configuration in which the voice interaction device according to the present embodiment utters when the silent time that has elapsed since the voice interaction device uttered exceeds the waiting time, the waiting time is calculated in real time with the individual difference of the person. It will be changed in consideration.

<Configuration of the embodiment>
Next, with reference to FIG. 1, the configuration of the voice interactive apparatus 1 according to the present embodiment will be described. As shown in FIG. 1, the voice interaction device 1 is a device that interacts with a user 2 who is a person, such as a conversation robot. The voice interaction apparatus 1 includes an utterance unit 11, an utterance reception unit 12, a measurement unit 13, a filter unit 14, an estimation unit 15, and an utterance control unit 16. The filter unit 14 is an example of a storage unit. FIG. 1 shows only the essential components of the present invention, and other components are omitted.

The utterance unit 11 utters the user 2 by outputting sound through a speaker (not shown).
The utterance reception unit 12 recognizes the utterance of the user 2 by inputting the voice uttered by the user 2 via a microphone (not shown).

  The measurement unit 13 measures a response time that is a time from when the utterance unit 11 speaks until when the user 2 responds and the utterance reception unit 12 accepts the utterance. Specifically, the response time of the user 2 is a time from when the utterance unit 11 finishes utterance to when the utterance reception unit 12 accepts a voice uttered in response to the utterance by the user 2. For example, in the example of FIG. 2, the timing when the utterance unit 11 finishes the utterance of “sure” is the time t1, and the voice “the physical condition is good” uttered in response to the utterance by the user 2 The timing received by the utterance receiving unit 12 is time t2. Therefore, the response time of the user 2 is 1 second which is a difference between the time t1 and the time t2.

  The filter unit 14 filters the response time data, which is the response time data of the user 2 measured by the measurement unit 13, removes outlier response time data, and stores the remaining response time data. For example, when the voice interactive device 1 is left unattended or when the user 2 ponders and responds, the response time of the user 2 becomes very long. The filter unit 14 removes such response time data as outlier response time data.

  The filter unit 14 stores N response time data for the past N times, and discards the oldest response time data when the number exceeds N. As will be described later, the estimation unit 15 calculates the variance and average of the response time of the user 2 using the response time data, and the silent time (the user 2 is uttered) To the time when the utterance receiving unit 12 does not accept the utterance). In general, at least three pieces of data for calculating the variance and the average are required. Therefore, the above N is an integer of 3 or more. If N is large, the estimation accuracy increases, but old data remains, and if it is small, it becomes easier to deal with the current situation, but the estimation accuracy decreases. Therefore, the above N may be determined to an appropriate numerical value of 3 or more in consideration of the current state and the priority of estimation accuracy.

  Any method can be used as a method for the filter unit 14 to remove outlier response time data. For example, the filter unit 14 may use a median filter to remove response time data that deviates from the vicinity of the median value. Alternatively, the filter unit 14 may use a filter that uses IQR (interquartile range) to remove response time data of the whisker portion of the box plot that deviates from IQR. A filter using IQR can be used even when the distribution is not normal, and has advantages such as a small amount of calculation and high-speed response. Alternatively, the measurement unit 13 may add a label to the response time data when measuring the response time, and the filter unit 14 may remove the response time data to which a specific label is added. For example, it is conceivable that the measurement unit 13 gives a specific label to response time data that can be determined that the user 2 has responded by pondering from an image captured by a camera (not shown). Further, it is conceivable that the measurement unit 13 recognizes the voice of the user 2 and gives a specific label to the response time data of the voice that does not include an independent word. Moreover, it is possible that the measurement part 13 gives a specific label to the response time data when it is waiting for the operation | movement completion of the voice interactive apparatus 1. FIG.

  The estimation unit 15 first calculates the average and variance of the response times of the user 2 using the N response time data stored in the filter unit 14. Subsequently, the estimation unit 15 creates a gamma distribution relating to the response time of the user 2 and the probability of speaking at the response time using the average and variance of the calculated response times of the user 2. The gamma distribution is an example of a parametric model. Subsequently, the estimation unit 15 estimates the silent time during which it can be determined that the user 2 is not willing to speak using the created gamma distribution.

