JP2003177788A - Audio interactive system and its method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve the recognition rate of vocabulary included in a user's speech and to enable recognition of a key word corresponding to a response with an indication word in a barge-in audio interactive system allowing the user to respond in the middle of a system speech. <P>SOLUTION: A scenario of the system speech is monitored, and weights to be added to key words included in individual parts of the scenario are determined on the basis of the time of the user's speech, the speech time of each scenario part, etc., and key words included in the user's speech are recognized. Further, a key word which the indication word means is recognized in accordance with the correspondence relation between the time of the user's speech with the indication word and a corresponding system speech. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice dialogue system that provides voice information to a user and performs processing based on a response from the user. The present invention relates to a voice dialogue system that can recognize the response content of a user even when responding.

[0002]

2. Description of the Related Art In a voice dialogue system, a system utters an option or the like, and a user responds by voice with a choice from the options. Then, the system converts this voice into a voice signal and then compares the voice signal with the voice information stored in the recognition dictionary to recognize which option the user has selected.

When this voice interactive system is used for program guides, the categories of programs are sequentially uttered, and the categories in which the user wants to receive guidance are uttered and designated. A concrete dialogue example is shown below.

System utterance: "Please select the service you want to use. 1 weather forecast, 2 sports news, 3 today's fortune." User utterance: "Ja, weather forecast".

As described above, in the method in which the system sequentially utters the options (the category name of the program in the above-mentioned dialog example) and then the user specifies, even if the user wants to be presented with the program of the category uttered first, Since the system has to wait until the utterance is finished and the user has to remember the option he / she wants to select until the system utterance is finished, the possibility of selecting the wrong option is increased. Also, for the system, until the utterance ends,
Since it is not possible to get the user's response, it will take time to move to the next step according to the scenario.
Here, in the above example, the “next scenario” is the presentation of choices such as baseball, soccer, and tennis that are more detailed choices of the “sports news” selected by the user, or the results of today's baseball game. It is the presentation of content such as.

In order to solve such a problem, there is known a barge-in dialogue system in which the user can respond even when the voice dialogue system is speaking. An example of a conversation in which a response can be made by this barge-in is shown.

(Example of dialogue with barge-in) System utterance: "Please select the service you want to use.

1 weather forecast, 2 sports news,
... "User utterance:" Sports news "or" 2 ".

In this barge-in voice interaction system, the user is allowed to respond even while the system is uttering, and the option can be responded immediately after the system speaks or is spoken by the system. There is no need to wait until the utterance ends, and it is less likely that the user will select an option by mistake, while the voice dialogue system can proceed to the next scenario. In this way, the barge-in voice dialogue system can be said to be a user-friendly voice dialogue system capable of proceeding with a smooth dialogue with the user.

However, in this barge-in dialogue system, since the user can speak while the system is speaking, a voice in which the system utterance is superimposed on the user's utterance is input to the system, and the user's utterance content is erroneously recognized. Or the recognition rate may decrease. Therefore, the barge-in spoken dialogue system employs a technique such as echo canceling that cancels the system utterance even if the system utterance is input to the system again. Furthermore, a recognition dictionary that weights the vocabulary that the user can predict to speak is used to facilitate recognition of the content of the speech.

[0011]

Even in the barge-in voice dialogue system equipped with the echo cancellation or weighted recognition dictionary described above, it is difficult to always correctly recognize the voice uttered by the user. Furthermore, in a conventional voice dialogue system, when the system utters a plurality of options, it is difficult for the user to recognize which option the user has selected, even if the user responds with a directive such as “that”. It was A specific example of the dialogue using this directive is shown below.

(Example of dialogue using directives) System utterance: "Please select the service you want to use.

1 weather forecast, 2 sports news,
・ ・ ・ ”User utterance:“ That ”.

Therefore, a first object of the present invention is to provide a voice interaction system and method capable of barge-in that recognizes a user's utterance more accurately.

A second object of the present invention is to provide a voice dialogue system and method that can recognize which option the user has selected, even if the user gives a response including a directive.

[0016]

In the following description, system utterances are referred to as system utterances, and user utterances are referred to as user utterances. An utterance that the user responds to during system utterance is called a barge-in utterance. The barge-in spoken dialogue system is characterized in that the user can respond even while the system is speaking based on the scenario.

In such a dialogue system, when the user utters a desired option, the user can predict that the option is often answered during or immediately after the utterance.

The present invention pays attention to such a characteristic of the user's response, and the time interval between the word included in the scenario during the system utterance or the continuous word and the user utterance in response to the word,
Furthermore, the vocabulary in the recognition dictionary is weighted according to the barge-in situation such as the strength of the user's utterance and the speed of utterance, and the recognition of which vocabulary in the recognition dictionary the user uttered It is characterized by improving the recognition of which option has been selected and instructed.

More specifically, in the invention of claim 1, a dialogue scenario section for storing a scenario defining a plurality of words to be uttered and an utterance order of the plurality of words, and a recognition dictionary storing vocabulary information. In a voice dialogue system connectable to a voice output unit for uttering voice according to the scenario of the dialogue scenario unit and a voice input unit for inputting voice uttered by a user, this voice output unit utters any part of the scenario. An output information management unit that detects whether
The output information management unit detects a scenario portion when the user speaks, and based on the detected scenario portion, the weighting calculation unit that weights the vocabulary information stored in the recognition dictionary and the voice input unit are input. A voice recognition unit that selects which of the vocabulary information stored in the recognition dictionary the voice uttered by the user is based on, from the voice information obtained by signal processing the processed voice and the weighted vocabulary information, The main point is a spoken dialogue system.

In this way, the system utterance corresponding to the time of the user utterance can be detected by the output information management unit, so that a word that is highly likely to be included in the user utterance is detected before the user utterance. The recognition rate is improved by increasing the weighting of the words spoken by this system and the continuous words, which are found by the correspondence with the words that have been uttered.

The above-mentioned word is, for example, "weather forecast".
May be treated as one word, or may be treated as two words “weather” and “forecast”.
Furthermore, in the case of "weather forecast", this "weather forecast" may be treated as one word.
"No", "forecast", "weather", and "forecast" may be treated as words.

Further, according to the invention of claim 1, since a voice dialogue system connectable to a system utterance scenario, a recognition dictionary, and a voice input unit and a voice output unit is constructed, the purpose of this voice dialogue system is as follows. It is possible to easily use a corresponding scenario, a recognition dictionary suitable for the scenario, and a voice input unit and a voice output unit suitable for the environment in which the voice dialogue system is used.

Further, in the invention of claim 2, a dialogue scenario section for storing a scenario defining a plurality of words to be uttered and an utterance order of the plurality of words, a recognition dictionary storing vocabulary information, and a user A voice input unit for inputting the spoken voice, a voice output unit for speaking the voice according to the scenario of the dialogue scenario unit, and an output information management unit for detecting which part of the scenario the voice output unit speaks. , A weighting calculation unit that weights vocabulary information stored in the recognition dictionary based on the scenario location detected by the output information management unit when the user speaks, and the voice input to the voice input unit is signal-processed. A voice for selecting which of the vocabulary information stored in the recognition dictionary is the voice uttered by the user, from the voice information and the weighted vocabulary information. The speech dialogue system and having a 識部 was gist.

