JP2000172483A - Method and system for speech recognition by common virtual picture and storage medium stored with speech recognition program by common virtual picture - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable efficient vocal transmission of an intension to an object in a common virtual picture by allowing a grammar management server to distribute difference information of grammar needed for speech recognition to a user terminal. SOLUTION: The system has a grammar management server which manages the grammar having a recognition object vocabulary registered in a text, etc., and a speech recognition result distribution server which distributes a speech recognition result. When a user terminal requests the grammar management server to update the difference information of the grammar needed for speech recognition (S1), the grammar management server distributes necessary difference information of the grammar to the user terminal at the start of the speech recognition (S2). The user terminal updates the grammar according to the distributed difference information (S3), recognizes an input voice by using the updated grammar (S4), and extracts (S5) and distributes (S6) the most likelihood vocabulary in the vocabulary in the recognition result to the distribution server. The distribution server distributes it to other user terminals (S7).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition method and system for a shared virtual screen and a storage medium storing a voice recognition program for the shared virtual screen, and more particularly to a scene viewed by a user according to the viewpoint of the user. In a system for recognizing a voice communicated via a shared virtual screen that can be updated in drawing, for example, a voice intention for a target in the shared virtual screen that is not directly related to the user. The present invention relates to a speech recognition method and system for a shared virtual screen that is effective for efficiently communicating a voice when performing communication, and a storage medium that stores a speech recognition program for a shared virtual screen.

[0002]

2. Description of the Related Art Conventionally, a plurality of user terminals are connected to a server device via a network, and a plurality of users can participate in their avatars in a three-dimensional virtual space constructed by three-dimensional CG. As a virtual space, for example, InterSpace (http://www.ntts.com/ispace.
html). In this system, a user can communicate by voice with other users participating in the three-dimensional shared virtual space using a voice input device such as a microphone.

[0003] On the other hand, speech recognition technology that allows a computer to understand the words spoken by humans is an important technology for exchanging information between humans and machines, and voice recognition technology for operability of multimedia. Expectations for recognition technology are great. As a voice recognition system via a network, there is a client-server system that transmits voice data input from a client to a server, executes voice recognition processing on the server, and returns a voice recognition result to the client. A low-cost personal computer that uses a high-speed (and expensive) workstation as a server when handling a task with a large processing amount such as continuous speech recognition or a task that requires immediacy. It is possible to use a form that is used by a plurality of clients.

In general speech recognition processing, a phoneme (vowel, consonant, or the like) which is a component of a linguistic speech is used as a unit, and an acoustic model in which acoustic properties are modeled for each phoneme is searched. The vocabulary and the grammar for defining the space, the language model which is a model of the statistical chain probability of the linguistic unit, and the vocabulary in which the vocabulary to be recognized is registered as text are used. Speech analysis of the input speech data (speech waveform), extracting speech features effective for speech recognition, limiting the search space in speech recognition processing with a language model, and performing acoustic matching with the acoustic model , The vocabulary having the highest likelihood among the vocabularies registered in the grammar is obtained as the speech recognition result.

[0005]

Recently, voice recognition processing can be performed on a personal computer by simplifying a voice recognition processing program by increasing the processing capability of a personal computer and gaining knowledge of voice recognition processing algorithms. It is becoming.

However, since an acoustic model for an unspecified speaker is still used, it is difficult to obtain a speech recognition performance equivalent to an acoustic model created for a specific speaker. In addition, it is difficult to communicate by voice to a target that has left a human operation (for example, to instruct a dog character by voice such as “hand” or “sit”).

As described above, in the conventional speech recognition technology via a network, an acoustic model for an unspecified speaker is prepared in the speech recognition server in order to recognize the voices of a plurality of users. Although it is an acoustic model for unspecified speakers, it is difficult to perform speech recognition that absorbs all the various factors due to individual differences in speech, and it is equivalent to an acoustic model created for a specific speaker. It is difficult to obtain the voice recognition performance.

