JPH086590A - Speech recognition method and apparatus for speech dialogue - Google Patents

Abstract

(57) [Abstract] [Purpose] Utilizing the causal relationship between the system announcement and the user's utterance start time, it is possible to detect the utterance start time with high accuracy and perform high-accuracy speech recognition. Perform highly accurate voice recognition without making a single decision. [Structure] In the announcement utterance device 2, a silent section is provided during a system announcement to control the utterance start time of the user, and the calculation unit 12 uses the predicted distribution of the utterance start times corresponding to this to generate a first utterance start point. Likelihood a (t) is calculated, and on the other hand, the arithmetic unit 14 calculates the second utterance start point likelihood b (t) from the utterance detection characteristic parameter 13a, and from these, the arithmetic unit 15 calculates the third utterance start point likelihood. α (t)
Is calculated, the speech start time is determined by comparing α (t) with the reference value _Ref, and the switch 17 is turned on to start voice recognition.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice dialog device for performing a dialogue with a user (user) by using voice, and more particularly, to improve detection accuracy with respect to a user's utterance start time and voice recognition for the user's utterance. It is useful for improving accuracy.

[0002]

2. Description of the Related Art In a voice dialog device, a system announcement is made by talking to the user from the device side using voice, and by recognizing the user's utterance, that is, the voice uttered by the user, the user's intention is understood, and the user and the device are recognized. Have a dialogue between. Therefore, the accuracy of voice recognition is important.

With reference to FIG. 7, a voice recognition method and a voice recognition device in a conventional voice dialogue system will be described. In FIG. 7, the voice dialogue device includes a dialogue management device 1, an announcement utterance device 2, and a voice recognition device 50. The voice output circuit 3 and the voice input circuit 5 may be built in the voice dialog device, or may be separately connected to the voice dialog device as appropriate. An example of the latter is a telephone handset, which is connected to a voice interaction device through a telephone line and telephone exchange. The voice recognition device 50 includes a voice signal passage switch 51 for utterance detection and an acoustic analysis unit 1 for utterance detection.
3, a speech detection unit 52, a voice signal passage switch 17 for voice recognition, a sound analysis unit 18 for voice recognition, and a voice recognition unit 19.

The operation of the voice interactive apparatus shown in FIG. 7 and the function of each section will be described below.

(I) In the announcement utterance device 2, the dialogue management device 1 creates an electric signal 2a of a voice to be uttered based on the text 1a of the system announcement designated by the code name or the like, and sends it to the voice output circuit 3. . Also, an announcement start signal 2b indicating the start of the system announcement,
Alternatively, an announcement end signal 2c indicating the end of the system announcement is sent to the speech detection voice signal passing switch 51 of the voice recognition device 50. The voice output circuit 3 is an electric signal 2
A is converted into voice, and the system announcement 3a is heard by the user. The voice input circuit 5 receives the user's utterance 4 to the system announcement 3a, converts it into an electric voice signal 5a, and sends it to the voice signal passing switches 51 and 17 of the voice recognition device 50 for utterance detection and voice recognition.

(Ii) In the voice recognition device 50, in the case of accepting the interruption utterance of the user who is making the system announcement, the voice signal is passed from a time when the announcement start signal 2b is given to a certain time after the announcement end signal 2c is given. When the switch 51 is closed (ON) and the interruption utterance of the user who is making the system announcement is not accepted, the voice signal passing switch 51 is closed (ON) for a fixed time from when the announcement end signal 2c is given. The voice signal sent while the voice signal passing switch 51 is closed is sent to the utterance detection acoustic analysis unit 13 as the utterance detection target signal 51a.

(Iii) The acoustic analysis unit 13 calculates a characteristic parameter 13a suitable for detecting a user's utterance, such as a power spectrum, from the voice signal 51a that has passed through the voice signal passing switch 51 and sends it to the utterance detection unit 52. The utterance detection unit 52 determines one point each of the utterance start time and the utterance end time of the user based on the characteristic parameter 13a and sends a signal 52a designating the time point to the voice signal passing switch 17 for voice recognition.

(Iv) The voice signal passing switch 17 is closed (ON) only during a period designated by the signal 52a from the utterance detecting section 52, and the voice signal transmitted while the voice signal passing switch 17 is closed is the signal 17a for voice recognition. To the acoustic analysis unit 18 for voice recognition. The acoustic analysis unit 18 calculates a characteristic parameter 18a suitable for voice recognition from the voice signal 17a that has passed through the voice signal passing switch 17, and sends it to the voice recognition unit 19. In the voice recognition unit 19, the characteristic parameter 18
Speech recognition is performed based on a, and the recognition result 19a is sent to the dialogue management device 1.

(V) In the dialogue management device 1, the voice recognition unit 1
Based on the recognition result 19a given from 9, the system announcement text 1a to be uttered next is determined, and the code name or the like is sent to the announcement utterance device 2.

By repeating the above operation, a dialogue using voice is performed between the person and the device. If necessary, the dialogue management device 1 recognizes the user's intention from the dialogue contents and outputs the information 1b to the outside.

[0011]

Although the accuracy of voice recognition is important in the voice interactive apparatus, when the above-mentioned conventional technique is used to receive the interrupt utterance of the user,
There are points to be improved such as (a) and (b) below.

(A) Since the utterance detection unit 52 detects the utterance of the user using only the characteristic parameter 13a such as the power spectrum, the utterance start time detection accuracy is not good. Further, if the user makes a meaningless utterance (redundant word) or coughs during the system announcement, there is a high possibility that the time point is erroneously detected as the utterance start time of the user. As a result, the speech recognition is performed by including the meaningless utterance and cough in the recognition target, and the speech recognition accuracy decreases.

(B) Furthermore, since the utterance detection unit 52 determines the utterance start time of the user to be only one point, if an error occurs in the detection of the utterance start time, an error that the voice recognition unit 19 cannot recover. Therefore, definite error propagation occurs that the accuracy of voice recognition is reduced.

Therefore, an object of the present invention is to provide a voice recognition method and apparatus capable of performing high-accuracy voice recognition by improving the detection accuracy of the user's utterance start time, and further, to start the user's utterance start. Another object of the present invention is to provide a voice recognition method and device capable of performing highly accurate voice recognition by reducing the influence of time detection error even if there is an error.