Since the voice dialogue system is configured in this manner, the system utterance corresponding to the user's utterance can be detected by the output information management unit as in the case of the invention of claim 1, so that the system is included in the user utterance. Highly probable words or consecutive words are identified by the correspondence with the system uttered words or consecutive words before this user utterance, and the system uttered words or consecutive words are weighted higher. This improves the recognition rate.

Furthermore, the weighting described in claim 1 or 2 may be performed based on the strength and / or speed of the voice uttered by the user.

With this configuration, the vocabulary in the recognition dictionary can be weighted according to the situation in which the user speaks.

Furthermore, the weighting according to claim 1 or 2 may be performed based on the utterance time of the user and the scheduled utterance time of the word included in the scenario.

With this configuration, the word predicted to be included in the user's utterance can be predicted more accurately,
Appropriate weighting of vocabulary in the recognition dictionary can be performed.

Furthermore, the weighting according to claim 1 or 2 is configured to be performed based on the user's utterance time and the scheduled utterance time of the word included in the scenario, and the weighting calculation unit determines the user's utterance time. Recognition based on the difference between the words included in the scenario and the scheduled time of speech Recognition of the weighting coefficient or weighting ratio stored in the weighting coefficient management unit that stores the weighting coefficient or the weighting ratio added to the words included in the dictionary A voice dialogue system (called a voice dialogue system A) characterized by weighting the vocabulary information stored in the dictionary may be the gist.

With this configuration, the time interval between the system uttered word already stored in the recognition dictionary and the word to be system uttered and the user uttered time is obtained, and the time interval is set to this time interval. Based on this, the vocabulary information stored in the recognition dictionary can be weighted.

Furthermore, the weighting according to claim 1 or 2 is configured to be performed based on the utterance time of the user and the scheduled utterance time of the words included in the scenario, and the weighting calculation unit determines the utterance time of the user. A weighting coefficient or a weighting coefficient stored in a weight table management unit that stores a table showing a weighting coefficient or a weighting ratio to be added to the words included in the recognition dictionary according to the difference between the utterance scheduled time of the words included in the scenario Spoken dialogue system characterized by weighting the vocabulary information stored in the recognition dictionary based on the ratio (referred to as speech dialogue system B)
May be the gist.

Since the weighting coefficient or the weighting ratio to be added to the words included in the recognition dictionary is stored in the table as described above, the weighting process can be easily performed.

Further, in the above-mentioned voice interaction system A or voice interaction system B, the weighting coefficient or the weighting ratio stored in the weighting coefficient management unit or the weight table management unit and the selection result selected by the voice recognition unit. It has a history management unit that stores history information indicating correspondence,
A voice dialogue system may be configured in which the weighting coefficient or the weighting ratio is changed based on the history information.

With this configuration, since the weighting coefficient or the weighting ratio can be changed based on the history information based on the responses of a large number of users, weighting with a better recognition rate can be performed.

Further, in the invention of claim 3, a dialogue scenario part storing a plurality of words to be uttered and a scenario defining the utterance order of the plurality of words, a recognition dictionary storing vocabulary information, and a dialogue scenario part. An output that detects which part of the scenario the voice output unit is speaking in a voice interaction system that can be connected to a voice output unit that speaks a voice according to a scenario and a voice input unit that inputs a voice uttered by a user Either the information management unit or the vocabulary information stored in the recognition dictionary for the voice uttered by the user from the voice information obtained by signal processing the voice input to the voice input unit and the vocabulary information stored in the recognition dictionary. Select whether or not, and if the selected vocabulary information is a vocabulary, the vocabulary is uttered from the voice output unit at the time when the vocabulary is uttered. The speech dialogue system and having a voice recognition unit for association with the word words or continuous contained in the scenario portion was gist.

With this configuration, even if the user utters an instruction word, there is a high possibility that the instruction word indicates the user's utterance time, or a word included in the system uttered scenario before the user utterance time. Therefore, it is possible to perform barge-in using the directive. In addition, here the directive is
For example, "this", "that", "that", "these",
Includes words such as "these."

Furthermore, in the invention of claim 4, a dialogue scenario part for storing a plurality of words to be uttered and a scenario defining the utterance order of the plurality of words, a recognition dictionary storing vocabulary information, and a dialogue scenario part. In a voice dialogue system connectable to a voice output unit for uttering voice according to a scenario and a voice input unit for inputting voice uttered by a user, a voice output for storing waveform data corresponding to a word uttered by the voice output unit One of the vocabulary information stored in the recognition dictionary, which is the voice uttered by the user, from the data storage unit, the voice information obtained by signal processing the voice input to the voice input unit, and the vocabulary information stored in the recognition dictionary. Is selected, and if the selected vocabulary information corresponds to a vocabulary, the vocabulary is uttered from the voice output unit corresponding to the utterance of this vocabulary. Re was summarized as speech dialogue system, characterized in that it comprises a voice recognition unit for association with the waveform data stored in the audio output data storage unit.

With this configuration, since the waveform data corresponding to the word included in the system utterance at the time of the user's utterance or immediately before the user's utterance is stored in the voice output data storage unit, the user uses the instruction word as the system. Even if the user responds to the utterance, the word included in the system utterance designated by the directional word or the continuous word can be recognized.

Further, in the invention of claim 5, the step of uttering based on the scenario, the step of inputting the utterance of the user, the step of detecting the part of the scenario corresponding to the utterance of the user, and the step of uttering the user Correlate the step of weighting the words or consecutive words contained in the scenario, and the correspondence between the input user utterance and the weighted or consecutive words,
A speech dialogue method of a speech dialogue system including a step of selecting either a word or a continuous word and a step of performing next processing based on a scenario based on the selected word or the continuous word.

Since the voice interaction method is configured as described above, the words included in the scenario or the continuous words can be weighted according to the time of the user's utterance, so that the recognition rate of the voice interaction system can be improved.

As described above, when the user utterance during the system utterance (hereinafter, such a user utterance is referred to as a barge-in utterance) is performed, the weight of the recognition vocabulary is determined based on the time information between the barge-in utterance and the system utterance. A change is obtained and the weight of the word information stored in the recognition dictionary is changed. Since the user's utterance is recognized based on this changed weighted recognition dictionary, even if the time of occurrence of the barge-in utterance and the timing of the system utterance based on the scenario are different, the recognition dictionary with appropriate weight is added. Since the user utterance can be recognized by using, the recognition rate of the user utterance is improved.

Further, the vocabulary information corresponding to the demonstrative word such as "that" is stored in the recognition dictionary as the recognition vocabulary,
When the barge-in utterance includes an utterance indicating an option by the directive, the output information management unit uses the vocabulary information corresponding to the option immediately before the utterance of the directive,
By using the vocabulary as the recognition result, it is possible to select the option by the vocabulary.

Furthermore, the waveform data of the system utterance is saved in the voice output data storage unit, and when a barge-in utterance is made by the directive, the system uttered word immediately before the barge-in utterance can be detected, and that word is designated as the directive word. Recognize as the option selected by.