[0008] In the conventional communication using voice in a shared virtual screen, humans directly listen to voices transmitted via a network, and perform reactions on the voices while understanding them. Here, real-time human intervention is indispensable. For example, “hand” or “sitting” is performed on an object such as a dog character that has left human operation as described above.
It is difficult to realize such a command by voice.

[0009] Therefore, according to the above-mentioned conventional technology, when communicating by voice to an object in the shared virtual screen which the user is not directly concerned with, it is not possible to efficiently communicate by voice. There is a problem. The present invention
In view of the above points, when a user communicates by voice to an object in a shared virtual screen that is not directly related to the user, the shared virtual machine is effective for efficiently communicating by voice. It is an object of the present invention to provide a method and system for voice recognition on a screen and a storage medium storing a voice recognition program for a shared virtual screen.

[0010]

Means for Solving the Problems The present invention (claim 1) provides:
In a voice recognition method on a shared virtual screen capable of drawing and updating a scene viewed by a user according to a user's viewpoint, a grammar management server that manages a grammar in which a vocabulary to be recognized is registered by text or the like, a voice recognition result And a voice recognition result distribution server that distributes the result to the user terminal,
The grammar management server distributes grammar difference information necessary for the voice recognition process to the user terminal when the voice recognition process of the user terminal starts.

According to the present invention (claim 2), in a user terminal, a user-specific acoustic model and a language model required for speech recognition are stored and managed together with a grammar. According to the present invention (claim 3), in the user terminal, the input speech is recognized using the grammar, the acoustic model, and the language model in the user terminal. According to the present invention (claim 4), in the user terminal, the result of speech recognition is transmitted to the speech recognition result distribution server.

FIG. 1 is a block diagram showing the principle of the present invention. According to a fifth aspect of the present invention, in a voice recognition method for a shared virtual screen capable of drawing and updating a scene viewed by a user in accordance with a user's viewpoint, a grammar in which recognition target vocabulary is registered as text or the like is managed. Grammar management server, and a speech recognition result distribution server that distributes speech recognition results to the user terminal. The user terminal updates the grammar difference information required for speech recognition to the grammar management server at the user terminal. Request (step 1), the grammar management server distributes grammatical difference information required for the speech recognition processing to the user terminal when the speech recognition processing of the user terminal starts (step 2). The grammar is updated by acquiring the difference information distributed from the grammar management server (step 3), and using the updated grammar, performs voice recognition of the input voice ( Step 4) Extract the maximum likelihood vocabulary from the vocabulary of the speech recognition result as the speech recognition result (step 5), transmit the speech recognition result to the speech recognition result distribution server (step 6), and deliver the speech recognition result. On the server,
The speech recognition result transmitted from the user terminal is delivered to another user terminal (step 7).

FIG. 2 is a diagram showing the principle of the present invention. The present invention (Claim 6) is a voice recognition system for a shared virtual screen capable of drawing and updating a scene viewed by a user in accordance with the viewpoint of the user, comprising: a user terminal 100 that performs voice recognition; The grammar management server 200 manages a grammar whose vocabulary is registered as text and the like, and a speech recognition result delivery server 300 that delivers speech recognition results to the user terminal 100. A grammar management unit 10 that manages grammar registered by the user, etc., and a difference information distribution unit 20 that distributes, to the user terminal, grammatical difference information required for the speech recognition process when the speech recognition process of the user terminal starts. Has,
The user terminal 100 stores and manages a user-specific acoustic model and a language model required for speech recognition together with a grammar, and a model storage unit 30 and stores input speech in the user terminal by using a grammar, an acoustic model, and a language model. The apparatus includes a speech recognition unit 40 for recognizing speech and a speech recognition result transmitting unit 50 for transmitting the speech recognition result obtained by the speech recognition unit 40.
Means for delivering the speech recognition result transmitted from 00 to other user terminals.