[0015]

[Means for Solving the Problems] First to achieve the above object
In a voice recognition method applied to a voice interaction device for interacting with a user using voice, a prediction distribution having a maximum point of a user's utterance start time for a system announcement of the voice interaction device is prepared in advance. Based on this prediction distribution, the expected value at which the user's utterance starts is calculated as the first utterance start point according to the time after the system announcement starts; the user's utterance converted into an electric signal is calculated. Acoustic analysis is performed to calculate a characteristic parameter for utterance detection, and based on the characteristic parameter, a likelihood that the utterance of the user is likely to be started is calculated as second utterance starting point likelihood according to time; The likelihood of the utterance start point of 2 is weighted by the likelihood of the first utterance start point, and the likelihood of the third utterance start point obtained by this weighting is compared with a reference value, The time when it becomes larger than the value is determined to be the user's utterance start time; the user's utterance converted into an electric signal is acoustically analyzed to calculate a characteristic parameter for speech recognition, and the speech recognition is performed based on this characteristic parameter. Performing the process according to the determination of the utterance start time of the user;

A second aspect of the present invention is a voice recognition method applied to a voice dialogue device for dialogue with a user by using voice: Prediction having a maximum point of a user's utterance start time with respect to a system announcement of the voice dialogue device. Prepare a distribution in advance, and calculate the expected value at which the user's utterance starts based on this predicted distribution as the first utterance start point according to the time after the system announcement starts; converted to an electrical signal The utterance of the user is acoustically analyzed to calculate a characteristic parameter for utterance detection, and based on the characteristic parameter, the likelihood that the utterance of the user may have started
Of the utterance start point of the user according to the time;
The first reference value is weighted according to the likelihood of the first utterance to calculate the second reference value that changes with time, and the second reference value is calculated.
The utterance starting point of the person is compared with the second reference value, and the second
Is determined to be the user's utterance start time; the time when the utterance of the user is converted into an electric signal is acoustically analyzed to calculate a characteristic parameter for voice recognition, and based on the characteristic parameter Performing a process of performing voice recognition according to the determination of the utterance start time of the user;

In a third aspect of the present invention, there is provided a voice recognition method applied to a voice dialogue device for dialogue with a user by using a voice: Prediction having a maximum point of a user's utterance start time with respect to a system announcement of the voice dialogue device. Prepare a distribution in advance, and calculate the expected value at which the user's utterance starts based on this predicted distribution as the first utterance start point according to the time after the system announcement starts; converted to an electrical signal The utterance of the user is acoustically analyzed to calculate a characteristic parameter for utterance detection, and based on the characteristic parameter, the likelihood that the utterance of the user is likely to be started is determined as the second utterance start point according to the time. Calculating; weighting the likelihood of the second utterance start point with the likelihood of the first utterance start point to calculate the likelihood of the third utterance start point; The acoustic parameters of the user's utterance converted to are calculated to calculate the characteristic parameters for voice recognition, and based on the characteristic parameters, the recognition start times are shifted one after another to perform the voice recognition, and each recognition start time is corresponded. Calculating the likelihood for each voice recognition result; calculating the sum or product of the likelihood of the voice recognition result for each recognition start time and the third utterance starting point likelihood obtained by the weighting together with the time. Determining the voice recognition result corresponding to the recognition start time at which the calculated value becomes maximum as the voice recognition result for the user's utterance.

In a fourth aspect of the present invention, there is provided a voice recognition method applied to a voice dialogue device for dialogue with a user by using a voice: a prediction having a maximum point of a user's utterance start time with respect to a system announcement of the voice dialogue device. Prepare a distribution in advance, and calculate the expected value at which the user's utterance starts based on this predicted distribution as the first utterance start point according to the time after the system announcement starts; converted to an electrical signal The utterance of the user is acoustically analyzed to calculate a characteristic parameter for utterance detection, and based on the characteristic parameter, the likelihood that the utterance of the user is likely to be started is determined as the second utterance start point according to the time. Calculating; weighting the likelihood of the second utterance start point with the likelihood of the first utterance start point to calculate the likelihood of the third utterance start point; Utterances converted user by acoustic analysis to calculate the characteristic parameters for voice recognition, on the basis of the characteristic parameter, leading to perform speech recognition by searching the probability finite-state network with a silent state in the;
Updating the probability of transition from the silent state at the head of the finite state network with probability to the head state of the sentence according to the time using the third utterance starting point likelihood obtained by the weighting. It is a recognition method.

A fifth aspect of the invention is characterized in that, in the first to fourth aspects, the maximum point of the predicted distribution of the utterance start time of the user exists in the silent section of the system announcement, and the sixth aspect of the invention. Further, the silent section has a length of 0.2 seconds or more and 3 seconds or less, and is present in at least one of a sentence of the system announcement and a sentence.

Next, a seventh aspect of the present invention is a voice recognition device applied to a voice dialogue device for dialogue with a user using voice; the maximum point of a user's utterance start time with respect to a system announcement of the voice dialogue device. A first means for storing a prediction distribution having the following; based on the stored prediction distribution, an expected value at which the user's utterance is started is calculated as the first utterance start point according to the time after the system announcement is started. Second means; acoustic analysis of the user's utterance converted into an electric signal, and third means for calculating characteristic parameters for utterance detection; and utterance of the user based on the characteristic parameter for utterance detection And a fourth means for calculating the likelihood as the second utterance start point likelihood according to the time; and weighting the second utterance start point likelihood by the first utterance start point likelihood. Is compared with a reference value of the third utterance start point likelihood of; was carried out, the weighted values fifth means and for calculating in response to time as a third utterance start point ness
Sixth means for determining a time point when the value becomes larger than the reference value as a user's utterance start time; acoustic analysis of the user's utterance converted into an electric signal to calculate a characteristic parameter for voice recognition, and this characteristic And a seventh means for performing a process of performing voice recognition based on a parameter according to the determination of the utterance start time of the user.

An eighth aspect of the present invention is a voice recognition device applied to a voice dialogue device for dialogue with a user using voice; a prediction having a maximum point of a user's utterance start time with respect to a system announcement of the voice dialogue device. First means for storing a distribution; second means for calculating an expected value at which a user's utterance starts based on the stored predicted distribution as a first utterance start point in accordance with a time after the system announcement is started And; Thirdly, acoustic analysis is performed on the user's utterance converted into an electric signal, and characteristic parameters for utterance detection are calculated.
Fourth means for calculating the likelihood that the user's utterance has started based on the characteristic parameter for utterance detection as the second utterance start point according to the time;
Means for weighting the reference value of the first utterance starting point likeness and calculating the weighted value as the second reference value according to the time; Sixth means for comparing with a second reference value obtained by weighting, and determining a time point when the second reference value is larger than the second reference value as the utterance start time of the user; And a seventh means for performing a process of performing characteristic analysis for voice recognition by acoustic analysis and performing voice recognition based on the characteristic parameter according to the determination of the utterance start time of the user; It is a recognition device.

A ninth aspect of the present invention is a switch, wherein the seventh means of the seventh or eighth aspect of the invention passes the user's utterance converted into an electric signal from the time when it is determined to be the user's utterance start time. Means; acoustic analysis means for voice recognition for acoustically analyzing the utterance of the user passing through the switch means to calculate a characteristic parameter for voice recognition; and a characteristic parameter for voice recognition calculated by the acoustic analysis means. Voice recognition means for performing voice recognition based on the above; A tenth invention is that the seventh means in the seventh or eighth invention starts the acoustic analysis of the user's utterance converted into an electric signal from the time when it is determined to be the user's utterance start time, and outputs the voice. A voice recognition acoustic analysis unit for calculating a recognition feature parameter; and a voice recognition unit for performing voice recognition based on the voice recognition feature parameter calculated by the sound analysis unit; To do. An eleventh aspect of the invention is that the seventh means of the seventh or eighth aspect of the invention is an acoustic analysis means for voice recognition, which acoustically analyzes the user's utterance converted into an electric signal to calculate a characteristic parameter for voice recognition. Of the characteristic parameters for voice recognition calculated by this acoustic analysis means,
And a voice recognition means for performing voice recognition based on the characteristic parameter after the time when it is determined to be the utterance start time of the user.