[0044]

DESCRIPTION OF THE PREFERRED EMBODIMENTS (First Embodiment) A first embodiment will be described with reference to FIGS.

First, an outline of the voice dialogue system 10 will be described with reference to FIG. 1 showing an example of the configuration of the first embodiment of the present invention. The dialogue control unit 14 is a ROM (Read Only M) that stores a microcomputer, a control program, and the like.
memory), a RAM (R) for storing processed data, etc.
and voice access system 1 including other storage devices, processing circuits, and the like.
The main control of 0 is performed. The dialogue scenario unit 12 connected to the dialogue control unit 14 is connected to the dialogue scenario unit 12 that stores a scenario describing the system utterance of the voice dialogue system 10, and the system utterance to the user is generated based on this scenario. Output information management unit 16
Is output from the audio output unit 18 via. The output information management unit 16 monitors which part of the scenario is the voice output of the system output from the voice output unit 18.

In response to the system utterance, the user of the voice dialogue system 10 utters the user's voice, and the uttered voice is input to the voice input unit 20 and undergoes signal processing such as A / D conversion to control the dialogue. It is sent to the unit 14, and in the present embodiment, the dialogue control unit 14 further receives signal processing such as feature extraction to obtain voice information, which is sent to the voice recognition unit 21. In the voice recognition unit 21, the voice information is compared with the vocabulary information stored in the recognition dictionary 22 weighted by the weight calculation unit 24 by a known method such as pattern matching, and corresponds to which vocabulary. Is recognized.

In FIG. 1, the voice dialogue system basic portion 11 enclosed by a chain line is the main portion of the present invention, and the dialogue scenario portion 1 according to the purpose of use of the voice dialogue system 10.
2. By using the recognition dictionary 22, the voice output unit 18, and the voice input unit 20, it is possible to configure a voice dialogue system more suited to the purpose.

Next, the recognition of the user's utterance when the user makes a barge-in utterance, which is a feature of the present invention, will be described in detail.

FIG. 2 shows an example of the dialogue scenario 30 stored in the dialogue scenario section 12 and the utterance timing when the scenario part is voice-outputted when the dialogue scenario 30 is voice-output as a system utterance. FIG. 4 is a diagram showing a system utterance 40 and a user utterance 45 indicating a timing at which the user uttered to select an option corresponding to the system utterance 40.

In the example of this dialogue scenario 30, "Please select the service you want to use. 1 weather forecast, 2
Sports news, fortunes today. , A part of the scenario, and before and after a keyword that is an option, for example, before and after “1 weather forecast”, a delimiter 36 “/” is added to indicate that the keyword is a keyword.

This dialogue scenario 30 includes the dialogue control unit 14
Then, the output information management unit 16 detects which part of the scenario has been uttered by the system, and the voice output unit 18 utters the system utterance 40.

In the example of FIG. 2, "Please select .." 31 and "1 weather forecast" that constitute the dialogue scenario 30.
32, "2 sports news" 33, "3 today's fortune" 34, respectively, correspond to the time sections 41, 42, 43, and 44, as indicated by the arrows.
It indicates that the system is spoken. Then, in the time section 43 of the system utterance 40, "2 sports news"
Indicates that the user has barged in the user utterance 47 “sports news” starting from the user utterance time point 46.

When the user responds to the system utterance 40 shown in FIG. 2 with a barge-in as shown by the user utterance 45, the voice interactive system 10 recognizes the processing flow for recognizing the option selected by the user. 3 and FIG. 4 showing an example of correspondence between the vocabulary of the recognition dictionary and the weight of the vocabulary, FIG. 5 showing an example of the recognition dictionary in which the weight is changed, and FIG. 6 showing an example of change of the weight with barge-in time. .

FIG. 3 is a diagram showing an example of the processing flow of the voice dialogue system 10 of the present invention. As shown in FIG. 2, when the user responds with a barge-in utterance, the timing of the barge-in corresponds. Output scenario part Information management unit 1
Detected in 6, guess the word etc. contained in the user's response,
The weighting of the inferred word or the like stored in the recognition dictionary 22 is changed to perform the process of improving the recognition rate.

The process starts at step 100. Each keyword of the scenario stored in the dialogue scenario unit 12 is described by being separated by the delimiter 36 “/” (see FIG. 2), and the keywords are sequentially provided to the output information management unit 16 via the dialogue control unit 14. Is transmitted, the number of the delimiter 36 is counted.
Therefore, the system utterance from the voice output unit 18 is executed while the output information management unit 16 monitors the number of the delimiter symbol corresponding to the keyword (step 10).
2). In this initial state, the vocabulary of the recognition dictionary 22 is stored with the same weight (1.0 in this embodiment) added to each stored vocabulary, as shown in FIG.

When the user starts barge-in utterance to select one of the options during the above-mentioned system utterance, the barged-in user's voice is input to the voice input unit 20 and the barge-in start signal is input via the dialogue control unit 14. Is transmitted to the output information management unit 16 (step 104).

Based on this barge-in start signal, the output information management unit 16 stops the system utterance and further detects which part of the scenario the system uttered. For example, as shown in FIG. 2, the user utterance 47 “Sports news” is displayed at the barge-in utterance start time point 46 indicated by the user in the system utterance portion 43 indicated by the system utterance of “sports news” in the scenario portion 33. If you do a barge-in with this user, the barge-in for this user is
The occurrence between the delimiters 36-2 and 36-3 of the dialogue scenario 30 is detected by the output information management unit 16. Here, where in the dialogue scenario the barge-in is performed is referred to as a barge-in attribute. When the barge-in occurs, the output information management unit 16 transmits the barge-in attribute to the weight calculation unit 24 via the dialogue control unit 14 (step 106 in FIG. 3).

In the weight calculation section 24, the keyword part included in each of the scenario parts before and after the barge-in based on the received barge-in attribute and each scenario part before and after the barge-in is detected based on the dialogue scenario 30. For each of the detected keywords, for example, the keyword included in the barged-in scenario part is given the highest weight, and the scenario part uttered by the system before the barge-in is a scenario part close to the barge-in scenario part. The keywords included in are weighted more. Then, the keywords included in the scenario portion that was scheduled to be system uttered after the barge-in time are weighted smaller than the weights added to the system uttered keywords (step 108 in FIG. 3).

Specific weighting is shown in FIG. 5 as an example of weights associated with the respective keywords when barging in at the user utterance time point 46 in FIG. In this Figure 5,
The vocabulary 61 is a keyword in the scenario, and a weight 62 is associated with each keyword. As shown in FIG. 5, the largest weight 1.5 is given to the keyword “sports news” included in the system utterance 43 corresponding to the user utterance time point 46 when the user barges in.
However, the keyword "weather forecast" that has already been uttered by the system is given a weight of 1.2, and the keyword "today's fortune" that has not been uttered by the system is given a lowest weight of 0.8.

As described above, each weight for the keyword group calculated by the weight calculating section 24 is added to the vocabulary corresponding to each keyword stored in the recognition dictionary 22, and the vocabulary weight in the recognition dictionary 22 is updated. (Step 11 in FIG. 3)
0).