According to a seventh aspect of the present invention, there is provided a storage medium storing a voice recognition program for a shared virtual screen mounted on a grammar management server capable of drawing and updating a scene viewed by a user in accordance with the viewpoint of the user. When the speech recognition processing of the user terminal is started, grammatical difference information required for the speech recognition processing is delivered to the user terminal.

The present invention (claim 8) is a storage medium storing a voice recognition program for a shared virtual screen mounted on a user terminal capable of drawing and updating a scene viewed by a user according to the viewpoint of the user. A process for storing and managing user-specific acoustic models and language models necessary for speech recognition together with grammar in storage means, and a grammar distributed from a grammar management server for managing grammar necessary for performing speech recognition A process of acquiring and storing the difference information of the utterance, a process of recognizing the input speech using a grammar, an acoustic model, and a language model, and a process of transmitting the result of the speech recognition to the speech recognition result distribution server. Have.

As described above, the present invention has a grammar management server that manages a grammar in which a vocabulary to be recognized is registered as text and the like, and a speech recognition result distribution server that delivers speech recognition results to user terminals. In a grammar management server provided for sharing a personally created grammar for a certain voice recognition among a large number of user terminals, a grammar necessary for the voice recognition process is started when the user terminal starts the voice recognition process. Distribute the difference information to the user terminal, perform voice recognition of the input voice using a grammar that updates the obtained difference information, transmit the voice recognition result to the voice recognition result distribution server, and perform voice recognition distribution. By distributing the transmitted speech recognition result to another user terminal in the server, it is possible to draw and update the scene viewed by the user.

[0017]

FIG. 3 shows the configuration of a speech recognition system according to the present invention. The system shown in FIG.
The user terminal 100, the grammar management server 200, and the speech recognition result distribution server 300 are connected to the network 400.

The user terminal 100 has a data management unit 11
0, grammar creation processing unit 120, acoustic model adaptation unit 130,
It comprises a speech recognition processing unit 140, a grammar update request unit 150, a display unit 160, and a communication control unit 170. Note that one user terminal 100 is connected to the network in FIG. 3 for simplicity of description, but originally,
It is assumed that n units are connected.

The data management unit 110 stores and manages data such as acoustic models, language models, and grammar required for speech recognition processing. The grammar creation processing unit 120 creates a grammar unique to the user in which the vocabulary to be recognized is registered as text or the like. The acoustic model adaptation unit 130 inputs the speech 18
The sound model for the unspecified speaker is adapted for the user according to the sound characteristics of the user 0. According to a specific method of adapting, in accordance with the instruction displayed on the display unit 160 by the acoustic model adapting unit 130, the user 180 performs pronunciation so as to cover all vowels and consonants in the acoustic model for an unspecified speaker. The acoustic model adaptation unit 130 captures the acoustic characteristics of the user 180 from the pronunciation and adapts the acoustic model for the unspecified speaker to the user.

The voice recognition processing unit 140 includes the data management unit 1
A speech recognition result is created by using data necessary for speech recognition processing managed by 10. Grammar update request unit 150
Requests the grammar management server 200 to update the grammar difference information for the target for which the speech recognition processing unit 140 performs speech recognition. The display unit 160 draws and updates the shared virtual scene for the user 180.

The communication control unit 170 controls the grammar management server 20
0 and the communication with the speech recognition result distribution server 300 are managed. The grammar management server 200 is a server that manages a grammar created by a user for a certain voice recognition target in order to share the grammar created personally with many other user terminals. It comprises a unit 230, an upload unit 240, and a communication control unit 250.

The grammar database 210 stores the user terminal 1
The grammar information transmitted from 00 is stored and managed. The search unit 220 searches the grammar database 210 based on a request specified by the grammar update request unit 140 of the user terminal 100. The loading unit 230 loads difference information from the grammar managed by the data management unit 110 of the user terminal 100 searched by the search unit 220.

