JP2003058188A - Voice interaction system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice interaction system with which a recognition rate in voice recognition is improved. SOLUTION: This voice interaction system is provided with a voice recognizing engine for recognizing the voice of a user, an interaction processing engine for preparing a speech to the user corresponding the recognized result, a voice synthesizing engine for synthesizing the prepared speech to the voice, a voice recognition dictionary for storing dictionary data containing voice patterns for voice recognition and a voice input/output means. Further, this voice recognizing engine is configured to search the last user voice recognized result in the voice recognition dictionary to extract dictionary data related to the searched recognized result and to set the data into its own storage area. Therefore, when recognizing the voice of the user the next time, the dictionary data related to the last recognized result are used so that the recognition rate can be improved.

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 for automatically operating various equipment such as a car navigation system mounted on a vehicle and an audio equipment based on a dialogue with a user.

[0002]

2. Description of the Related Art In a complicated operation such as a car navigation system and an audio device, a voice dialogue system is used to operate the device. In this system,
The system utters a question created based on a predetermined step, recognizes the user's answer to this utterance by voice, and creates a new utterance based on the recognition result. Information is obtained from the user.

[0003]

However, the user's utterance has various utterance speeds, sound qualities, intonations, and the like. Therefore, in voice recognition by comparing and collating voice patterns, a recognition error often occurs. When such an error occurs, the user has to utter to correct the error and then repeat the same utterance, which significantly reduces the progress speed of the dialogue. Therefore, if a recognition error occurs frequently, the operability of the device is significantly reduced.

The present invention has been made in view of the above points of the conventional voice dialogue system, and an object of the present invention is to provide a voice dialogue system having a high recognition rate and therefore good operability.

[0005]

In order to solve the above problems, according to a first aspect of the present invention, a voice recognition engine for recognizing a user's utterance and a utterance to the user according to the recognition result. A dialogue processing engine for creating, a voice synthesizing engine for synthesizing the created utterance into voice, a voice recognition dictionary for storing dictionary data including a voice pattern for the voice recognition, and a voice input / output unit. The voice recognition engine searches the recognition result of the previous user's utterance in the voice recognition dictionary, extracts dictionary data related to the searched recognition result, and sets it in the voice recognition engine to set the voice of the next user's utterance. A spoken dialogue system is provided that provides recognition.

In this speech dialogue system, the speech recognition engine searches the speech recognition dictionary for the previous recognition result, extracts only the dictionary data related to the searched recognition result from the speech recognition dictionary, and extracts it from the speech recognition dictionary. Set as a dictionary in the storage area. The next voice recognition is performed using the dictionary thus set in the voice recognition engine. Therefore, it is possible to prevent a recognition error that a recognition result not related to the previous recognition result is obtained. In addition, since the dictionary data for recognition is significantly narrowed down, the recognition speed is improved, which in turn improves the response speed of the dialogue.

In the voice system of the above aspect, when there are a plurality of recognition results of the previous user's utterance, the voice recognition engine searches the voice recognition dictionary for any of them.
Dictionary data related to any recognition result is set in the voice recognition engine.

As a result, the data incorporated in the voice recognition engine as a dictionary is further narrowed down, the occurrence rate of recognition errors is reduced, and the response speed of the dialogue is improved.

In the speech system of the above aspect, the speech recognition engine has a list of word attributes to be searched for recognition results. Therefore, when a word not in this list is recognized in the previous voice recognition, the word is excluded and only the word in the list is searched in the dictionary. As a result, even if a word that does not need to be a search target is included in the user's utterance, it can be dealt with.

In the voice system according to the above aspect, the voice recognition engine stores the word attribute corresponding to the list in the recognition result of the previous user's utterance, if the word attribute exists. Thus, when an error occurs in the recognition of the next user's utterance, the stored word attributes can be used to search the dictionary and extract the related dictionary data. As a result, the recognition error can be dealt with promptly.