By this step 110, the vocabulary that may be included in the user's utterance, and further included in the user's utterance among this vocabulary, may be included depending on the situation in which the user virginized (in this case, the virgin time for the dialogue scenario). Greater weight was given to the likely vocabulary. The recognition dictionary 22 with the updated weighting is used to perform voice recognition of the user's utterance (step 112 in FIG. 3). Here, for the voice recognition, a well-known dynamic programming matching method, a method using a hidden Markov model, a method based on the characteristics of the characteristic parameters of the speech included in the user's utterance, or the like can be used.

Here, in order to explain this speech recognition in a simple manner, the characteristic parameter U extracted from the speech of the user's speech and the characteristic parameters K1, K2, ... Of each keyword (vocabulary) stored in advance. , Kn is compared, and the keyword having the smallest | U−Kj | is set as the keyword uttered by the user. Here, assuming that the above-mentioned weight is W, in the present embodiment, the keyword uttered by the user is recognized to have the smallest value of | U−Kj | / W.

Next, the dialogue control unit 14 determines whether or not the dialogue scenario recorded in the dialogue scenario unit 12 has a dialogue scenario to be provided to the user next, and if there is a dialogue scenario to be provided subsequently. (Fig. 3, step 11
If YES in 4), based on the keyword selected by the user, that is, the recognized result, the process proceeds to the process of uttering the next dialogue scenario (step 116 in FIG. 3).

On the other hand, if there is no next dialogue scenario (NO in step 114 of FIG. 3), processing corresponding to the selected option, for example, if "sports news" has been selected, sports Content such as news is handed over to a process presented to the user from, for example, the audio output unit 18 via a content presentation device (not shown), and this process flow ends (step 11 in FIG. 3).
8).

In the processing flow shown in FIG. 3, the response from the user is confirmed, for example, after the user's response, the word "sports news is desired" corresponding to the recognized word. The system utterance processing with may be inserted. When such confirmation processing is performed, it is preferable to insert a confirmation processing step after step 112 in the processing flow of FIG.

In this confirmation step, the determination process based on the user's response is also performed, and if the confirmation result is YES,
In the case of NO in step 114, it is preferable to return to step 102 and continue the processing after uttering a comment such as "I will repeat again, so please select the desired item".

Further, "weather forecast", "sports news", "today's fortune", etc. are respectively registered as the vocabulary in the recognition dictionary of FIG. 5 or FIG. It may be registered in more detail by dividing into “weather”, “forecast”, “sports”, “news”, “today”, “fortune” and the like. Even in the case of fine division, it is preferable to add the same weight to, for example, “weather” and “forecast”.

When the system utters the system utterances with the numbers in parallel, such as "1 weather forecast",
In the same way as above, "1" is also registered, weather ",
It is preferable to add the same weight to the “forecast”.

As can be seen from the above description of the processing flow, the weighting added to each keyword shown in FIG. 5 changes depending on the time when the user barges in. FIG. 6 shows how the weight assigned to each keyword changes. In FIG. 6, reference numerals 72, 73, and 74 denote time information when barge-in is performed, for example, time information 7 indicated by a symbol.
2 is a system utterance point 4 in the system utterance 40 of FIG.
It is shown that it was barge-in in 2nd. Barging in at this timing is when barging in during system utterance with "weather forecast". The keyword "weather forecast" has the largest weight of 1.5, and system utterance is made after this keyword. A small weight of 0.8 is added to the keywords "sports news", "fortune of today", etc. Further, in the case of the time information 74, since there was a barge-in when the system utterance "Today's fortune" was uttered, the keyword "Today's fortune" was added with the largest weight of 1.4, and In the spoken keywords "weather forecast" and "sports news",
A small weight of 1.0 is added. Here, time information 7
The weight in the case of 4 is set to a smaller weight for a keyword having a long time elapsed from the time of barge-in, such as "weather forecast" 0.9, "sports news" 1.0, "today's fortune" 1.5. It may be added.

Regarding the above-described processing when performing voice recognition, the keyword included in the user's utterance is estimated based on the time information at the time of barge-in, and the relationship between the estimated keyword and the scenario information corresponding to the time information. Weight is added based on this. In addition to weighting by the time information at the time of barge-in, the following weighting in consideration of the status of barge-in is also effective for more accurately recognizing the keyword spoken by the user. A recognition method considering the situation of the barge-in will be described with reference to FIGS. 7 and 8. FIG. 7 is a diagram showing weighting based on the speed of the user utterance, and FIG. 8 is a diagram showing weighting based on the strength of the user utterance.

In FIG. 7, time waveform 8 of barge-in utterance
The barge-in utterance duration T is calculated from 0. The utterance time To stored in advance for the keyword included in the dialogue scenario portion corresponding to the time of the barge-in utterance
And T, the coefficient is obtained from the correspondence 82 between the speed ratio and the weighting coefficient in accordance with the speed ratio R81, and the initial weight value 83 is multiplied by this coefficient to obtain the changed weight 84. . These series of processes are performed by the dialogue control unit 14 or the weight calculation unit 2
4 may be performed, or both may be shared.

Further, the correspondence 82 between the speed ratio and the weighting coefficient is stored as a table in either the dialogue control unit 14, the weighting calculation unit 24 or the recognition dictionary 22.

In place of the above-mentioned utterance speed, in FIG. 8, when the acoustic power of the user utterance is used, the spectrum 91 of the barge-in utterance obtained by frequency analysis is obtained from the time waveform 80 of the barge-in utterance. From the spectrum 91, the power P of the barge-in speech is obtained. Then, with respect to the keyword included in the dialogue scenario portion corresponding to this barge-in utterance time, the pre-stored ratio of power Po and P is obtained, and the correspondence between the power ratio and the weighting coefficient is obtained according to this power ratio 92. The coefficient is obtained from 93, and the initial value 94 of the weight is multiplied by this coefficient to obtain the changed weight 95. These series of processes may be processed by the dialogue control unit 14 or the weight calculation unit 24, or may be shared by both. Further, the correspondence 93 between the power ratio and the weight coefficient is stored as a table in either the dialogue control unit 14, the weight calculation unit 24, or the recognition dictionary 22.

(Second Embodiment) FIG. 9 shows a second embodiment of the present invention.
The voice dialogue system 200 will be described with reference to FIG. In FIG. 9 showing the second embodiment, components having the same functions as those of the components in FIG. 1 showing the configuration example of the first embodiment are designated by the same reference numerals. Unlike the voice dialogue system 10 of the first embodiment, in the voice dialogue system 200, the weight calculation unit 2
The weighting factor management unit 201 is connected to 4-2, and the weighting factor added to the vocabulary stored in the recognition dictionary 22 by the weighting calculation unit 24-2 is stored in the weighting factor management unit 201. .

More specifically, when the response to the system utterance is uttered by the user, which part of the dialogue scenario is uttered by the system corresponding to the user utterance time is output information management unit as in the first embodiment. Detected by 16. The keyword included in the system utterance at the time of this barge-in has the highest weight, and the keyword included in the system utterance at the time of the barge-in has a large time lapse with the barge-in time. The smaller the weight of the keyword included in the keyword, and the smaller the weight of the keyword included in the unspoken system utterance. Here, the ratio of each weight or each weight is stored as a table in the weight coefficient management unit 201.