The upload unit 240 is provided for the user terminal 10
The grammar information transmitted from 0 is uploaded to the grammar database 210. The communication control unit 250 controls the user terminal 1
Manage communication with 00. Speech recognition result distribution server 30
Reference numeral 0 denotes a server that distributes the speech recognition result obtained from the user terminal 100 to the selected other user terminals, and includes a speech recognition result management unit 310, a distribution user management unit 320, and a communication control unit 330.

The speech recognition result management section 310 stores and manages a speech recognition result transmitted from the user terminal 100.
The distribution user management unit 320 includes a speech recognition result management unit 310
Manages the user terminal 100 that distributes the voice recognition result managed by the user terminal 100. The selection of the user terminal to be distributed is based on Japanese Patent Application
It is assumed that the method depends on the method of selecting a master terminal disclosed in US Pat.

The communication control unit 330 manages communication with the user terminal 100. Next, an outline of the operation of the present invention will be described. According to the common rules among the acoustic model, the language model, and the grammar required for the speech recognition processing, the language model and the acoustic model reflecting the acoustic characteristics of the individual, not for the unspecified speaker, are stored in the individual user terminals 100. The grammar created by each user shared by other users is stored in the user terminal 100 itself that created the grammar, and is also stored in the grammar management server 200 so that other users can share the grammar.

The grammar creation processing unit 120 of the user terminal 100
The grammar created by each user by the user terminal 10
0 is updated and stored in the grammar database 210 via the communication control unit 170 of the grammar management server 200, the communication control unit 250 of the grammar management server 200, and the upload unit 240. Note that the grammar of the common rule is already in the data management unit 11 of the user terminal 100.
0 and may be stored in the user terminal 100.
May be stored in both the data management unit 110 and the grammar database 210.

The process of reflecting the latest grammar on each user terminal is performed as follows. As reflection processing of the first grammar, the user 180 selects a speech recognition target,
A case where a grammar update request is issued from the user terminal 100 will be described. The grammar update in this case updates only the grammar difference information relating to the speech recognition target. The user terminal 100 transmits a grammar version related to the speech recognition target to the grammar management server 200.

Next, the grammar management server 200 compares the transmitted grammar version regarding the speech recognition target with the latest grammar version regarding the speech recognition target stored in the grammar database 210. Only when it is different, the difference information from the latest version of the grammar of the grammar database 210 is returned to the user terminal 100. As a second grammar reflection process, when a common grammar is required for the entire world (for example, in a world reflecting the Kansai cultural sphere, the Kansai dialect needs to be communicated), the grammar management server 200 sends the user terminal 100 A method for updating the grammar of the grammar will be described.

In this case, the grammar update time is determined, for example, when the user terminal 100 joins the world via the world management server (a server (not shown) for managing the login and logout of the user to the world). The grammar management server 200 compares the entire grammar version sent via the world management server (not shown), and sends the latest grammar information to the user terminal 100 when they are different. I do.

As a third grammar reflection process, a method of passively updating the grammar necessary for approaching the user's avatar from the voice recognition target will be described. In this case, the grammar updating method includes, for example, a grammar version for the speech recognition target stored in the user terminal 100 when the speech recognition target accompanies a grammar version for the user and approaches a distance within a range of the user's avatar. If the grammar version is different, the grammar version of the user terminal is transmitted to the grammar management server 200, and grammar difference information is obtained and updated. Alternatively, the above-described first grammar reflection processing is performed when the distance to the user's avatar approaches a certain range.

Next, the operation of the speech recognition processing according to the above configuration will be described. FIG. 4 is a flowchart of the voice recognition processing of the present invention. Step 101) The grammar update request unit 150 requests the grammar update request unit 150 to update grammar difference information required for performing speech recognition, and determines whether or not grammar update is required. If it is necessary to update, go to step 102; if not, go to step 1
Shift to 03.

Step 102) Grammar management server 200
The grammar managed by the data management unit 110 of the user terminal 100 is updated based on the grammar difference information loaded by the loading unit 230. Step 103) Next, voice recognition is performed by the voice recognition processing unit 140 using data necessary for the voice recognition process managed by the data management unit 110 of the user terminal 100.