  For example, it is assumed that the estimation unit 15 creates a gamma distribution as shown in FIG. FIG. 3 shows an example in which the number of response time data is 15 and the threshold of the utterance probability is 95%. In the example of FIG. 3, the utterance probability that the user 2 utters within a response time of 2.4 seconds or less is 95%, which is a threshold value. In other words, if the silent time that has elapsed since the utterance unit 11 uttered is 2.4 seconds or more, the user 2 can determine that the user 2 is not willing to utter with a probability of 95%. Therefore, the estimation unit 15 estimates the silent time that can be determined that the user 2 is not willing to speak as 2.4 seconds. Note that the number of response time data and the threshold of the utterance probability are not limited to the numerical values shown in FIG. 3, but may be other numerical values.

  The utterance control unit 16 controls the utterance timing of the utterance unit 11. Specifically, the utterance control unit 16 controls the utterance unit 11 to utter in response to the utterance when the user 2 utters. In addition, when the utterance unit 11 utters, the utterance control unit 16 controls the utterance unit 11 to utter when the current silent time after the utterance unit 11 utters exceeds the waiting time.

  Here, since no response time data is stored in the filter unit 14 when the dialogue with the user 2 is started, the estimation unit 15 cannot estimate the silent time. Therefore, the utterance control unit 16 uses a predetermined waiting time as a waiting time in the uttering unit 11 when starting a conversation with the user 2. Thereafter, when the dialogue with the user 2 proceeds and three or more response time data are stored in the filter unit 14, the estimation unit 15 can estimate the silent time. Therefore, thereafter, each time the estimation unit 15 estimates the silent time, the speech control unit 16 updates the waiting time in the speaking unit 11 with the estimated silent time as a new waiting time.

  The utterance control unit 16 also controls the utterance content of the utterance unit 11. For example, when the speech control unit 16 speaks when the current silent time exceeds the waiting time, the speech control unit 16 utters, “Did you go to play the last consecutive holiday?”, Etc., and provides a new topic. To control. However, what kind of content is determined as utterance content is not an essential part of the present invention, and since utterance content can be determined by applying a known method, detailed description thereof is omitted.

  In addition, each component with which the voice interactive apparatus 1 is provided is realizable with the processor (not shown) and memory (not shown) with which the voice interactive apparatus 1 which is a computer is provided, for example. Specifically, each component can be realized by the processor reading and executing software (program) from the memory. Further, each component is not limited to being realized by software by a program, and may be realized by any combination of hardware, firmware, and software.

  The above-described program can be stored using various types of non-transitory computer readable media and supplied to a computer. Non-transitory computer readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (for example, flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (for example, magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / W and semiconductor memory (for example, mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (random access memory)) are included.

  Further, the program may be supplied to the computer by various types of temporary computer readable media. Examples of transitory computer readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer readable medium can supply the program to the computer via a wired communication path such as an electric wire and an optical fiber, or a wireless communication path.

<Operation of the embodiment>
Then, with reference to FIG. 4, the operation | movement after the utterance part 11 utters is demonstrated in the voice interactive apparatus 1 which concerns on this Embodiment.
As shown in FIG. 4, first, the utterance control unit 16 determines whether or not the utterance unit 11 has uttered (step S1). The utterance of the utterance unit 11 in step S1 includes the utterance at the start of the dialogue, the utterance with respect to the utterance of the user 2, and the utterance when the silent time elapsed after the utterance of the utterance unit 11 exceeds the waiting time.

  In step S1, when the utterance unit 11 speaks (Yes in step S1), the utterance control unit 16 determines whether the user 2 responds to the utterance of the utterance unit 11 (step S2), and the response of the user 2 is If there is not (No in step S2), it is then determined whether or not the current silent time that has elapsed since the utterance of the utterance unit 11 has exceeded the waiting time (step S3). The waiting time here is a predetermined waiting time at the start of the dialogue with the user 2, and when the dialogue with the user 2 proceeds and the estimation unit 15 can estimate the silent time, the estimation unit 15 estimated the waiting time. It becomes a new waiting time updated with silence time. In step S3, when the current silent time does not exceed the waiting time (No in step S3), the process returns to step S2. That is, steps S2 and S3 are repeated until the current silent time exceeds the waiting time or the user 2 responds during the waiting time.

  In step S3, when the current silent time exceeds the waiting time (Yes in step S3), the utterance control unit 16 determines that the user 2 is not willing to utter and controls the utterance unit 11 to utter ( Step S4). After step S4, the process returns to step S1.