FIG. 10 shows an example of a weighting coefficient table stored in the weighting coefficient management unit 201 as a table.
In the example shown in FIG. 10, the keyword included in the scenario portion at the time of barge-in is given the highest weight of 1.8, and the keyword included in the scenario portion immediately before the keyword has a weight of 1.5 and further 1. The keyword included in the previous scenario part was set to 1.4. That is, with respect to the keywords included in the scenario portion that has already been uttered by the system, the smaller weight is associated with the keyword included in the scenario portion whose time elapses from the time of barge-in. Also,
A small weight of 0.8 is associated with the keyword included in the scenario portion that was supposed to be uttered after the barge-in time.

Thus, with the barge-in time as the base point,
The weighting factor management unit 201 stores the weight to be added to the keyword estimated to be uttered by the user based on the elapsed time and whether or not the system uttered, detects the barge-in time, and detects the barge-in time and the scenario uttered portion. And the process of adding the corresponding weight to the keyword are performed by the weight calculation unit 24-2.

Since the weight table is constructed in this way,
The weights added to the keywords are easily associated with each other, and the processing can be performed quickly.

(Third Embodiment) Next, a third embodiment will be described with reference to FIG. 11 showing a configuration example of the third embodiment of the present invention. This FIG. 11 shows the voice dialogue system 22 of the third embodiment.
0 shows the outline of 0. Also in this figure, components having the same functions as those of the first and second embodiments are designated by the same reference numerals, and the description thereof will be omitted.

The feature of the third embodiment is that a dialogue history management unit 224 is provided, and the dialogue history management unit 224 stores the correct ratio of the result of recognition of the user utterance by the voice dialogue system 220 as a history. The weight given to the keyword is updated based on this correctness rate.

The user utterance is barged in according to the system utterance, and the weight table management unit 222 is added to the keyword included in the scenario portion corresponding to the barge-in time.
The weight calculation unit 24-3 gives the weights stored in the table.

The weight coefficient table stored in the weight table management section 222 may be the same as the table stored in the weight coefficient management section 201 of the second embodiment.

Next, a process for updating the weight stored in the weight table management unit 222 based on the dialogue history will be described below. When the user makes a barge-in utterance, the output information management unit 16 detects the scenario part corresponding to the barge-in time, and detects the keywords included in the scenario part, the scenario parts before and after the scenario part, and the like. Then, as in the case of the second embodiment, the keywords included in the scenario portion that was uttered at the time of barge-in includes the highest weight in the scenario portion that has already been uttered by the system. With respect to keywords, the smaller weight is added to the keywords included in the scenario portion whose time elapses up to the time of barge-in. An example of the management table stored in the dialogue history management unit 224 by associating the keywords with the weights in this way is shown in FIG.

The table shown in FIG. 12 shows a case where the dialogue scenario includes five keywords, a, b, c, d, and e, and the scenario portion corresponding to the time of barge-in includes the keyword c. Indicates the case.

The keyword c included in the scenario portion at the time of barge-in has a weight of 1.8, the keyword b included in the scenario portion immediately before that has a weight of 1.5, and the keyword a has a weight of 1.4. , The longer the time elapsed until the time of barge-in is, the smaller weight is added to the keyword.
In this example, a weight of 0.8 is added to the keywords d and c that were supposed to be uttered after the barge-in time. Then, such a weight is added to each keyword, and the voice recognition unit 21-3 performs voice recognition by comparing these keywords with the voice of the user utterance input from the voice input unit 20. When this speech recognition is performed on a large number of users, the correctness rate 228, which is the ratio of correctly recognizing the weighted keywords, is updated as a dialogue history for each recognition process, and the dialogue history management unit 22
It is stored in 4.

By this comparison, a keyword having a large weight may not always be recognized as a keyword included in the user's utterance. For example, in the system utterance 40 shown in FIG. 2, “weather forecast”, “sports news”, and “fortune of today” are sequentially uttered, and as shown in FIG. Even if there is a barge-in utterance, the user's barge-in utterance may deviate from the system utterance if the user is confused about the choice or is an elderly user. The valid rate of the keyword b in FIG. 12 is 7% because it is the voice dialogue system 220.
It is shown that the user's barge-in timing and the system utterance may be slightly deviated from the user's response in many times.

In the diagram showing an example of the management table of the dialog history management unit shown in FIG. 13, as in FIG. 12, in the case of a dialog scenario of five keywords (five choices), an initial weight 234 is added. The method is also similar. However, the correct rate showing the result of a large number of user responses is shown in FIG.
In this example, the correct answer rate 236 is higher for the keyword b immediately before the keyword b even when the keyword included in the scenario portion at the time of barge-in is given the highest weight. .

For example, if the keyword b is a relatively short word, the system utters only "famous place" for the purpose of, for example, sight-seeing, or if it is a confusing word, for example, "scenic spot" for the purpose of guiding a scenic place. This may occur when a word is used, or the system utterance is difficult to hear due to ambient noise and the barge-in timing is shifted, or when the above-mentioned user is lost in selecting an option.

As described above, the initial weight and the correct rate are shown in FIG.
If it does not correspond to, the weight is changed and updated as indicated by the changed weight 238.

In this way, in the voice dialogue system 220 of this embodiment, the weight added to the keyword can be changed to a better weight based on the dialogue history, so that the voice recognition rate can be improved.

(Fourth Embodiment) Referring to FIG. 14 showing an example of the configuration of the fourth embodiment of the present invention and FIG. 15 showing an example of the processing flow of a response including a directive, the user utterance is “that”. The voice interaction system 240 in the case of barging in the response including the directives such as

Also in this embodiment, the same reference numerals are given to the components having the same functions as in the first, second and third embodiments,
The description is omitted.

The main difference between the fourth embodiment and each of the above-described embodiments is that even if a user's uttered voice including an instruction word such as "that" is input from the voice input unit 20, which option is selected by the instruction word. The feature is that it can be recognized whether it is selected. Here, the instruction word is uttered by the user, and the voice input unit 20
When input to the dialogue control unit 14-4 via the, the feature extraction of this directive is performed, the feature is compared with the standard pattern in the voice recognition unit 21-4, and the directive is input. Is detected. The standard pattern information of these various directives is stored in the recognition dictionary 22 in this embodiment.

Next, a detailed processing flow of the voice interaction system 240 when an option is selected by the user's utterance including this directive is described below.

When the voice interactive system 240 becomes operable, the process starts in step 300 of FIG. Also in this embodiment, the system utterance will be described assuming that the dialogue scenario shown in FIG. 2 is used.

Similar to the first embodiment, the dialogue scenario section 12
The dialogue scenario stored in the dialogue control unit 14-4.
When the keywords are sequentially transmitted to the output information management unit 16-4 via the, the number of the delimiter 36 shown in FIG. 2 is counted. Therefore, the system utterance from the voice output unit 18 is the order of the delimiter 36.
The keyword corresponding to is executed while being monitored by the output information management unit 16-4 (step 302 in FIG. 15).