Step 104) Then, the user terminal 1
The vocabulary with the highest likelihood among the vocabulary registered in the grammar managed by the data management unit 110 of 00 is extracted as a speech recognition result in a text format. Step 105) Finally, the speech recognition result expressed in the text format is transmitted to the speech recognition result distribution server 300, and the process ends.

[0034]

Embodiments of the present invention will be described below with reference to the drawings. As a premise of this embodiment, the data management unit 110 of the user terminal 100 has an action table 600 in which an instruction command and an action corresponding to the instruction are described, and the user terminal 100 becomes a master terminal ( When the user terminal becomes the master terminal,
Japanese Patent Application No. 10-347396) describes in detail
It is assumed that processing corresponding to the action described in the action table 600 is performed.

In the following embodiment, as a shared virtual screen,
A description will be given taking a three-dimensional shared virtual space as an example. FIG. 5 shows an example of speech recognition in a three-dimensional shared virtual space according to one embodiment of the present invention. As shown in FIG.
A user 51 participating in a certain world 500 using 00
0 is a dog character 5 existing in the world 500
A case where voice communication is performed with respect to 20 will be described. The dog character 520 operates based on the action table 600 as shown in FIG.

FIG. 7 shows the structure of a grammar database according to one embodiment of the present invention. Grammar database 210 shown in FIG.
Is composed of a plurality of files to which one file is assigned for each speech recognition target. The figure shows a state in which one grammar is stored in one file, and the grammar identifier and the start are a delimiter of one grammar, and each of many grammars is surrounded by the grammar identifier and the start. Are stored in one file.

The word declaration defines all meaningful words used in one grammar, how words recognized using the acoustic model and the language model correspond to meaningful words, It indicates whether there is a word with the same meaning. In this example, words having the same meaning are represented by an OR symbol with a "|" mark, and "Ido" and "Ido Please" are the same meaningful words.

A grammar declaration is one meaningful statement.
It shows how meaningful word groups are composed in what order. The notation of this grammar is a known method. In this example: • $ number = ($ 1 | $ 2 | $ 3); "1" or "2" or "3" is "number". $ Num = $ number $ steps; "one step", "two steps" or "three steps" is "num".・ $ Direction = $ before; “before” is “direction”.・ In $ dir = $ direction $; “before” is “dir”.・ Number = ($ num $ dir | $ dir $ num); "one step" or
“2 steps”, “3 steps” and “before”, or “1 step” or “2 steps” or “3 steps” from “before” is “number”.・ $ Sentence = $ numdir $ move; "one step" or "two steps"
Alternatively, "one step" or "two steps" or "three steps" and "movement" from "three steps" and "before" or "before"
Is "sentence".

Therefore, "move one step before", "move one step before", and "move one step before" have the same meaning.
Also, there are two version information for each file, and there is version information for each grammar separated by the grammar identifier and the start. When these are used, as one of two version information for each file of the grammar database 210, one is for updating one entire file under the control of a control device or the like that maintains and manages the grammar management server 200. Is version information to be updated / used. The version information includes a version number and a creation date. When a common grammar is required for the entire world (in a world reflecting the Kansai cultural sphere, the Kansai dialect needs to be communicated), the grammar management server 200 It is used when updating the entire grammar for the user terminal 100.

The version information for each file of the other grammar database 210 includes the latest update date / time, etc., and each time the file of the grammar database 210 is updated by creating a grammar from the user terminal 100, the update date / time is updated. Updated at time. This version information is used when transmitting the version information from the grammar update request unit 150 of the user terminal 100 to the voice recognition target recorded in the data management unit 110 having the same configuration format as the database. is there. This version information is used to obtain the latest update date and time faster than the process of obtaining the latest update date and time by searching from version information in each grammar. Has a high processing capability, and even if a version information search process for each grammar is performed without using the version information, the latest update date / time can be easily obtained, and other processes are not affected. If so, the version information for each file may not be necessary.