Next, a twelfth aspect of the invention is a voice recognition device applied to a voice dialogue device for dialogue with a user using voice; the maximum point of a user's utterance start time with respect to a system announcement of the voice dialogue device. A predictive distribution storage unit for storing the predictive distribution having; an expected value at which the user's utterance starts is calculated based on the stored predictive distribution as a first utterance start point in accordance with a time after the system announcement is started. First computing means; speech analysis means for utterance detection for acoustically analyzing the utterance of the user converted into an electric signal, and calculating characteristic parameters for utterance detection; user based on the characteristic parameters for utterance detection Second computing means for calculating the likelihood that the utterance of ”is likely to have started as the second utterance start point likelihood according to the time; for the second utterance start point likelihood Performs weighting by the first utterance start point ness, and third arithmetic means for calculating according to the time the weighted value as the third utterance start point ness;
Acoustic analysis of the user's speech converted into the electrical signal,
An acoustic analysis unit for voice recognition for calculating a feature parameter for voice recognition; based on the feature parameter for voice recognition, the recognition start time is shifted one after another to perform voice recognition, and corresponding to each recognition start time. A voice recognition means for calculating a likelihood for each voice recognition result; a sum or product of the likelihood of the voice recognition result for each recognition start time and the likelihood of the third utterance start point is calculated according to the time, and A voice recognition device, comprising: a voice recognition result determining means for determining a voice recognition result corresponding to a recognition start time at which the calculated value becomes maximum as a voice recognition result for a user's utterance.

A thirteenth aspect of the present invention is a voice recognition device applied to a voice dialogue device for dialogue with a user using voice; prediction having a maximum point of a user's utterance start time with respect to a system announcement of the voice dialogue device. A predictive distribution storage means for storing a distribution; a first calculating an expected value at which a user's utterance starts based on the stored predictive distribution as a first utterance start point in accordance with a time after a system announcement is started An utterance detecting acoustic analysis means for acoustically analyzing a user's utterance converted into an electric signal to calculate a utterance detecting characteristic parameter; and a user utterance based on the utterance detecting characteristic parameter. Second computing means for calculating the likelihood that the utterance has started as the second utterance starting point likelihood according to the time; Third computing means for performing weighting according to the utterance start point likelihood of the user and calculating the weighted value as the third utterance start point likelihood in accordance with time; acoustic analysis of the user's utterance converted into the electric signal. And an acoustic analysis unit for voice recognition for calculating a feature parameter for voice recognition;
A speech recognition means for performing speech recognition by searching a finite state network with a probability having a silent state at the beginning; a probability of transition from a silent state at the beginning of the finite state network with probability to a beginning state of a sentence, a third utterance And a transition probability updating means for updating according to the time using the likelihood of the starting point;

The fourteenth invention is the seventh to thirteenth inventions.
In the invention, the maximum point of the predicted distribution of the utterance start time of the user exists in the silent section of the system announcement, and the fifteenth invention further has a length of 0.2 seconds or more in the silent section. It is less than a second, and is characterized by being present in at least one of the sentence of the system announcement and the sentence and the phrase.

A sixteenth aspect of the present invention is the speech recognition apparatus according to the seventh to fifteenth aspects of the invention, which converts the designated text of the system announcement into an electrical voice signal and starts the system announcement by the speech recognition apparatus. An announcement voicing device for notifying, and a dialogue management device for managing a dialogue with a user by using a voice by designating a text of a system announcement for the announcement voicing device and inputting a voice recognition result from the voice recognition device. Is a voice interaction device.

[0027]

In the first, second and seventh to eleventh inventions, the likelihood that the user's utterance may have started from the characteristic parameter for utterance detection obtained by the acoustic analysis (the second utterance starting point ) Is determined to determine the utterance start time, the second utterance start point likelihood or the reference value is weighted by the first utterance start point likelihood obtained from the prediction distribution. As a result, it is possible to highly accurately determine one point of the speech start time of the user, and the accuracy of voice recognition is improved. Further, since the user's utterance start time can be determined with high accuracy, the interrupt utterance of the user during the system announcement can be recognized with high accuracy, and the use time of the voice dialog device can be shortened.

In the third, fourth, twelfth and thirteenth inventions, highly accurate voice recognition is possible without deciding the user's utterance start time at one point.

First, in the third and twelfth inventions, a large number of speech recognitions are performed with their start times being shifted one after another, and the likelihood for each speech recognition result corresponding to each recognition start time is obtained. The optimal speech recognition result is determined from the third utterance start point likelihood obtained by weighting the second utterance start point likelihood by the utterance start point likelihood. As a result, highly accurate voice recognition can be performed. It is also possible to start the voice recognition from the time when the likelihood of the third utterance start point exceeds a predetermined level, which reduces the processing amount of the voice recognition. In the third and twelfth inventions, even if it is not possible to highly accurately determine one point of the utterance start time of the user, the accuracy of voice recognition is improved as a result.

Next, in the fourth and thirteenth inventions, it is assumed that speech recognition is performed by searching a probability-added finite state network having a silent state at the head. At that time, the probability of transition from the silent state at the head to the head state of the sentence is
The third utterance start point likelihood obtained by weighting the second utterance start point likelihood is changed using the utterance start point likelihood. Therefore, substantial voice recognition is not performed while the utterance start is uncertain, and is started at the most probable utterance start time, and highly accurate voice recognition can be performed. In the fourth and thirteenth inventions, even if the user's utterance start time cannot be determined with high accuracy, the accuracy of the voice recognition is improved as a result, and compared with the third and twelfth inventions, Since it is not necessary to perform voice recognition in parallel by shifting start times one after another, high-speed processing is possible and the memory capacity can be reduced.

In the fifth, sixth, fourteenth and fifteenth inventions, a prediction distribution for obtaining an expected value at which the utterance of a user with higher reliability is started is obtained. The inventors investigated what kind of causal relationship exists between the system announcement and the user's utterance start time. This is because, if there is a characteristic causal relationship, it is considered that the user's utterance start time can be accurately detected by using this.

More specifically, an experiment was conducted in which a large number of users were made to use the voice dialog device and the time and frequency at which the user started speaking after the system announcement was started were examined. As a result, it was found that the utterance start time of the user has a distribution having a maximum point. In particular, if the user speaks after interrupting the system announcement, the fifth
It can be seen that the utterance start time is distributed mainly in the silent section of the system announcement as in the fourteenth invention, and further between the sentences or between the phrases as in the sixth and fifteenth inventions. When a certain silent section is positively provided, the distribution peak becomes steep, and this tendency is remarkable when the silent section is preferably 0.2 seconds to 3 seconds (more preferably 0.4 to 1.5 seconds). Was found. It was also found that when the user starts uttering after the system announcement ends, the utterance start time has a specific distribution around the time immediately after the system announcement ends. It should be noted that the silent section is a silent section not only when there is no sound at all but also when a chime or background music is played and when it can be said to be a substantially silent state for the system announcement. The silent section is controlled so as to prompt the user to start speaking.

Therefore, based on such an experiment, a prediction distribution 100 having the maximum point of the user's utterance start time for the system announcement as shown in FIG. 2 is prepared in advance and prepared, or no sound is produced regardless of the experiment. Prepare a prediction distribution that has the maximum point of the user's utterance start time for the system announcement by using a probability distribution such as a normal distribution, Poisson distribution, and chi-square distribution so that it has a maximum point before or after the interval. Thus, it is possible to obtain the expected value (likely as the first utterance start point) at which the user's utterance will start depending on the time after the system announcement is started.