According to the second aspect of the present invention, a voice recognition engine for recognizing a user's utterance, a dialogue processing engine for creating a utterance to the user according to the recognition result, and the created utterance as a voice. A voice synthesis engine for synthesizing, a voice recognition dictionary for storing dictionary data including voice patterns for voice recognition, and a voice input / output unit are provided, and the voice recognition dictionary has a plurality of types of dictionaries. However, the speech recognition engine has a correspondence table of a dialogue state and a dictionary of a type used in the dialogue state, and in recognizing a user's utterance in each dialogue state, a dictionary of a corresponding type according to the correspondence table is selected from the speech recognition dictionary. Extract and use,
A spoken dialogue system is provided.

Further, in this voice dialogue system, the voice recognition engine extracts a plurality of types of dictionaries from the voice recognition dictionary according to the correspondence table and uses them.

According to this speech dialogue system, the speech recognition engine extracts only the required type of dictionary from the speech recognition dictionary according to the correspondence table preset according to the dialogue state, and extracts it from its own storage area. Set to. Therefore, in speech recognition, since only the required dictionary is subject to recognition work, the load on the speech recognition engine is reduced, and as a result, the processing time of speech recognition is shortened and the response of the speech dialogue is improved. To be done.

[0014]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the basic configuration of a voice dialogue system 10 according to an embodiment of the present invention. This system is a system for providing a voice guidance function to a car navigation system, and includes a display 12 as a video output device, a speaker 14 as a voice output device, a microphone 16 as a voice input device, and a voice dialogue start switch. 18 and a processing device 20.

The processing device 20 has, as hardware,
As is well known, it comprises a central processing unit (CPU), a main storage device, and the like. In addition, the processing device 20, as software that can be run on the main storage device, as shown in FIG. 1, an image display engine 202 that performs a process of displaying an image on the display 12, and a pattern of a sound input from the microphone 16 are generated. A voice recognition engine 204 that performs a process for recognizing, and a voice synthesis engine 2 that performs a process for electronically synthesizing a voice to be output from the speaker 14.
It is equipped with 06.

Further, as shown in FIG. 1, the processor 20 includes, as software, an interactive processing engine 210 for controlling the above-described image display engine, voice recognition engine and voice synthesis engine in response to an instruction from the application program 208. I have it. Further, the processing device 20
Includes a dialogue database 212 referred to by the speech recognition engine 204, the speech synthesis engine 206, and the dialogue processing dedicated engine 210, and a speech recognition dictionary 214 mainly referred to by the speech recognition engine 204.

Of course, an operating system (OS) 216 that comprehensively manages and controls hardware and software also runs in the processing device 20. In addition, the interactive database 212 and the voice recognition dictionary 214 described above may of course be configured on an external memory such as a CD-ROM, a DVD-ROM, or a memory card without being incorporated in the processing device 20. .

As shown in FIG. 2, the voice recognition engine 204 includes a processing program (not shown) for voice recognition,
It has a dictionary 204a for basic terms used for operating the car navigation system, and a RAM area 204b for extracting and setting a predetermined portion from the voice recognition dictionary 214. The basic term dictionary 204a includes basic terms necessary for dialogue, such as “yes”, “no”, “return”, “go”, prefecture names, and frequently used big city names. Facility names such as department stores, restaurants, etc. are stored together with their voice patterns.
The basic term dictionary 204a is always set in the RAM area 204b, and the word can be recognized by voice. It is also possible to remove the prefecture name and the metropolitan city name from the basic term dictionary 204a and treat them as the highest hierarchy of the dictionary concerning place names.

The voice recognition engine recognizes a search target word attribute list, a region 204c in which a table in which a dialog state and a dictionary type used for the dialog are stored, and word attributes in the search target word attribute list. A storage area 204d or the like for storing the storage area when the storage area is stored is described in detail in the section of description of each embodiment.

As shown in FIG. 3, the voice recognition dictionary 214 stores an address dictionary, a facility name dictionary, and the like. In the present invention, as shown in FIG. 4, the address dictionary has a hierarchical structure in which words of attributes having conceptually upper and lower relations are arranged in a hierarchical manner, and each word in the upper hierarchy is used as a node. Has a structure in which words related to the lower hierarchy are branched, that is, a tree structure.