The version information for each grammar is updated each time the grammar created for the speech recognition target by the grammar creation processing unit 120 of the user terminal 100 is transmitted. This version information is the latest update date and time (date sent)
And so on. Also, in the data management unit 110 of the user terminal 100 that has transmitted the creation grammar, the version information is updated at the time of transmission together with the creation grammar. The data management unit 1 at the time of grammar creation in the user terminal 100 at this time
In the method of updating the creation grammar and the version information of No. 10, if only the created grammar and its version information are updated by the processing of the user terminal 100, the version of another grammar may remain in the old generation (because the old grammar and the version are not updated) The version information that is the latest update date and time of the entire grammar for the speech recognition target is the latest, but the other grammars are still the contents of the old generation, and the date of the latest update date before correction From the modification date to the latest update date are no longer reflected), so version information that is the latest update date and time of the entire grammar for the speech recognition target before modification stored in the data management unit 110 is also transmitted. It is necessary to obtain and update difference information by processing of the grammar management server 200 described below.

The grammar management server 200 stores the version information for each grammar in the grammar database 210,
Compared with the version information (latest update date / time) for the speech recognition target stored in the data management unit 110 transmitted from the grammar update request unit 150 of the user terminal 100, Only the grammar in the grammar database 210 of the time version is used as difference information,
Transmit to the user terminal 100. In the user terminal 100,
The individual grammar of the data management unit 110 is updated with the transmitted grammar difference information for the speech recognition target, and is updated. Similarly, the version information (latest update date / time) is also updated.

FIG. 8 is a sequence chart showing an example of the operation of the voice recognition processing according to one embodiment of the present invention. Step 201) In FIG. 5, when speech recognition processing is started by, for example, clicking the dog character 520 in the world 500 with a mouse, the user terminal 100
Of the grammar management server 200 by the
Is requested to update the grammatical difference information required for performing the voice recognition process on the dog character 520.

Step 202) Upon receiving the update request, the grammar management server 200 searches the grammar database 210 for a grammar necessary for performing the voice recognition processing on the dog character 520 by the search unit 220, and the load unit 230 Data management unit 110 of user terminal 100
Load the difference information with the grammar managed by. Step 203) Next, the grammar difference information loaded by the loading unit 230 is updated and stored by the data management unit 110 of the user terminal 100, and actual speech recognition processing can be performed using this grammar. Here, a grammar update request from the client side (user) when the user selects a speech recognition target is taken as an example. For example, in a world that reflects the Kansai cultural sphere, the Kansai dialect is communicated. If a common grammar is required for the entire world, the grammar is updated from the server side (grammar management server 200) to the client (user), or the grammar is passively approached from the voice recognition target. Various application examples are conceivable, such as updating the grammar.

Here, the grammar refers to a combination of a character string describing a voice uttered by a user in a kana and a corresponding character string notation. For example, in the action table 600 as shown in FIG. 6, when a text command of “move one step ahead” is obtained as a speech recognition result, at least the grammar “Immediately before” is included in the grammar.
Cannot be recognized unless it is registered that the utterance corresponds to the notation of the character string “move forward one step”. In addition, “move one step forward” and “move one step forward”
If you are not registered in the grammar that the utterance "Please move forward one step" is the same as the voice recognition result as "Move one step ahead", each will obtain a different voice recognition result. Would. Although these utterances are all the same in content, they lead to obtaining different speech recognition results depending on how the grammar is registered. Conversely, even if the utterances are exactly the same, different speech recognition results can be obtained depending on how the grammar is registered, so a different grammar is registered for each object to be subjected to speech recognition processing, and the grammar is stored in the user terminal. By sequentially reflecting, it is possible to flexibly communicate the user's voice.

Step 204) The voice recognition process for the dog character 520 is performed using the grammar updated by the update request of the grammar update request unit 140 of the user terminal 100.
This is performed by the voice recognition processing unit 130. Step 205) The speech recognition result expressed in the text format is transmitted to the speech recognition result distribution server 300.