During the above system utterance, at the user utterance time point 46 in FIG. 2, when the user utters a barge-in utterance indicating "it" to select one of the options, the barge-in user's voice is input by voice input. It is input to the unit 20, feature extraction is performed by the dialogue control unit 14-4, and the extracted feature information is input to the voice recognition unit 21-4 and is a standard of each directive stored in the recognition dictionary 22 in advance. The pattern is compared. When the voice recognition unit 21-4 recognizes that the user utterance is an instruction word, a barge-in signal based on this recognition is transmitted to the selection instruction word management unit 242, and the selection instruction word management unit 242 uses the instruction word. An instruction word barge-in signal indicating that the barge-in has been performed is output via the dialogue control unit 14-4 to the output information management unit 16-4.
(Step 304 in FIG. 15).

Based on this directive word barge-in signal, the output information management unit 16-4 stops the system utterance and further detects which part of the scenario the system uttered.
For example, as shown in FIG. 2, the user utters "that" at the barge-in utterance start time point 46 indicated by the user in the system utterance portion 43 indicated by the system utterance of "sports news" in the scenario point 33. Since the barge-in was performed with the directive, the fact that the barge-in of this user occurred between the delimiters 36-2 and 36-3 of the dialogue scenario 30 (barge-in attribute) is the output information management unit 16-.
4 (step 306 in FIG. 15).

Since the detected barge-in attribute includes the keyword "sports news", the output information management unit 16-4 determines that the keyword selected by the user with the instruction word is "sports news". The determination is made (308 in FIG. 15), and the keyword information corresponding to this keyword “sports news” is set as the dialogue control unit 14-4.
Via the selected instruction word management unit 242.

The selection instruction word management unit 242 determines that the instruction word "that" indicates "sports news" based on the keyword information corresponding to the transmitted keyword "sports news" (see FIG. 15). Step 31
0), the result of recognition that the keyword selected by the user is “sports news” is transmitted to the dialogue control unit 14-4.

Next, the dialogue control unit 14-4 determines whether or not the dialogue scenario recorded in the dialogue scenario unit 12 includes a dialogue scenario to be provided to the user next, and the dialogue scenario to be provided subsequently. If there is any (YES in step 312 in FIG. 15), the process moves to the utterance of the next dialogue scenario based on the keyword selected by the user, that is, the recognized result (step 314 in FIG. 15).

On the other hand, if there is no next dialogue scenario (NO in step 315 of FIG. 15), processing corresponding to the selected option, for example, if “sports news” has been selected, sports The content such as news is delivered to the process presented to the user from, for example, the audio output unit 18 via the content presentation device (not shown), and the process flow ends (step 31).
6).

As described above, even when the option is selected by barge-in with the directive word, it is recognized whether or not the user's utterance is the directive word, and when it is the directive word, the time when the barge-in is detected is detected. Since it is configured to detect the keywords included in the scenario part corresponding to this barge-in time,
Even if the user's utterance is an instruction word, the user's choice can be recognized.

(Fifth Embodiment) With reference to FIG. 16 showing a configuration example of a fifth embodiment of the present invention and FIG. 17 showing another example of the processing flow of a response including a directive, the user utterance is “that. "
The voice interaction system 250 in the case of barging in the response including the directives such as

Also in the present embodiment, the constituents having the same functions as those in the first, second, third, and fourth embodiments are designated by the same reference numerals, and the description thereof will be omitted.

The main feature of the present embodiment is that, with regard to the construction of the dialogue scenario, the dialogue system can be used without using a delimiter or the like for distinguishing the choices. Can be used as a dialogue scenario. That is, in the first to fourth embodiments, the dialogue scenario specifies the scenario part with the delimiter "/" for specifying the scenario part. However, in the present embodiment,
It is possible to allow a user to recognize a keyword that indicates a choice from a uttered user uttered in a barge without including information for specifying a scenario part in the scenario.

The outline of the characteristic points which are different from the present embodiment and the above-mentioned embodiments will be mainly described. In the voice dialogue system 250 of FIG. 16, the dialogue scenario section 12-5 of the present embodiment records and stores the dialogue scenario as shown in FIG. 2, for example. The dialogue scenario section 12-5 records the dialogue scenario. The recorded dialogue scenario is system-uttered from the voice output unit 18-5 via the dialogue control unit 14-5, and the uttered voice waveform data is sequentially stored in the voice output data storage unit 252. go.

Next, when the voice dialogue system 250 having such characteristics has a plurality of choices uttered by the system utterance, the user makes a response to the selection of one of the choices by the instruction word. The flow will be described.

When the voice interactive system 250 becomes operable, the process starts at step 400 in FIG. Also in this embodiment, the contents of the system utterance are 3 such as "weather forecast", "sports news", and "fortune of today" shown in FIG.
Described as using a dialogue scenario in which one of the two options is selected.

In this embodiment, the dialogue scenarios stored in the dialogue scenario section 12-5 are uttered by the system from the voice output section 18-5 via the dialogue control section 14-5, and are sequentially uttered. The voice waveform data of the system utterance is stored in the voice output data storage unit 252 (step 402 in FIG. 17).

During the above-mentioned system utterance, at the user utterance time point 46 in FIG. 2, when the user barge-in utters an instruction word "that" to select one of the options, the barge-in user's voice is input by voice input. The dialogue control unit 14-5 inputs the feature information to the unit 20, and the extracted feature information is input to the voice recognition unit 21-5 and is a standard for each directive stored in the recognition dictionary 22 in advance. The pattern is compared. When the voice recognition unit 21-5 recognizes that the user utterance is an instruction word, a barge-in signal based on this recognition is transmitted to the selection instruction word management unit 242-5, and the selection instruction word management unit 242-5. From the instruction word barge-in signal indicating that the barge-in is performed by the instruction word from the dialogue control unit 14-5, and the voice output unit 1
The system utterance from 8-5 is stopped (step 406 in FIG. 17).

Based on this directive word barge-in signal, the dialogue control unit 14-5 reads the voice waveform data within the predetermined time interval before the barge-in time recorded in the voice output data storage unit 252, and the voice recognition unit 21- Send to 5. The voice recognition unit 21-5 performs voice recognition processing on the voice waveform data and the voice waveform data stored in the recognition dictionary 22 by a method such as pattern comparison. (Fig. 1
7 step 408).

Here, a specific example will be described with reference to FIG.
If the user performs a barge-in with the user utterance "that" at the barge-in utterance start point 46 indicated by a ▼ mark in the system utterance part 43 shown in System utterance of "sports news" of the scenario point 33 The voice recognition unit 21-5 recognizes that the user has selected "sports news" by comparing the voice waveform data (step 408 in FIG. 17). Then, the recognition result “sports news” is selected by the selection word management unit 242-5.
(Step 410 in FIG. 17).

Then, the selection instruction word management unit 242-5 transmits the recognition result "sports news" to the dialogue control unit 14-5, and the dialogue control unit 14-5 determines whether or not there is a next dialogue scenario.
If it is determined in -5 and there is a next scenario (YES in step 412 of FIG. 17), the process proceeds to step 414, and the next dialogue scenario is performed based on the recognition result.