In the illustrated example, the highest hierarchy is "prefecture name".
Is used as an attribute, the second level is made up of words having "city name" as an attribute, and the third level is made up of words having "ward name" as an attribute. Furthermore, the prefecture names in the highest hierarchy are used as nodes, and the city names in each prefecture are branched. Similarly, each word belonging to the hierarchy of the city name is used as a node, and the ward name is branched. As a result, an image of a tree-shaped three-layer grammar as shown in FIG. 4 is constructed.

With respect to the voice recognition dictionary 214 which stores the address recognition dictionary having such a grammatical structure, the voice recognition engine 204 refers to the result of the voice recognition by the previous user's utterance, and refers to The grammar is extracted from the dictionary 214 and set in its own RAM area 204b. The voice recognition engine 204 recognizes the next user utterance using the dictionary set in this way. Therefore, in this voice recognition, data other than the word attribute data after the node defined by the previous recognition result is not set in the voice recognition engine 204. For example, even if a city name in Hyogo prefecture is uttered, it is set in Osaka prefecture. There is no situation in which it is mistakenly recognized as the city name. Therefore, in speech recognition, the recognition rate is improved.

The operation of various embodiments of the voice dialogue system having the above basic structure will be described below in detail with reference to flowcharts.

FIG. 5 is a flow chart for explaining the operation of the voice dialogue system according to the first embodiment of the present invention. Further, FIG. 6 is an example of a dialogue between the voice dialogue system and the user for explaining the flowchart (dialogue example 1).
Indicates.

When the system is driven and set, the dialogue engine 210 operates to create the first question "Where do you want to go?" For setting the destination, and this question is voiced by the voice synthesis engine 206. And is output from the speaker 14. When the user answers "Hyogo prefecture" to this question, the voice recognition engine 204 recognizes the user's utterance by referring to the basic dictionary 204a of its own (step S1 in FIG. 5).

Next, the voice recognition engine 204 converts the word attribute "Hyogo", which is the first recognition result, into the voice recognition dictionary 2.
04 in the address dictionary (step S2), and the dictionary data after the searched word attribute is extracted and set in the RAM area 204b of the voice recognition engine 204 (step S).
3). Specifically, the dictionary data below the node 10 in FIG. 4 is set in the area 204b.

The system then asks "where in Hyogo Prefecture?" In response to this, when the user answers "Kobe City", the voice recognition engine 204 recognizes this using the dictionary data set in the RAM area 204b (step S4). When it is confirmed in the next step S5 that the dialogue has not ended, the voice recognition engine 204 returns to step S2 again, and searches the dictionary of FIG. In S3, the dictionary data of the node 20 and below is set in the RAM area 204b.

The system then asks "Where is Kobe?" And the user replies "Nada Ward." This user utterance is recognized using the dictionary data of the node 20 and below set in the RAM area 204b. The setting of the destination is completed by repeating these operations until the dialogue is finished, that is, the utterance of the destination is finished.

In the above embodiment, the RAM area 2 of the voice recognition engine 204 is used for voice recognition of "Kobe City".
Since only the dictionary of each city name under Hyogo prefecture is set in 04b, there is no mistakenly recognizing "Kobe city" as the city name of other prefectures. Similarly, even when recognizing "Nada Ward", only the ward name in Kobe city is set in the RAM area 204b, so that there is no possibility of erroneously recognizing this as a ward name in another city.

Therefore, between the system and the user, for example, the system "where are you going?", The user "Osaka Prefecture", the system "Where is Osaka Prefecture", the user "Sakai City", the system "Where is Akashi City?" The occurrence of an unnatural conversation such as (a user utterance "Sakai city" is mistakenly recognized as "Akashi city") is prevented. Further, since the words to be recognized are narrowed down in each recognition, the recognition speed is significantly improved.

FIG. 7 is a flow chart for explaining the operation of the voice dialogue system according to the second embodiment of the present invention, and FIG. 8 shows a dialogue example (dialogue example 2) used for the explanation of this flow chart.

In this embodiment, the system has a configuration capable of responding to the question "Where do you want to go?" On the system side when the user answers with two words, "Kobe City, Hyogo Prefecture". There is.