In the sixteenth invention, the user and the device can talk with each other under highly accurate voice recognition.

[0035]

Embodiments of the invention will be described below with reference to the drawings. In the drawings, FIG. 1 shows a block configuration of a voice dialogue apparatus according to the first embodiment. FIG. 2 shows an example obtained by experimentally observing the predicted distribution of the user's utterance start time for the system announcement. Further, FIG. 3 shows a block configuration of a voice dialog device according to the second embodiment.
FIG. 4 shows a block configuration of a voice dialogue apparatus according to the third embodiment, and FIG. 5 shows a block configuration of a voice dialogue apparatus according to the fourth embodiment. FIG. 6 shows an example of a finite state network with probability having a silent state at the beginning.

<First Embodiment> As shown in FIG. 1, a voice dialogue apparatus according to the first embodiment comprises a dialogue management apparatus 1, an announcement utterance apparatus 2, and a voice recognition apparatus 10. Therefore, the voice output device 3 and the voice input device 5 may be built in the voice dialogue device as necessary, or may be appropriately connected as a separate object apart from the voice dialogue device. When the voice dialogue device is used in the extension telephone reception system, the handset of the telephone corresponds to the voice output circuit 3 and the voice input circuit 5, and is connected to the voice dialogue device through the telephone line and the telephone exchange. The speech recognition device 10 includes a prediction distribution storage unit 11
A first utterance starting point likelihood calculating section 12, an utterance detecting acoustic analysis section 13, a second utterance starting point likelihood calculating section 14, and a third utterance starting point likelihood calculating section 15. An utterance start time determination unit 16, a voice signal passage switch 17 for voice recognition, a sound analysis unit 18 for voice recognition, and a voice recognition unit 19 are provided.

The announcement utterance device 2 is the dialogue management device 1.
Generates an electric signal 2a of a voice to be uttered based on the system announcement text 1a designated by the code name or the like and sends it to the voice output circuit 3. At this time, as shown in FIG. 2, the announcement voicing device 2 has a certain silent section 2 between sentences of the system announcement or between phrases.
00 is provided to create a voice electric signal 2a. In the present embodiment, the length of the silent section 200 is set to about 0.5 seconds, but generally 0.2 seconds or more and 3 seconds or less is appropriate, and more preferably 0.4 seconds or more and 1.5 seconds or less. To do.
If the silent section is too long, the user feels uneasy. The silent section is a silent section not only when there is no signal at all but also when there is substantially no sound for the system announcement, for example, when a chime or background music is playing. Further, the announcement utterance device 2 sends an announcement start signal 2b indicating the start of the system announcement to the voice recognition device 10. It should be noted that the start of the system announcement does not mean only the time itself at which the voice starts to be output to the user, but the system announcement may be started at a certain time before the start of the voice output.

The voice output circuit 3 converts the electric signal 2a sent from the announcement utterance device 2 into a voice and makes the user hear it as a system announcement 3a. Since there is a user utterance 4 for this system announcement 3a,
The speech input circuit 5 converts the utterance 4 into an electric signal 5a and sends it to the speech recognition apparatus 10.

In the voice recognition device 10, the prediction distribution storage unit 1
1 stores a predicted distribution 100 of user's utterance start times for system announcements as shown in FIG. This predicted distribution 100 is made by a system announcement in which about 500 users have previously used the voice conversation device of the extension telephone reception system and provided a silent interval 200 of about 0.5 seconds between sentences from the device. It is created from the experimental result of observing the distribution of the utterance start time of each user in the case of making it. In FIG. 2, the horizontal axis represents time t when the start of the system announcement is time 0, and the vertical axis represents the expected value at which the user's utterance starts at each time t. There is a maximum in the distribution. It should be noted that, regardless of the experiment, by using a probability distribution such as a normal distribution, a Poisson distribution, or a chi-double distribution having a maximum point before or after the silent section, each time when the system announcement is started at time 0 It is also possible to create a distribution of expected values at which the user's utterance starts at t and use it as a predictive distribution having a maximum point of the user's utterance start time with respect to the system announcement.

Based on the predicted distribution in the predicted distribution storage unit 11, the arithmetic unit 12 determines the time t from the time when the announce start signal 2b is received from the announcer 2 based on the predicted distribution in the predicted distribution storage unit 11. An expected value a (t) at which the user's utterance will start is calculated and sent to the calculation unit 15.

The speech detection acoustic analysis unit 13 inputs the electrical voice signal 5a provided from the voice input circuit 5 to constantly perform acoustic distribution, successively calculates the characteristic parameters 13a for speech detection, and the arithmetic unit 14 Send to.

Based on the utterance detection characteristic parameter 13a, the calculation unit 14 determines the likelihood b (t) at which the utterance of the user is likely to be started at time t as the second utterance starting point.
According to the above, and sends it to the calculation unit 15. However, the start of the system announcement is time 0.

The arithmetic unit 15 determines whether the first utterance starting point is a.
The second speech start point likelihood b (t) is weighted by (t) to calculate the third speech start point likelihood α (t),
It is sent to the utterance start time determination unit 16. Equations (1) to (3) are given here as examples of weighting. However, in the formula (2), 0 <k ₁ <1.

Α (t) = a (t) + b (t) Equation (1) α (t) = k ₁ · a (t) + (1-k ₁ ) · b (t) Equation (2) ) Α (t) = a (t) · b (t) (3)

The utterance start time determination unit 16 compares the third utterance start point likelihood α (t) with a reference value R _ef fixed in advance, and when α (t)> R _ef first, or α
When (t)> R _ef continues for a certain period of time, first α (t)>
The time when it becomes _Ref is determined as the utterance start time of the user,
An utterance start signal 16a indicating that is sent to the voice signal passing switch 17 for voice recognition.

The switch 17 is turned on from the time when the utterance start time 16a is given, passes the voice signal 5a, and sends it as the voice recognition target signal 17a to the voice analysis acoustic analysis section 18.

The acoustic analysis unit 18 acoustically analyzes the voice signal 17a that has passed through the switch 17, successively calculates the characteristic parameters 18a for voice recognition, and sends them to the voice recognition unit 19.

The voice recognition unit 19 performs voice recognition based on the feature parameter 18a for voice recognition. The recognition result 19a is sent to the dialogue management device 1.

In the dialogue management device 1, the text 1 of the system announcement to be spoken next is based on the recognition result 19a.
A is determined, and the code name or the like is sent to the announcement voicing device 2. Further, the user's intention is recognized from the content of the dialogue with the user, and, for example, in the case of an extension telephone reception system, the extension number information 1b is output to the outside. A dialogue is carried out by repeating the above-described operation of each of the devices 1, 2 and 10.

In the above description of the first embodiment, the switch 1
Although only the signal 17a of the voice recognition target is given to the acoustic analysis unit 18 using the switch 7, it may be changed as follows without using the switch 17. (1) It is assumed that the voice signal 5a from the voice input circuit 5 is constantly sent to the acoustic analysis unit 18, and the utterance start time determination unit 16 sends the utterance start signal 16a to the acoustic analysis unit 18, which starts the utterance. The acoustic analysis is started from the time when the signal 16a is given. (2) Alternatively, the voice signal 5a from the voice input circuit 5 is constantly sent to the acoustic analysis unit 18, and the acoustic analysis unit 18 always performs acoustic analysis to determine the characteristic parameter 18a as the voice recognition unit 1.
9, and the speech start time determination unit 16 further transmits the speech start signal 16a to the speech recognition unit 19. The speech recognition unit 19 performs speech recognition using the characteristic parameter 18a from the time when the speech start signal 16a is given. To start.