Step 206) Upon receiving the speech recognition result, the speech recognition result distribution server 300 causes the speech recognition result management unit 310 to temporarily store the received speech recognition result. Step 207) Next, the distribution user management unit 320 selects a user terminal to distribute the received voice recognition result. Step 208) The communication control unit 330 distributes the speech recognition result to the selected user terminal 100.

Step 209) Upon receiving the speech recognition result distributed from the speech recognition result distribution server 300, the user terminal 100 causes the dog character 520 to start the corresponding action defined by the action table 600. Accordingly, it is possible to efficiently communicate with voice when communicating with voice to a target in the shared virtual space that the user is not directly aware of.

Step 210) On the other hand, the user terminal 10
The user-specific grammar created by the grammar creation unit 120 of 0 is updated and stored by the data management unit 110. Step 211) At the same time, the communication control unit 170
It is transmitted to the grammar management server 200. Step 212) The grammar management server 200 that has received the grammar information uploads the grammar information to the grammar database 210 by the upload unit 240.

Thus, the grammar unique to the user can always be reflected on other user terminals as the latest grammar. Here, a case where an action is performed based on the definition of the action table 600 in step 209 will be described. The user terminal 100 that has recognized the command for the dog character 520 transmits the voice-recognized command to the voice recognition result distribution server 300. The speech recognition result distribution server 300 distributes the command for the dog character 520 to the master terminal that manages the behavior of the dog character 520. In the master terminal that manages the action of the dog character, the shared object processing means (not shown) of the master terminal uses the action table stored in the data management unit to perform an action corresponding to the command for the dog character. .

Also, by managing the acoustic models required for the speech recognition processing individually for each user, the acoustic models can be flexibly adapted for each user who actually inputs speech, so that the best speech model for each user can be obtained. It is possible to obtain recognition performance. As described above, according to the present invention, in a shared virtual screen in which a scene viewed by a user can be drawn and updated in accordance with the viewpoint of the user, audio is output to an object in the shared virtual screen that is not directly related to the user. When the communication is performed by the user, the communication by voice can be efficiently performed.

Although the above embodiment has been described based on the configuration of FIG. 3, each component of the user terminal 100, the grammar management server 200 and the speech recognition result distribution server 300 is constructed as a program, and The terminal is easily stored in a disk device connected to a computer used as a server or a portable storage medium such as a floppy disk, a CD-ROM, or the like, and is easily installed by implementing the present invention. The invention can be realized.

The present invention is not limited to the above embodiment, but can be variously modified and applied within the scope of the claims.

[0054]

As described above, according to the present invention, in voice recognition on a shared virtual screen, a user directly perceives a shared virtual screen in which a scene viewed by the user can be drawn and updated according to the user's viewpoint. When communicating by voice to an object in the shared virtual screen that is not
It is possible to efficiently communicate by voice.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining the principle of the present invention.

FIG. 2 is a principle configuration diagram of the present invention.

FIG. 3 is a configuration diagram of a speech recognition system of the present invention.

FIG. 4 is a flowchart of a voice recognition process according to the present invention.

FIG. 5 is an example of speech recognition in a three-dimensional shared virtual space according to one embodiment of the present invention.

FIG. 6 is an example of an action table according to an embodiment of the present invention.

FIG. 7 is an example of a grammar according to an embodiment of the present invention.

FIG. 8 is a sequence chart showing an example of a speech recognition operation according to an embodiment of the present invention.