First, the user's first utterance "Kobe, Hyogo" is recognized using, for example, the basic dictionary 204a (step S10). Next, the voice recognition engine 204 recognizes the word attribute “Hyogo” of the recognition result as the recognition dictionary 2
14 is searched (step S11), and a dictionary below the searched word attribute, that is, a dictionary below the node 10 in FIG. 4 is extracted (step S12). Next, it is determined whether or not there is an unsearched recognition result (step S13). If there is an unsearched recognition result, the process returns to step S11 and a search is performed in the dictionary for the unsearched recognition result. In the case of FIG. 8, since an unsearched recognition result “Kobe City” exists, this is searched from the dictionary of nodes 10 and below, and the dictionary of nodes 20 and below is extracted.

This operation is repeated until there is no unsearched recognition result, and when there is no recognition result, the dictionary data finally extracted in step S14 is used as the voice recognition engine 20.
4 RAM area 204b. After that, the next user utterance "Nada Ward" is recognized using the set dictionary data (step S15).

By repeating the above operation until the end of the dialogue (step S16), the setting of the destination in the system is completed.

In this embodiment, since the voice recognition engine narrows down the dictionary to be repeatedly set for each word attribute with respect to the utterance of a plurality of word attributes by the user, as in the case of the first embodiment, the destination Occurrence of erroneous recognition upon voice input can be prevented, and a high response speed can be obtained.

FIG. 9 is a diagram for explaining the third embodiment of the present invention, and specifically, the voice recognition engine 20.
6 shows a search target word attribute list set in advance in FIG.
In this embodiment, for example, in response to a user's utterance such as “go to Osaka prefecture”, “Osaka” is detected by the first voice recognition,
This is to deal with the case where a recognition result consisting of three word attributes "ni" and "go" is obtained.

In this embodiment, each recognized word attribute is first searched for in the search target word attribute list. In the illustrated example, since only "Osaka Prefecture" exists in the list, the process shown in the flowchart of FIG. 5 is performed for the word attribute "Osaka Prefecture".
As a result, it is possible to prevent the result having the word attribute other than the municipality of Osaka from being recognized by the next user's utterance.

FIG. 10 is a diagram showing an interaction example (interaction example 3) for explaining the configuration of the fourth embodiment according to the present invention. In this embodiment, when there is a word attribute belonging to the search target word attribute list as shown in FIG. 9 in the recognition result recognized in the past, this is recognized by the recognition engine 204.
It is characterized in that the saved result is used in addition to the previous recognition result when the dictionary data for recognizing the user's utterance is created.

That is, as shown in FIG. 10, when the user's answer "Hyogo prefecture" to the question "Where are you going?" On the system side is recognized, as shown in the flowchart of FIG. 5 or 7, voice recognition is performed. RAM area 2 in engine 204
In 06b, the subsequent dictionary data having “Hyogo Prefecture” as a node is set. The recognized word attribute "Hyogo" is
This word attribute “Hyogo” is included in the search target word attribute list shown in FIG.
4 storage areas 204d.

The system then asks, "Where is Hyogo Prefecture?"
Ask. The user answers "ramen shop" to this. Since the word attribute "ramen shop" is included in the dictionary of the facility type in the basic dictionary in the voice recognition engine 204, "ramen shop" is recognized using the waveform data.

When the "ramen shop" is recognized, the voice recognition engine extracts the ramen shop dictionary included in the dictionary of the facility names in the voice recognition dictionary 214, and further stores it in the storage area 204d. Use the word attribute "Hyogo" to further narrow down the ramen shop dictionary. As a result, a dictionary of ramen restaurants in Hyogo prefecture is extracted and set in the RAM area 204b of the voice recognition engine 204. The next user utterance "OO ramen" is recognized using the dictionary data set in this way.

Since each address is attached as data to the word attribute of the facility name, a ramen shop in a specific area can be specified by searching this data.

FIG. 11 is a diagram showing an interaction example (interaction example 4) for explaining the configuration of the fifth embodiment according to the present invention. In this embodiment, when there is a word attribute belonging to the search target word attribute list as shown in FIG. 9 in the recognition result recognized in the past, this is recognized by the recognition engine 204.
It is characterized in that it is stored in the storage area 204d of the dictionary and then, if a recognition error occurs in the process of the dialogue between the system and the user, the dictionary is narrowed down using the word attribute stored in the storage area 204d. .