On the other hand, if there is no next dialogue scenario (NO in step 412 of FIG. 17), a process corresponding to the selected option, for example, if "sports news" has been selected, sports Content such as news is delivered to the processing presented to the user from, for example, the audio output unit 18-5 via the content presentation device (not shown), and the processing flow ends (step 416 in FIG. 17).

(Sixth Embodiment) A sixth embodiment will be described with reference to FIG. 18 showing an outline of a portable telephone to which the present invention is applied. The mobile phone 500 of the sixth embodiment has a function of communicating with a service provider, a service provider (for example, a public facility such as a public office), a product sales company, etc., and receiving a desired service by a voice response from a user. It is a thing. A functional unit corresponding to the voice interaction system basic unit 11 shown in FIG. 1 is housed in the main body 510 of the mobile phone 500, and the speaker 520 as a voice output unit, the microphone 530 as a voice input unit, the keyboard 530 and the display panel. 54
0 is arranged on the front surface of the mobile phone 500. Functions corresponding to the dialogue scenario unit 12 and the recognition dictionary 22 in FIG. 1 are installed on the information providing side such as the service provider via the antenna 550.

In this way, the mobile phone 500 which is a communication terminal
Since the present invention is applied to the above, there is an effect that the user can enjoy a desired service by voice conversation at any time.

Note that, like the mobile phone 500, the PDA
(Personal Digital Assistant
nt) may be equipped with the voice dialogue system.

(Supplementary Note 1) A dialogue scenario part for storing a plurality of words to be uttered and a scenario defining the utterance order of the plurality of words, a recognition dictionary for storing vocabulary information, and a voice utterance according to the scenario of the dialogue scenario part. In a voice dialogue system connectable to a voice output unit for inputting and a voice input unit for inputting a voice uttered by a user, an output information management unit for detecting which part of the scenario the voice output unit is speaking. And a weighting calculation unit for weighting the vocabulary information stored in the recognition dictionary based on the scenario location detected by the output information management unit when the user speaks, and the voice input unit. A voice uttered by the user is stored in the recognition dictionary based on the voice information obtained by signal processing the voice and the weighted vocabulary information. Speech dialogue system, characterized in that it comprises a voice recognition unit for selecting which of the lexical information.

(Supplementary Note 2) A dialogue scenario section that stores a plurality of words to be uttered and a scenario that defines the utterance order of the plurality of words, a recognition dictionary that stores vocabulary information, and a voice uttered by a user are input. A voice input unit, a voice output unit that speaks a voice according to the scenario of the dialogue scenario unit, an output information management unit that detects which part of the scenario the voice output unit is speaking, and the user Based on the scenario location detected by the output information management unit at the time of utterance, a weighting calculation unit for weighting the vocabulary information stored in the recognition dictionary, and signal processing of the voice input to the voice input unit. From the voice information and the weighted vocabulary information, it is selected whether the voice uttered by the user is the vocabulary information stored in the recognition dictionary. Speech dialogue system, characterized in that it comprises a voice recognition unit for, a.

(Supplementary Note 3) The voice interaction system according to Supplementary Note 1 or Supplementary Note 2, wherein the weighting is performed based on the strength and / or speed of the voice uttered by the user.

(Supplementary note 4) The voice dialogue system according to Supplementary note 1 or Supplementary note 2, wherein the weighting is performed based on the speech time of the user and the scheduled speech time of the word included in the scenario.

(Supplementary Note 5) The weighting calculation unit adds the weighting coefficient or the weighting factor to the word included in the recognition dictionary based on the difference between the utterance time of the user and the utterance scheduled time of the word included in the scenario. The vocabulary information stored in the recognition dictionary is weighted on the basis of the weighting coefficient or the weighting ratio stored in the weighting coefficient management unit that stores the ratio. system.

(Supplementary Note 6) The weighting calculation unit adds the weighting coefficient to the word included in the recognition dictionary according to the difference between the utterance time of the user and the scheduled utterance time of the word included in the scenario, or Note 4 is characterized in that the vocabulary information stored in the recognition dictionary is weighted based on the weighting coefficient or the weighting ratio stored in a weight table management unit that stores a table representing the weighting ratio. Spoken dialogue system described in.

(Supplementary Note 7) In the voice dialogue system according to Supplementary Note 5 or Supplementary Note 6, the weighting coefficient or the weighting ratio stored in the weighting coefficient management unit or the weight table management unit and the voice recognition unit are selected. A voice interaction system, comprising: a history management unit that stores history information indicating a correspondence with the selected result, and changing the weighting coefficient or the weighting ratio based on the history information.

(Supplementary Note 8) A dialogue scenario part for storing a plurality of words to be uttered and a scenario defining the utterance order of the plurality of words, a recognition dictionary storing vocabulary information, and a voice utterance according to the scenario of the dialogue scenario part. In a voice dialogue system connectable to a voice output unit for inputting and a voice input unit for inputting a voice uttered by a user, an output information management unit for detecting which part of the scenario the voice output unit is speaking. Of the vocabulary information stored in the recognition dictionary, the voice uttered by the user from the voice information obtained by signal processing the voice input to the voice input unit and the vocabulary information stored in the recognition dictionary. Is selected, and when the selected vocabulary information is a reference word, the reference word corresponds to the time when the reference word is uttered, and Speech dialogue system, characterized in that it comprises a voice recognition unit for association with words that the word or the continuous contained in the scenario locations uttered from the force unit.

(Supplementary Note 9) A dialogue scenario part for storing a plurality of words to be uttered and a scenario defining the utterance order of the plurality of words, a recognition dictionary storing vocabulary information, and a voice utterance according to the scenario of the dialogue scenario part. A voice output system and a voice input system for inputting a voice uttered by a user, in the voice dialogue system, a voice output data storage unit for storing waveform data corresponding to a word uttered by the voice output unit, The voice uttered by the user from the voice information obtained by signal-processing the voice input to the voice input unit and the vocabulary information stored in the recognition dictionary is any of the vocabulary information stored in the recognition dictionary. Is selected, and if the selected vocabulary information corresponds to a reference word, the reference word is output from the voice output unit corresponding to the utterance of the reference word. A voice recognition unit for association with the waveform data stored in the audio output data storage unit is,
A spoken dialogue system comprising:

(Supplementary Note 10) A step of uttering based on a scenario, a step of inputting a utterance of a user, a step of detecting a part of the scenario corresponding to the utterance of the user, and a step of correlating with the utterance of the user A step of weighting a word included in a part of a scenario or a continuous word, and a correspondence between the input user's utterance and the weighted word or a continuous word, and either the word or the continuous word; And a step of performing the following processing based on the scenario based on the selected word or continuous words, a voice interaction method for a voice interaction system.

(Supplementary Note 11) A dialogue scenario section that stores a plurality of words to be uttered and a scenario that defines the utterance order of the plurality of words and a recognition dictionary that stores vocabulary information can be connected via a base station. A communication terminal device, wherein a voice input unit for inputting a voice uttered by a user, a voice output unit for uttering voice according to the scenario of the dialogue scenario unit, and a voice output unit for uttering any part of the scenario An output information management unit that detects whether or not the vocabulary information is stored, and a weighting calculation unit that weights the vocabulary information stored in the recognition dictionary based on the scenario location detected by the output information management unit when the user speaks. And the voice uttered by the user is recognized from the voice information obtained by signal-processing the voice input to the voice input unit and the weighted vocabulary information. Speech dialogue system, characterized in that it comprises a voice recognition unit for selecting which of the lexical information stored in the book, the.