<Second Embodiment> As shown in FIG. 3, the voice dialogue apparatus according to the second embodiment comprises a dialogue management apparatus 1, an announcement utterance apparatus 2, and a voice recognition apparatus 20. Therefore, the voice output device 3 and the voice input device 5 may be built in the voice dialogue device as necessary, or may be appropriately connected as a separate object apart from the voice dialogue device. The voice recognition device 20 includes a prediction distribution storage unit 11, a first utterance start point likelihood calculation unit 12, and an utterance detection acoustic analysis unit 13.
A second utterance starting point likelihood calculating unit 14, a reference value calculating unit 21, a utterance start time determining unit 22, a voice signal passing switch 17 for voice recognition, and a sound analyzing unit 18 for voice recognition. , And a voice recognition unit 19. Of these devices, the calculation units 12 and 14, the reference value calculation unit 21, and the utterance start time determination unit 22 are different from the first embodiment shown in FIG. 11, 13 and 1
Since 7 to 19 have the same functions as those of the same reference numerals in the first embodiment, the description will be simplified.

The calculation unit 12 calculates the first utterance start point likelihood a (t) calculated at the time t based on the prediction distribution stored in the prediction distribution storage unit 11 as shown in FIG. It is sent to the calculation unit 21. The calculation unit 14 calculates the second utterance start point likelihood b (t) calculated according to the time t based on the utterance detection feature parameter 13a from the acoustic analysis unit 13.
Is sent to the utterance start time determination unit 22.

The reference value calculation unit 21 _weights the first reference value _Refo by the first speech start point likelihood a (t),
A second reference value _Ref (t) that changes according to the time t is calculated and sent to the utterance start time determination unit 22. Equations (4) to (5) are given here as examples of weighting. However, in the formula (5), 0 <k ₂ .

[Equation 2] R _ef (t) = S / α (t) Equation (4) R _ef (t) = S−k ₂ · α (t) Equation (5)

The utterance start determination unit 22 weights the second utterance start point likelihood b (t) and the second reference value R _ef (t).
And b (t)> R _ef (t) for the first time, or b (t)> R _ef (t) continues for a certain period of time, then b (t)> R _ef (t) for the first time. The time point t) is determined as the utterance start time of the user, and the utterance start signal 22a indicating that is sent to the voice signal passing switch 17.

The switch 17 is turned on from the point of time when the utterance start signal 22a is given, and sends only the voice signal 17a sent while it is on to the acoustic analysis unit 18 as a voice recognition target. The acoustic analysis unit 18 calculates a characteristic parameter 18a suitable for voice recognition from the voice signal 17a that has passed through the voice signal passing switch 17, and sends it to the voice recognition unit 19. The voice recognition unit 19 performs voice recognition based on the characteristic parameter 18a and sends the recognition result 19a to the dialogue management device 1. The dialogue management device 1 determines the text 1a of the system announcement to be uttered next based on the recognition result 19a provided from the voice recognition unit 19 and sends the code name or the like to the announcement utterance device 2.

The switch 1 is also used in the description of the second embodiment.
Although only the signal 17a of the voice recognition target is given to the acoustic analysis unit 18 using the switch 7, it may be changed as follows without using the switch 17. (1) It is assumed that the voice signal 5a from the voice input circuit 5 is always sent to the acoustic analysis unit 18, and the utterance start time determination unit 22 sends the utterance start signal 22a to the acoustic analysis unit 18, which starts the utterance. The acoustic analysis is started from the time when the signal 22a is given. (2) Alternatively, the voice signal 5a from the voice input circuit 5 is constantly sent to the acoustic analysis unit 18, and the acoustic analysis unit 18 always performs acoustic analysis to determine the characteristic parameter 18a as the voice recognition unit 1.
9, and the utterance start time determination unit 22 further sends the utterance start signal 22a to the voice recognition unit 19. The voice recognition unit 19 performs the voice recognition using the characteristic parameter 18a from the time when the utterance start signal 22a is given. To start.

<Third Embodiment> As shown in FIG. 4, the voice dialogue apparatus according to the third embodiment comprises a dialogue management apparatus 1, an announcement utterance apparatus 2, and a voice recognition apparatus 30. Therefore, the voice output device 3 and the voice input device 5 may be built in the voice dialogue device as necessary, or may be appropriately connected as a separate object apart from the voice dialogue device. The speech recognition device 30 includes a prediction distribution storage unit 11, a first utterance start point likelihood calculation unit 12, and an utterance detection acoustic analysis unit 13.
The second utterance starting point likelihood calculating unit 14, the third utterance starting point likelihood calculating unit 15, the voice analysis acoustic analysis unit 18, the voice recognizing unit 31, and the voice recognition result determining unit 32.
Is provided.

Among the components of each device of the third embodiment, the arithmetic unit 15, the voice recognition unit 31, and the voice recognition result determination unit 32.
Is different from the first embodiment shown in FIG. 1, and the utterance start time determining unit 16 and the switch 17 in the first embodiment are not present, but the other ones 1, 2, 3, 5, 11 to 14 and 18 Has the same function as that of the same reference numeral in the first embodiment, the description will be simplified.

The calculation unit 15 uses the equations (1) to (3) described above.
By using the first utterance starting point likelihood a (t)
The utterance starting point like b (t) is weighted,
The third utterance start point likelihood α (t) is calculated according to the time t, and this is sent to the voice recognition result determination unit 32. It should be noted that the first utterance start point likelihood a (t) is calculated by the prediction distribution storage unit 11
2 is calculated by the calculating unit 12 calculating an expected value at which the user's utterance will start at time t, based on the predicted distribution shown in FIG. The second utterance start point likelihood b (t) is the characteristic parameter 13a for utterance detection, which is obtained by the acoustic analysis unit 13 constantly performing acoustic analysis.
Based on, the calculation unit 14 calculates the likelihood that the user's utterance will have started at time t. However, the announcement start signal 2a from the announcement vocalization device 2
Is given as time 0.

The voice-recognition acoustic analysis unit 18 constantly acoustically analyzes the voice signal 5a supplied from the voice input circuit 5 to successively calculate the voice-recognition characteristic parameters 18a, and sends the feature parameters 18a to the voice recognition unit 31.

The voice recognition unit 31 performs a plurality of voice recognitions by shifting the voice recognition start time one after another by assuming that the time t is the user's utterance start time at each time t of every 10 milliseconds, for example. Each voice recognition result w (t) when starting from each time t is sent to the voice recognition result determination unit 32, and the likelihood p (t) is calculated for each voice recognition result w (t) to calculate the voice recognition result. It is sent to the determination unit 32.

The voice recognition result judging section 32 uses the following equation (6), equation (7) or equation (8) to recognize each recognition start time t.
A value q (t) is calculated by integrating the likelihood p (t) of each speech recognition result and the third utterance start point likelihood α (t), and the time at which this value q (t) is maximized is calculated. find t _max ,
Of all the speech recognition results w (t), the speech recognition result w (t _max ) corresponding to the time t _max is determined as the recognition result 32a for the user's utterance. Only the voice recognition result 32a is sent to the dialogue management device 1. However, in equation (7),
For example, 0 <k ₃ <1. As a result, even if the user's utterance start time cannot be determined with high accuracy,
As a result, the speech of the user can be recognized with high accuracy.