That is, as shown in FIG. 11, when the user's answer “Hyogo prefecture” to the question “Where are you going?” On the system side is recognized, as shown in the flowchart of FIG. 5 or 7, voice recognition is performed. RAM area 2 in engine 204
In 06b, the subsequent dictionary data having “Hyogo Prefecture” as a node is set. The recognized word attribute "Hyogo" is
Since it is included in the search target word attribute list shown in FIG. 9, it is stored in the storage area 204 d of the voice recognition engine 204.

The system then asks, "Where is Hyogo Prefecture?"
Ask. The user answers "Kobe city" to this, but the system misrecognizes this as "Akashi city" and utters the next question "Where is Akashi city?". The user knows that this question in the system misrecognized his previous answer and answers "yes". When the system recognizes this utterance using the basic dictionary, it searches the storage area 204d to find the stored word attribute "Hyogo prefecture".
Next, while narrowing down the dictionary using this word attribute, the question "Where is Hyogo prefecture?" Is created again.

This operation indicates that the dialogue returns to the location where the recognition is incorrect, as indicated by the double-lined arrow in FIG. After the re-inquiry of the system, the processing of each of the above-described embodiments is executed as indicated by the double-lined arrow in the figure.

12 to 14 are views for explaining the sixth embodiment of the present invention. FIG. 12 shows a dialogue example (dialog example 5) applied to this embodiment, and FIG. 14 is a diagram conceptually showing the structures of the voice recognition engine 204 and the voice recognition dictionary 214 in this embodiment.

In the present embodiment, the type of dictionary to be used is set in advance according to the dialogue state between the user and the system (see FIG. 12), and when each dialogue state is reached, the dictionary of the preset type is set. Extract the data from the voice recognition dictionary,
It is characterized in that it is set in the RAM area 204b of the voice recognition engine 204.

For example, if the correspondence table 204e as shown in FIG. 13 is previously formed for the dialogue states 1 to 4, in the dialogue state 4, the data of the facility name dictionary is converted into the voice recognition dictionary 214 according to the correspondence table. It is taken out from and set in the RAM area 204b, and the voice recognition of the next user's utterance is executed using this dictionary.

Alternatively, in the case of the interactive state 2, as shown in FIG. 14, an address dictionary and a facility type dictionary are retrieved from the voice recognition dictionary 214 according to the correspondence table 204e,
Both of them are set in the RAM area 204b. As a result, even if the user's next utterance is a “ramen shop”, the voice is easily recognized.

According to the present embodiment, in each state of the dialogue, by extracting only the dictionary having the utterance pattern which the user is likely to generate from the voice recognition dictionary and setting it in the voice recognition engine, the load on the engine ( The processing amount) can be reduced. As a result, the response time of voice recognition can be shortened, and the response of voice dialogue is improved.

As still another embodiment of the present invention, in a certain state, a speech recognition dictionary including word attributes necessary for accomplishing an interactive task other than the already obtained word attributes is automatically selected, It can also be set as a voice recognition engine. That is, when the recognition result having the attribute of "prefecture name" has already been obtained in the dialogue for setting the destination, the other attributes necessary for accomplishing the task are "city name", "town name", and "street address". "
For example, a dictionary including these word attributes is searched from a plurality of dictionaries, and matching dictionaries are combined and set in the voice recognition engine.

As a result, the same effect as the embodiment in this case can be obtained without having the correspondence table as in the sixth embodiment, and the reduction of the processing amount (data amount) can be expected.

[0055]

As described above with reference to each embodiment, according to the voice dialogue system of the present invention, the dictionary data relating to the already obtained result of the voice recognition is extracted from the voice recognition dictionary and the Since the recognition engine is set and the voice recognition of the next user's utterance is performed, the false recognition rate of the voice recognition is lowered. At the same time, the amount of data set in the voice recognition engine decreases, so the load on the recognition engine decreases and the recognition speed increases. As a result, the response speed of the dialogue is improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a voice dialogue system according to an embodiment of the present invention.

FIG. 2 is a diagram showing a configuration of a voice recognition engine shown in FIG.