[Explanation of symbols]

 Reference Signs List 10 grammar management means 20 difference information distribution means 30 model storage means 40 speech recognition means 50 speech recognition result transmission means 100 user terminal 110 data management section 120 grammar creation processing section 130 acoustic model adaptation section 140 speech recognition processing section 150 grammar update request section 160 display unit 170 communication control unit 180 user 200 grammar management server 210 grammar database 220 search unit 230 load unit 240 upload unit 250 communication control unit 300 voice recognition result distribution server 310 voice recognition result management unit 320 distribution user management unit 330 communication control unit

Claims (8)

[Claims] 1. A speech recognition method for a shared virtual screen capable of drawing and updating a scene viewed by a user according to a viewpoint of a user, wherein a grammar management server for managing a grammar in which vocabulary to be recognized is registered as text or the like. And a speech recognition result delivery server that delivers a speech recognition result to the user terminal, wherein the grammar management server includes, at the start of speech recognition processing of the user terminal, difference information of grammar required for the speech recognition process. And distributing the same to the user terminal. 2. The speech recognition method for a shared virtual screen according to claim 1, wherein the user terminal stores and manages a user-specific acoustic model and a language model required for speech recognition together with a grammar. 3. The voice recognition method for a shared virtual screen according to claim 1, wherein at the user terminal, the input voice is voice-recognized at the user terminal using the grammar, the acoustic model, and the language model. 4. The voice recognition method for a shared virtual screen according to claim 1, wherein the user terminal transmits a voice recognition result to the voice recognition result distribution server. 5. A grammar management server for managing a grammar in which a vocabulary to be recognized is registered as a text or the like in a voice recognition method on a shared virtual screen capable of drawing and updating a scene viewed by the user in accordance with the viewpoint of the user. And a speech recognition result distribution server that delivers a speech recognition result to a user terminal. The user terminal requests the grammar management server to update grammatical difference information required for speech recognition. In the grammar management server, when the speech recognition process of the user terminal is started, grammar difference information required for the speech recognition process is delivered to the user terminal, and in the user terminal, the grammar management server delivers the grammar difference information. Acquiring the difference information and updating the grammar, performing speech recognition of the input speech using the updated grammar, and obtaining a result of the speech recognition. Extracting the maximum likelihood vocabulary in the vocabulary as a speech recognition result, transmitting the speech recognition result to a speech recognition result distribution server, and, in the speech recognition result distribution server, the speech recognition result transmitted from the user terminal. A speech recognition method for a shared virtual screen, which is distributed to another user terminal. 6. A speech recognition system for a shared virtual screen capable of drawing and updating a scene viewed by a user according to a viewpoint of the user, wherein a user terminal for performing speech recognition and a vocabulary to be recognized include text or the like. A grammar management server that manages the registered grammar; and a speech recognition result distribution server that distributes the speech recognition result to the user terminal, wherein the grammar management server manages the grammar in which the vocabulary to be recognized is registered as text or the like. A grammar management unit, and a difference information distribution unit that distributes grammar difference information necessary for the speech recognition process to the user terminal when the speech recognition process of the user terminal starts, the user terminal includes: Model storage means for storing and managing user-specific acoustic models and language models required for speech recognition together with grammar, and input speech stored in the user terminal A voice recognition unit that performs voice recognition using the grammar, the acoustic model, and the language model; and a voice recognition result transmitting unit that transmits a voice recognition result obtained by the voice recognition unit. The server has means for distributing a speech recognition result transmitted from the user terminal to another user terminal. 7. A storage medium storing a voice recognition program for a shared virtual screen mounted on a grammar management server capable of drawing and updating a scene viewed by a user in accordance with the viewpoint of the user, wherein the user terminal A storage medium storing a voice recognition program for a shared virtual screen, wherein at the start of the voice recognition process, grammatical difference information required for the voice recognition process is delivered to the user terminal. 8. A storage medium storing a voice recognition program for a shared virtual screen mounted on a user terminal capable of drawing and updating a scene viewed by a user in accordance with a viewpoint of the user, the storage medium being necessary for voice recognition. Process of storing and managing the user-specific acoustic model and language model together with the grammar in the storage means, and acquiring the difference information of the grammar distributed from the grammar management server that manages the grammar necessary for performing speech recognition. Storing the input speech using the grammar, the acoustic model, and the language model, and transmitting the speech recognition result to the speech recognition result distribution server. A storage medium storing a voice recognition program for a shared virtual screen as a feature.