[Number 3] q (t) = α (t ) + p (t) ... formula (6) q (t) = (1-k 3) · α (t) + k 3 · p (t) ... (7) q (t) = α (t) · p (t) (8)

In the dialogue management device 1, the text 1a of the system announcement to be uttered next is determined based on the recognition result 32a given from the voice recognition result judging section 32, and the code name or the like is sent to the announcement utterance device 2.

<Fourth Embodiment> As shown in FIG. 5, the voice dialogue apparatus according to the fourth embodiment comprises a dialogue management apparatus 1, an announcement utterance apparatus 2, and a voice recognition apparatus 40. Therefore, the voice output device 3 and the voice input device 5 may be built in the voice dialogue device as necessary, or may be appropriately connected as a separate object apart from the voice dialogue device. The speech recognition device 40 includes a prediction distribution storage unit 11, a first utterance start point likelihood calculation unit 12, and an utterance detection acoustic analysis unit 13.
A second utterance start point likelihood calculating unit 14, a third utterance start point likelihood calculating unit 15, a voice recognition acoustic analysis unit 18, a voice recognizing unit 41, and a transition probability updating unit 42. It is equipped with.

Among the components of the apparatus of the fourth embodiment, the arithmetic unit 15, the voice recognition unit 41 and the transition probability updating unit 42 are different from those of the first embodiment shown in FIG. 1, and the utterance in the first embodiment is different. Although the start time determination unit 16 and the switch 17 are not present, the other units 1, 2, 3, 5, 11 to 14 and 18 have the same functions as those of the same reference numerals in the first embodiment, and therefore the description will be simplified. .

The calculation section 15 uses the equations (1) to (3) described above.
By using the first utterance starting point likelihood a (t)
The utterance starting point like b (t) is weighted,
The third utterance starting point likelihood α (t) is calculated according to the time t, and this is sent to the transition probability updating unit 42. The first utterance starting point likelihood a (t) is based on the prediction distribution stored in the prediction distribution storage unit 11 as shown in FIG.
The calculation unit 12 calculates the expected value at which the user's utterance will start at time t. Further, the second utterance starting point likelihood b (t) would have started the utterance of the user at time t based on the utterance detection characteristic parameter 13a obtained by the acoustic analysis unit 13 performing acoustic analysis at all times. It is obtained by calculating the likelihood by the calculation unit 14. However, the time when the announcement start signal 2a is given from the announcement voicing device 2 is time 0.

The voice recognition acoustic analysis unit 18 constantly performs acoustic analysis on the voice signal 5a supplied from the voice input circuit 5 to successively calculate the voice recognition characteristic parameters 18a, and sends the feature parameters 18a to the voice recognition unit 41.

The voice recognition unit 41 always searches the sequence of the characteristic parameters 18a given from the acoustic analysis unit 18 for a probability-limited finite state network having a silent state 300 at the head as shown in FIG. The route from which the likelihood is obtained is output as the voice recognition result 41a and sent to the dialogue management device 1.

In general, a finite state network with probability is an HMM (Hidden Markov Model: Hidden Marv) of phonemes and words.
kov Model), and each state of the HMM holds the likelihood according to the feature parameter,
Alternatively, the probability distribution for calculating the likelihood according to the characteristic parameter is held.

When constructing this finite network with probability, a silent model 300 is provided at the beginning of a sentence, as shown in FIG. The silent model is a model corresponding to a section without voice, but it is possible to deal with such noise by using background noise or line noise during learning. In addition, by learning non-voices such as cough and breath, it is possible to prevent the non-voices from being mistaken as voices. It is also possible to separately learn noise and non-voice models and arrange them in parallel with the silence model 300. As a result, it becomes possible to always receive the input of the characteristic parameter 18a for voice recognition from the acoustic analysis unit 18 before the user starts speaking.

The transition probability updating unit 42 gives the probability of transition from the head silence model 300 of the finite state network with probabilities used in the speech recognition unit 41 to the sentence head state 303 to the third utterance starting point given from the arithmetic unit 15. Likeness α
Using (t), it is changed according to time t. That is, as shown in FIG. 6, the beginning silence model 300 has a transition 301 to a self-state and a transition 302 to a sentence start state, and each arc has an associated state transition probability. Therefore, at time t at which the third utterance start point likelihood α (t) is large, the state transition probability of transition to the sentence head state 303 is increased according to α (t). This facilitates the transition from the head silent model 300 to the sentence head state 303 at the user's utterance start time, and improves the accuracy of voice recognition.
In this case, at the time t when α (t) is large, the state transition probability of transitioning to the self-state 300 may be decreased at the same time according to α (t).

On the other hand, at time t when the third utterance start point likelihood α (t) is small, the state transition probability of transition to the sentence head state 303 is reduced according to α (t). As a result, before the user's utterance start time, it becomes difficult to transition from the head silent model 300 to the sentence head state 303, and it becomes difficult to perform erroneous voice recognition, so that the accuracy of voice recognition is improved. In this case, at time t when α (t) is small, the self-state 30
The state transition probability of transition to 0 may be increased according to α (t). Thus, by changing the state transition probability from the silent state 300 at the head to the sentence head state 303 by the third utterance start point likelihood α (t), a user's utterance start time is determined with high accuracy. Even if it is not possible, the user's utterance can be recognized with high accuracy as a result. Further, since the voice recognition is substantially performed once, the processing speed is increased and the memory capacity can be reduced as compared with the third embodiment.

The feature parameter 18a for voice recognition is used.
Is not optimal for detecting the start of speech, so the silent state 30
If the state transition probability from 0 to the sentence head state 303 is fixed, the accuracy of the transition from the head silent state 300 to the sentence head state 303 becomes low, and the accuracy of speech recognition deteriorates.

The dialogue management device 1 determines the text 1a of the system announcement to be uttered next based on the voice recognition result 41a given from the voice recognition unit 41, and sends the code name and the like to the announcement utterance device 2.

[0074]

According to the first, second and seventh to eleventh aspects of the invention, the utterance of the user is started from the prediction distribution prepared in advance, paying attention to the causal relationship between the system announcement and the utterance start time of the user. The expected value (likely as the first utterance starting point) that is likely to be calculated, and the likelihood that the utterance of the user obtained from the characteristic parameter for utterance detection is likely to start (likely as the second utterance starting point) Since the user's utterance start time is determined in combination with this, the utterance start time can be detected with one point and high accuracy, and therefore highly accurate voice recognition can be realized.

According to the third to fourth and twelfth to thirteenth inventions, the likelihood of the voice recognition result is also used to determine the utterance starting point by always performing the voice recognition. For example, it is possible to avoid an error such as meaningless utterance or determining a cough as the utterance starting point, and as a result, highly accurate voice recognition can be performed.