FIG. 3 is a diagram showing a configuration of a voice recognition dictionary shown in FIG.

FIG. 4 is a diagram showing a configuration of an address dictionary having a tree-like hierarchical structure.

FIG. 5 is a flowchart for explaining the operation of the voice dialogue system according to the first embodiment of the present invention.

FIG. 6 is a first example interaction provided in the description of the flowchart of FIG.

FIG. 7 is a flowchart for explaining the operation of the voice dialogue system according to the second embodiment of the present invention.

FIG. 8 is a second example interaction provided in the description of the flowchart of FIG.

FIG. 9 is a diagram showing an example of a search target word attribute list.

FIG. 10 is a third diagram for explaining a third embodiment of the present invention.
Dialogue example.

FIG. 11 is a fourth diagram for explaining the fourth embodiment of the present invention.
Dialogue example.

FIG. 12 is a fifth diagram for explaining the fifth embodiment of the present invention.
Dialogue example.

FIG. 13 is a diagram for explaining the sixth embodiment of the present invention.

FIG. 14 is a diagram which is used for describing a sixth embodiment of the present invention.

[Explanation of symbols]

10 ... Spoken dialogue system 12 ... Display 14 ... speaker 16 ... Mike 18 ... switch 20 ... Processing device 202 ... Image display engine 204 ... Voice recognition engine 206 ... Speech synthesis engine 208 ... Application program 210 ... Engine for dialogue processing 212 ... Interactive database 214 ... Voice recognition dictionary

─────────────────────────────────────────────────── ─── Continued Front Page (51) Int.Cl. ⁷ Identification Code FI Theme Coat (Reference) G10L 15/18 G10L 3/00 537F 15/28 521W

Claims (7)

[Claims] 1. A voice recognition engine for recognizing a user's utterance, a dialogue processing engine for creating a utterance to the user according to the recognition result, and a voice synthesizing engine for synthesizing the created utterance into a voice. And a voice recognition dictionary that stores dictionary data including a voice pattern for the voice recognition, and a voice input / output unit, and the voice recognition engine searches the voice recognition dictionary for a recognition result of a previous user utterance. Then, the dictionary data related to the searched recognition result is extracted and set in the voice recognition engine to perform voice recognition of the next user's utterance. 2. The voice interaction system according to claim 1, wherein when the voice recognition engine has a plurality of recognition results of the previous user's utterance, the voice recognition engine searches the voice recognition dictionary for any of the recognition results, A voice dialogue system for setting the relevant dictionary data in the voice recognition engine. 3. The voice dialog system according to claim 1, wherein the voice recognition engine has a list of word attributes to be searched for a recognition result. 4. The voice dialogue system according to claim 3, wherein the voice recognition engine stores a word attribute corresponding to the list in the recognition result of the previous user's utterance, if the word attribute exists. , Voice dialogue system. 5. A voice recognition engine for recognizing a user's utterance, a dialogue processing engine for creating a utterance to the user according to the recognition result, and a voice synthesizing engine for synthesizing the created utterance into a voice. And a voice recognition dictionary that stores dictionary data including a voice pattern for the voice recognition, and a voice input / output unit, the voice recognition dictionary has a plurality of types of dictionaries, and the voice recognition engine is A table having a correspondence table of a dialogue state and a dictionary of a type used in the dialogue state, and in recognizing a user utterance in each dialogue state, a dictionary of a corresponding type is extracted from the voice recognition dictionary according to the correspondence table and used. There is a voice dialogue system. 6. The voice dialogue system according to claim 5, wherein the voice recognition engine extracts and uses a plurality of types of dictionaries from the voice recognition dictionary according to the correspondence table. 7. A voice recognition engine for recognizing a user's utterance, a dialogue processing engine for creating a utterance to the user according to the recognition result, and a voice synthesizing engine for synthesizing the created utterance into a voice. And a voice recognition dictionary that stores dictionary data including a voice pattern for the voice recognition, and a voice input / output unit, and the voice recognition engine has a plurality of word attributes that achieve a predetermined interactive task. The voice interaction system, wherein dictionary data including word attributes other than the word attributes used when recognizing the user's utterance is extracted from the voice recognition dictionary and used for the next recognition of the user utterance.