[0130]

EFFECT OF THE INVENTION By monitoring the system utterance scenario and detecting the location of the system utterance at the time when the user utters, the keyword included in the user utterance responding during the system utterance is predicted, and the user utterance Since the keyword predicted from the time relationship between the time point and the corresponding system utterance is weighted to recognize the keyword being uttered by the user, the keyword being uttered by the user can be recognized more accurately. Furthermore, even in the response of the user by the directive, the keyword corresponding to the directive can be accurately recognized from the association with the scenario.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration example of a first embodiment of the present invention.

FIG. 2 is a diagram showing a relationship between a dialogue scenario, a system utterance, and a user utterance.

FIG. 3 is a diagram showing an example of a processing flow of a voice dialogue system.

FIG. 4 is a diagram showing an example of correspondence between a vocabulary of a recognition dictionary and a weight of the vocabulary.

FIG. 5 is a diagram showing an example of a recognition dictionary in which weights are changed.

FIG. 6 is a diagram showing an example of a weight table that changes weights over time.

FIG. 7 is a diagram showing weighting based on the speed of user speech.

FIG. 8 is a diagram showing weighting based on the strength of a user's utterance.

FIG. 9 is a diagram showing a configuration example of a second embodiment of the present invention.

FIG. 10 is a diagram showing an example of a weighting coefficient table.

FIG. 11 is a diagram showing a configuration example of a third embodiment of the present invention.

FIG. 12 is a diagram showing an example of a management table of a dialogue history management unit.

FIG. 13 is a diagram showing an example of a management table of a dialogue history management unit.

FIG. 14 is a diagram showing a configuration example of a fourth embodiment of the present invention.

FIG. 15 is a diagram showing an example of a processing flow of a response including a directive.

FIG. 16 is a diagram showing a configuration example of a fifth embodiment of the present invention.

FIG. 17 is a diagram showing another example of a processing flow of a response including a directive.

FIG. 18 is a diagram showing an outline of a mobile phone to which the present invention has been applied.

[Explanation of symbols]

10 Spoken dialogue system 11 Spoken dialogue system basics 12 Dialog scenario section 14 Dialog control unit 16 Output information management section 18 Audio output section 20 Voice input section 21 Speech recognition unit 22 Recognition dictionary 24 Weight calculator 30 Dialogue scenario 40 system utterance 45 User utterance 200 Spoken dialogue system 220 Spoken dialogue system 222 Weight table management unit 224 Dialog History Management Department 240 Spoken dialogue system 242 Selection Directive Management Section 250 Spoken dialogue system 252 voice output data storage 500 mobile phones

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁷ Identification code FI Theme Coat (reference) G10L 3/00 R (72) Inventor Tatsuro Matsumoto 4-1, 1-1 Ueodachu, Nakahara-ku, Kawasaki-shi, Kanagawa Fujitsu Incorporated (72) Inventor Shigeru Yamada 4-1-1 Kamiodanaka, Nakahara-ku, Kawasaki-shi, Kanagawa F-term (reference) within Fujitsu Limited 5D015 HH05 HH13 LL06 LL10 5D045 AB04 AB30

Claims (5)

[Claims] 1. A dialogue scenario section for storing a plurality of words to be uttered and a scenario defining a utterance order of the plurality of words, a recognition dictionary storing vocabulary information, and a voice for uttering a voice according to the scenario of the dialogue scenario section. In a voice dialogue system connectable to an output unit and a voice input unit for inputting a voice uttered by a user, an output information management unit that detects which part of the scenario the voice output unit is speaking, A weighting calculation unit for weighting the vocabulary information stored in the recognition dictionary based on a scenario location detected by the output information management unit when the user speaks; and a voice input to the voice input unit. The vocabulary information in which the speech uttered by the user is stored in the recognition dictionary from the signal-processed voice information and the weighted vocabulary information. Speech dialogue system and having a speech recognition unit for selecting which of. 2. A dialogue scenario section for storing a plurality of words to be uttered and a scenario defining a utterance order of the plurality of words, a recognition dictionary for storing vocabulary information, and a voice for inputting a voice uttered by a user. An input unit, a voice output unit that speaks a voice according to the scenario of the dialogue scenario unit, an output information management unit that detects which part of the scenario the voice output unit is speaking, and the user speaks At this time, based on the scenario location detected by the output information management unit, a weighting calculation unit for weighting the vocabulary information stored in the recognition dictionary, and voice information obtained by signal processing the voice input to the voice input unit. And which of the vocabulary information stored in the recognition dictionary the voice uttered by the user is selected from and the weighted vocabulary information. Speech dialogue system and having a, a voice recognition unit. 3. A dialogue scenario section for storing a scenario defining a plurality of words to be spoken and a speech order of the plurality of words, a recognition dictionary storing vocabulary information, and a voice for uttering a voice according to the scenario of the dialogue scenario section. In a voice dialogue system connectable to an output unit and a voice input unit for inputting a voice uttered by a user, an output information management unit that detects which part of the scenario the voice output unit is speaking, The voice uttered by the user from the voice information obtained by signal-processing the voice input to the voice input unit and the vocabulary information stored in the recognition dictionary is any of the vocabulary information stored in the recognition dictionary. And the selected vocabulary information is a directive word, the directive word corresponds to the time when the directive word is uttered by the voice output unit. Speech dialogue system, characterized in that it comprises a voice recognition unit for association with words that the word or the continuous contained in the scenario points that have been spoken, the. 4. A dialogue scenario section for storing a plurality of words to be uttered and a scenario defining a utterance order of the plurality of words, a recognition dictionary storing vocabulary information, and a voice for uttering a voice in accordance with the scenario of the dialogue scenario section. In a voice interaction system connectable to an output unit and a voice input unit for inputting a voice uttered by a user, a voice output data storage unit for storing waveform data corresponding to a word uttered by the voice output unit; Whether the voice uttered by the user is the vocabulary information stored in the recognition dictionary from the voice information obtained by signal-processing the voice input to the input unit and the vocabulary information stored in the recognition dictionary. If the selected vocabulary information corresponds to a vocabulary, the vocabulary is not uttered from the voice output unit corresponding to the utterance of the vocabulary. Speech dialogue system, characterized in that it comprises a voice recognition unit for association with the waveform data stored in the audio output data storage unit. 5. A utterance based on a scenario, a step of receiving a utterance of a user, a step of detecting a portion of the scenario corresponding to the utterance of the user, and a step of correlating the utterance of the user with the utterance of the user. A step of weighting a word included in the place of or a continuous word, and the correspondence between the input user's utterance and the weighted word or a continuous word, and either the word or the continuous word A voice interaction method for a voice interaction system, comprising: a step of selecting; and a step of performing the following processing based on the scenario based on the selected word or continuous words.