Particularly, according to the fifth and fourteenth inventions, there is a maximum point of the predicted distribution in the silent section of the system announcement, and according to the sixth and fifteenth inventions, the silent section is intentionally or positively provided. , The causal relationship between the system announcement and the utterance start time of the user is further clarified, and the utterance start time detection accuracy and the voice recognition accuracy are further improved.
Further, by intentionally or positively providing the silent section during the system announcement, the user can be controlled so that the user starts the utterance in the silent section, so that the use time of the voice interaction device can be shortened. . In other words, when confirming the voice recognition result in the dialogue, for example, "Is it OK in Yamamoto, otherwise answer No?" Compared to when making a system announcement, "Is it okay in Yamamoto (1 second silence) Yes or No? By answering `` Please answer '', the user who is accustomed to the device will speak in the silent section, and the system announcement after "Yes" will be unnecessary, so the time to listen to the system announcement will be reduced to less than half, It is possible to enhance the convenience for the user and efficiently use the device.
In addition, if necessary, it is possible to stop the system announcement when the utterance start time is determined so as not to disturb the user's utterance. Also, by setting the silent section, let the beginner hear a sufficient system announcement,
A spoken dialogue system that can be used by experienced users will be realized only by listening to a short system announcement. Furthermore, when the utterance start time can be determined with high accuracy, such a device capable of shortening the use time can work more effectively.

According to the sixteenth invention, since a dialogue using voice is performed between the user and the device under high-accuracy voice recognition, smooth dialogue can be realized.

[Brief description of drawings]

FIG. 1 is a block configuration diagram of a voice interaction device according to a first embodiment.

FIG. 2 is a diagram showing an example of a predicted distribution.

FIG. 3 is a block configuration diagram of a voice dialog device according to a second embodiment.

FIG. 4 is a block configuration diagram of a voice dialog device according to a third embodiment.

FIG. 5 is a block configuration diagram of a voice dialog device according to a fourth embodiment.

FIG. 6 is a diagram showing an example of a finite state network with probability having a silent state at the head.

FIG. 7 is a diagram showing a conventional example.

[Explanation of Codes] 1 Dialog management device 1a Text 2 Announcement speech device 2a, 5a Voice signal 2b Announcement start signal 2c Announcement end signal 3 Voice output circuit 3a System announcement 4 User's speech 5 Voice input circuit 10, 20, 30, 40 Speech recognition device 11 Prediction distribution storage unit 12 First utterance start point likelihood calculation unit 13 Speech detection acoustic analysis unit 13a Speech detection characteristic parameter 14 Second utterance start point likelihood calculation unit 15 Third utterance Start point likelihood calculation unit 16, 22 Speech start time determination unit 17 Voice signal passage switch for voice recognition 18 Acoustic analysis unit for voice recognition 18a Characteristic parameter for voice recognition 19, 31, 41 Voice recognition unit 19a, 32a, 41a Recognition result 21 Reference value calculation unit 32 Speech recognition result determination unit 42 Transition probability updating unit 100 Predicted distribution 200 Silence interval 300 Silence state 303 Sentence state a (t) First utterance start point likelihood b (t) Second utterance start point likelihood α (t) Third utterance start point likelihood p (t) speech recognition result likelihood R _ef reference value R _EFO first reference value R _ef (t) second reference value

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Seiichi Yamamoto 2-3-2 Nishishinjuku, Shinjuku-ku, Tokyo International Telegraph and Telephone Corporation

Claims (16)