  In step S2, when the user 2 responds to the utterance of the utterance unit 11 (Yes in step S2), the measurement unit 13 responds after the utterance unit 11 utters and the user 2 responds to the utterance reception unit. The response time until 12 accepts the utterance is measured (step S5). Subsequently, the filter unit 14 filters the response time data of the user 2 measured by the measurement unit 13 to remove outlier response time data, and stores the remaining response time data (step S6). Subsequently, the estimation unit 15 calculates the average and variance of the response time of the user 2 using the response time data stored in the filter unit 14, and uses the calculated average and variance to determine the response time of the user 2 and A gamma distribution related to the utterance probability is created, and the silent time during which the user 2 can determine that he / she does not want to utter is estimated using the created gamma distribution. Subsequently, the utterance control unit 16 updates the waiting time in the utterance unit 11 with the silent time estimated by the estimating unit 15 as a new waiting time (step S7). In step S7, when three or more response time data are not stored in the filter unit 14, the estimation unit 15 cannot estimate the silent time. Therefore, in that case, the utterance control unit 16 does not update the waiting time in the uttering unit 11 and keeps the current waiting time. After step S7, the process returns to step S1.

  The above is the operation after the utterance unit 11 utters in the voice interaction apparatus 1. In the voice interaction apparatus 1, when the user 2 speaks, an operation of speaking in response to the speech is performed. However, the operation after the user 2 speaks is not an essential part of the present invention, and a known operation can be applied.

<Effect of Embodiment>
As described above, according to the voice interaction apparatus 1 according to the present embodiment, the response time from when the utterance unit 11 utters until the user 2 responds and accepts the utterance is stored, and the response time of the user 2 is stored. Create a gamma distribution related to response time and utterance probability using average and variance, estimate the silent time that can be determined that a person is not willing to utter using the created gamma distribution, and use the estimated silent time as a new waiting time Then, the waiting time in the utterance unit 11 is updated.

  Thus, since the waiting time in the utterance part 11 is updated according to the response time of the user 2, the waiting time can be changed in consideration of individual differences between people in real time. As a result, it is possible to provide the voice interactive apparatus 1 capable of uttering in consideration of individual differences in real time.

Moreover, since it is possible to utter in consideration of individual differences in real time, it is possible to avoid interruption of the dialogue with the user 2 and to continue the dialogue.
Also, since the gamma distribution, which is a parametric model, is used to estimate the silent time when it is determined that a person is not willing to speak, the silent time is estimated with a small amount of response time data (minimum of three response time data is sufficient). can do. Thereby, it is possible to shorten the time from the start of dialogue until the silent time is estimated.

  Also, since outlier response time data is filtered out, even if the response time happens to be long due to factors such as the user 2 pondering, the silent time is estimated by excluding such response time data. can do. Thereby, it can avoid that the estimation precision of silence time falls.

<Modification of Embodiment>
Note that the present invention is not limited to the above-described embodiment, and can be changed as appropriate without departing from the spirit of the present invention.
For example, in the above-described embodiment, the waiting time in the utterance unit 11 is set as a waiting time set in advance at the start of the conversation with the user 2, but the present invention is not limited to this. For example, if the dialogue with the user 2 is started in a state where the response time data of the user 2 at the previous dialogue with the user 2 is stored in the filter unit 14, the estimation unit 15 estimates the silent time from the beginning of the dialogue. be able to. As a result, the silence time estimated by the estimation unit 15 from the start of the conversation can be set as the waiting time.

  In the above embodiment, the estimation unit 15 uses the gamma distribution, which is an example of a parametric model, in order to estimate the silent time that can be determined that the user 2 is not willing to speak. However, the estimation unit 15 is not limited thereto. Not. The estimation unit 15 may use a model having skewness instead of the parametric model. The skewness is an index indicating the distribution asymmetry.

DESCRIPTION OF SYMBOLS 1 Voice dialogue apparatus 11 Utterance part 12 Utterance reception part 13 Measurement part 14 Filter part 15 Estimation part 16 Utterance control part 2 User

Claims (1)

An utterance receiving unit for recognizing a person's utterance, and an uttering unit for uttering, wherein the utterance unit determines a silent time which is a time when the utterance receiving unit does not accept an utterance after the utterance unit utters A spoken dialogue device that utters when a specified waiting time is exceeded,
A storage unit that stores a response time from when the utterance unit utters until a person responds and the utterance reception unit receives the utterance;
The average and variance of the response times stored in the storage unit are calculated, a gamma distribution relating to the response time and the utterance probability is created using the calculated average and variance, and a person speaks using the created gamma distribution. An estimation unit for estimating a silent time that can be determined to be absent,
A speech dialogue apparatus comprising: an utterance control unit that updates the waiting time in the utterance unit using the silent time estimated by the estimation unit as a new waiting time.

Images