[Claims] 1. A voice recognition method applied to a voice interaction device for interacting with a user using voice, wherein: a prediction distribution having a maximum point of a user's utterance start time with respect to a system announcement of the voice interaction device is prepared in advance. Well,
Based on this predicted distribution, calculate the expected value at which the user's utterance starts as the first utterance start point according to the time after the system announcement starts; perform acoustic analysis of the user's utterance converted into an electrical signal. Calculating a feature parameter for utterance detection based on the feature parameter, and calculating the likelihood that the user's utterance has started as the second utterance starting point likelihood according to the time based on the feature parameter; The likelihood of the start point is weighted by the likelihood of the first utterance start point, the likelihood of the third utterance start point obtained by this weighting is compared with a reference value, and the time point at which the likelihood becomes larger than the reference value is the utterance start time of the user. A process of performing a voice recognition based on this characteristic parameter by acoustically analyzing the user's utterance converted into an electric signal to calculate a characteristic parameter for voice recognition. Speech recognition method according to claim; be carried out according to the decision of the speech start time of the user. 2. A voice recognition method applied to a voice interaction device for interacting with a user by using a voice: A prediction distribution having a maximum point of a user's utterance start time for a system announcement of the voice interaction device is prepared in advance. Well,
Based on this predicted distribution, calculate the expected value at which the user's utterance starts as the first utterance start point according to the time after the system announcement starts; perform acoustic analysis of the user's utterance converted into an electrical signal. Calculating a feature parameter for utterance detection based on the feature parameter, and calculating a likelihood that the user's utterance has started as a second utterance starting point likelihood according to the time; the first utterance. The first reference value is weighted according to the starting point likelihood to calculate a second reference value that changes with time, and the second utterance starting point likelihood is compared with the second reference value to obtain the second reference value. The time when the value becomes larger than the value is determined to be the user's utterance start time; the user's utterance converted into an electric signal is acoustically analyzed to calculate a characteristic parameter for voice recognition, and based on this characteristic parameter Speech recognition method comprising; a process for performing voice recognition, be carried out as determined by the speech start time of the user. 3. A voice recognition method applied to a voice interaction device for interacting with a user by using a voice: A prediction distribution having a maximum point of a user's utterance start time with respect to a system announcement of the voice interaction device is prepared in advance. Well,
Based on this predicted distribution, calculate the expected value at which the user's utterance starts as the first utterance start point according to the time after the system announcement starts; perform acoustic analysis of the user's utterance converted into an electrical signal. Calculating a feature parameter for utterance detection based on the feature parameter, and calculating the likelihood that the user's utterance has started as the second utterance starting point likelihood according to the time based on the feature parameter; The likelihood of the start point is weighted by the likelihood of the first utterance start point to calculate the likelihood of the third utterance start point; a characteristic parameter for voice recognition by acoustically analyzing the utterance of the user converted into an electric signal. Based on this characteristic parameter, the recognition start time is shifted one after another to perform voice recognition, and the likelihood for each voice recognition result corresponding to each recognition start time is calculated. The sum or product of the likelihood of the speech recognition result for each recognition start time and the likelihood of the third utterance start point obtained by the weighting is calculated by combining the times, and the calculated start time is the maximum. A voice recognition result corresponding to the above is determined as a voice recognition result for a user's utterance; 4. A voice recognition method applied to a voice interaction device for interacting with a user by using a voice: A prediction distribution having a maximum point of a user's utterance start time for a system announcement of the voice interaction device is prepared in advance. Well,
Based on this predicted distribution, calculate the expected value at which the user's utterance starts as the first utterance start point according to the time after the system announcement starts; perform acoustic analysis of the user's utterance converted into an electrical signal. Calculating a feature parameter for utterance detection based on the feature parameter, and calculating the likelihood that the user's utterance has started as the second utterance starting point likelihood according to the time based on the feature parameter; The likelihood of the start point is weighted by the likelihood of the first utterance start point to calculate the likelihood of the third utterance start point; a characteristic parameter for voice recognition by acoustically analyzing the utterance of the user converted into an electric signal. And probing a finite-state network with a probability having a silent state at the head for speech recognition based on this characteristic parameter; Speech recognition method comprising; the probability of transition from the beginning of the silent state of the click to the top state of the sentence, be updated according to the time by using the third utterance start point likelihood of that obtained in the weighting. 5. The voice recognition method according to claim 1, wherein the maximum point of the predicted distribution of the utterance start times of the users exists in a silent section of the system announcement. Method. 6. The voice recognition method according to claim 5, wherein
The voice recognition method, wherein the silent section has a length of 0.2 seconds or more and 3 seconds or less and is present in at least one of a sentence of the system announcement and a sentence or a sentence. 7. A voice recognition device applied to a voice interaction device for interacting with a user using voice; storing a prediction distribution having a maximum point of a user's utterance start time with respect to a system announcement of the voice interaction device. First means; second means for calculating an expected value at which a user's utterance starts based on the stored predicted distribution as a first utterance start point in accordance with a time after the system announcement is started; and an electric signal Acoustic analysis of the user's utterance converted to
Third means for calculating a characteristic parameter for utterance detection; calculating a likelihood that a user's utterance has started based on the characteristic parameter for utterance detection as a second utterance start point according to time Fifth means for calculating the second utterance starting point likelihood by weighting the first utterance starting point likelihood, and calculating the weighted value as the third utterance starting point likelihood according to time. Means: A third utterance start point likelihood is compared with a reference value, and a sixth means for determining a time point when the utterance start point is greater than the reference value to be the utterance start time of the user; Acoustic analysis is performed to calculate a feature parameter for voice recognition, and voice recognition is performed based on the feature parameter according to a determination of the utterance start time of the user. Voice recognition device that. 8. A voice recognition device applied to a voice interaction device for interacting with a user by using a voice; storing a prediction distribution having a maximum point of a user's utterance start time with respect to a system announcement of the voice interaction device. First means; second means for calculating an expected value at which a user's utterance starts based on the stored predicted distribution as a first utterance start point in accordance with a time after the system announcement is started; and an electric signal Acoustic analysis of the user's utterance converted to
Third means for calculating a characteristic parameter for utterance detection; calculating a likelihood that a user's utterance has started based on the characteristic parameter for utterance detection as a second utterance start point according to time Fourth means for performing; weighting according to the likelihood of the first utterance starting point with respect to the first reference value,
Fifth means for calculating the value obtained by this weighting as the second reference value according to the time; and comparing the second utterance start point likelihood with the second reference value obtained by the weighting, and Sixth means for determining a time point when it becomes larger than a reference value as a user's utterance start time; acoustic analysis of the user's utterance converted into an electric signal to calculate a characteristic parameter for voice recognition, and the characteristic parameter Seventh, a process of performing voice recognition based on is determined according to the determination of the utterance start time of the user.
A speech recognition apparatus comprising: means. 9. The voice recognition device according to claim 7 or 8, wherein the seventh means is a switch for passing the user's utterance converted into an electric signal from a time point determined to be the user's utterance start time. Means; acoustic analysis means for voice recognition for acoustically analyzing the utterance of the user passing through the switch means to calculate a characteristic parameter for voice recognition; and a characteristic parameter for voice recognition calculated by the acoustic analysis means. A voice recognition means for performing voice recognition based on the voice recognition device; 10. The voice recognition apparatus according to claim 7 or 8, wherein the seventh means performs an acoustic analysis of the user's utterance converted into an electric signal from the time when it is determined to be the user's utterance start time. A voice recognition acoustic analysis unit that starts to calculate a voice recognition characteristic parameter; and a voice recognition unit that performs voice recognition based on the voice recognition characteristic parameter calculated by the acoustic analysis unit; A voice recognition device characterized by the above. 11. The voice recognition apparatus according to claim 7 or 8, wherein the seventh means is for voice recognition for acoustically analyzing a user's utterance converted into an electric signal to calculate a feature parameter for voice recognition. Acoustic analysis means; speech recognition means for performing speech recognition based on the characteristic parameters after the time point when it is determined that the speech start time of the user is among the characteristic parameters for speech recognition calculated by the acoustic analysis means; A voice recognition device comprising: 12. A voice recognition device applied to a voice interaction device for interacting with a user using voice; storing a prediction distribution having a maximum point of a user's utterance start time with respect to a system announcement of the voice interaction device. Predictive distribution storing means; first calculating means for calculating an expected value at which the user's utterance starts based on the stored predictive distribution as a first utterance start point according to the time after the system announcement is started. An utterance detecting voice analysis means for acoustically analyzing the user's utterance converted into an electric signal and calculating a utterance detecting characteristic parameter; and an utterance of the user started based on the utterance detecting characteristic parameter. Second computing means for calculating the likelihood as the second utterance starting point likelihood according to the time; the first utterance starting point with respect to the second utterance starting point likelihood Third computing means for performing weighting according to the likelihood, and calculating the weighted value as the third speech start point likelihood according to the time; acoustic analysis of the user's speech converted into the electric signal, and voice recognition Acoustic analysis means for voice recognition for calculating feature parameters for voice recognition; voice recognition is performed by sequentially shifting recognition start times based on the voice recognition feature parameters, and voice recognition corresponding to each recognition start time. A voice recognition means for calculating the likelihood for each result; a sum or product of the likelihood of the voice recognition result for each recognition start time and the likelihood of the third utterance start point is calculated according to the time, and this calculation is performed. A voice recognition device, comprising: a voice recognition result determining means for determining a voice recognition result corresponding to a recognition start time having a maximum value as a voice recognition result for a user's utterance. 13. A voice recognition device applied to a voice interaction device for interacting with a user using voice; storing a prediction distribution having a maximum point of a user's utterance start time with respect to a system announcement of the voice interaction device. Predictive distribution storing means; first calculating means for calculating an expected value at which the user's utterance starts based on the stored predictive distribution as a first utterance start point according to the time after the system announcement is started. Utterance detection acoustic analysis means for acoustically analyzing the utterance of the user converted into an electrical signal and calculating a utterance detection characteristic parameter; utterance of the user started based on the utterance detection characteristic parameter Second computing means for calculating the likelihood as the second utterance starting point likelihood according to the time; the first utterance starting point with respect to the second utterance starting point likelihood Third computing means for performing weighting according to the likelihood, and calculating the weighted value as the third speech start point likelihood according to the time; acoustic analysis of the user's speech converted into the electric signal, and voice recognition Acoustic analysis means for voice recognition for calculating a feature parameter for voice recognition; voice recognition means for performing voice recognition by searching a finite state network with probability having a silent state at the head based on the feature parameter for voice recognition; Transition probability updating means for updating the probability of transition from the silent state at the head of the probability-limited finite state network to the head state of the sentence according to the time using the third utterance start point likelihood. And a voice recognition device. 14. The voice recognition apparatus according to claim 7, wherein the maximum point of the predicted distribution of the utterance start time of the user exists in the silent section of the system announcement. apparatus. 15. The voice recognition apparatus according to claim 14, wherein the silent section has a length of 0.2 seconds or more and 3 seconds or less, and is between the sentences of the system announcement and between the phrases and the phrases. A voice recognition device characterized by being present in at least one of them. 16. The voice recognition device according to claim 7, wherein the designated text of the system announcement is converted into an electrical voice signal, and the start of the system announcement is notified to the voice recognition device. An announcement utterance device, and a dialogue management device for managing a dialogue with a user using voice by designating a text of a system announcement for the announcement utterance device and inputting a voice recognition result from the voice recognition device. Spoken dialogue device.