JP5500647B2 - Method and apparatus for generating dynamic speech recognition dictionary - Google Patents

Description

  The present invention relates to a method for generating a dynamic dictionary for speech recognition and an apparatus for generating the dynamic dictionary.

  In recent years, in the field of speech recognition, it has become possible to dynamically (online) create a dictionary for speech recognition using GTP (Grapheme To Phoneme) technology that converts written characters to the smallest speech units (phonemes). It has become. As an example of use of this technology, there is a realization of a music search function by voice operation that is put into practical use in an in-vehicle navigation system. This function was created after reading information was created based on music information (music title, artist, etc.) sucked from the device (hard disk drive / portable player) and a dynamic dictionary for voice recognition was created. By performing voice recognition using a dictionary, music information can be selected and operated.

  However, in order to perform these music searches, it is indispensable that the creation of the dynamic dictionary is completed, and it takes a long time to process this dictionary creation.

  To solve this problem, when creating a new voice recognition dictionary (or differential update), the dictionary is created by dividing it into a plurality of sets, and loading is performed in order from the completed dictionary until the start of recognition. A method for shortening the time is known (for example, see Patent Document 1).

  The dynamic dictionary creation by GTP is not limited to music information, but it is a predetermined search such as POI (Point Of Interest: Search for facilities such as convenience stores near the current position of the vehicle, search for facilities such as gas stations along the set route, etc. It is possible to generate a dictionary based on various character strings such as facility names such as searching facilities according to conditions) and address book names. For example, a method of performing POI / address search by voice recognition by automatically generating a word list of street names and street names near the current position of the vehicle and using it for voice recognition is also known (see, for example, Patent Document 2).

JP 2010-72081 A JP 2002-318136 A

  However, in the speech recognition dictionary creating apparatus disclosed in Patent Document 1, although it is possible to recognize a part of vocabulary (songs) in a short time by dividing it, all vocabularies are recognized. There is no change in the time until it becomes possible compared to the case where it is created without dividing. In addition, in the method for forming a spatially constructed word list disclosed in Patent Document 2, dynamic dictionary creation using GTP requires many devices such as POI or other facility names and address book names to be created. Even when processing is requested from a function, the same problem as described above occurs.

  If there are multiple dictionary creation requests, either waiting until the current dictionary creation request is completed or canceling the current dictionary creation request for the current dictionary creation request It is necessary to process with. When waiting for the completion of the current dictionary creation request, if dynamic dictionary creation processing is performed for a large number of cases, it will take time to complete the current request and processing will start immediately. I can't. Also, when canceling the current dictionary creation request, it is possible to immediately start the GTP dictionary creation for POI search by canceling the current dictionary creation request, but the current dictionary creation has been suspended. After the creation of the GTP dictionary for POI search is completed, the dictionary for which creation has been stopped must be created again from the beginning. For this reason, when GTP dictionary creation for POI search is continuously requested at regular intervals, the current dictionary creation request may not be completed.

  The present invention has been made in view of such circumstances, and among the search functions by voice, it is possible to shorten the time until the function that responds to the user's utterance becomes effective, and to respond to the user's utterance. It is an object of the present invention to provide a method for generating a dynamic speech recognition dictionary and a device for generating the same, which can shorten the time until a function similar to a function becomes effective.

The dynamic speech recognition dictionary generation method according to the present invention includes a step (S11) of newly creating a dictionary generation thread when a request for generation of a speech recognition dictionary is received from a user, and a task of the dictionary generation thread. A dynamic speech recognition dictionary generating method for generating a speech recognition dictionary necessary for a task set in the list, the step of adding to a list of tasks during dictionary generation (S12). A step (S14) of determining whether there is a search category that classifies commands corresponding to tasks that require a task for each purpose, and whether there is a search category that is being prioritized, The voice recognition dictionary required for the task belonging to the search category is given priority over the voice recognition dictionary required for the task belonging to the search category not in the priority process. A step of forming (S19), when it recognizes a command that requires a dynamic generation of speech recognition dictionary when there is speech of the user, the search in the priority processing a search category task corresponding to the command belongs And a step of setting the category (S30) .

In a method of generating a dynamic speech recognition dictionary according to the present invention, when the voice recognition dictionary generated is completed, a step (S41) for deleting the completed tasks of said voice recognition dictionary generated from a list of dictionary generation in task, If the task has been completed of the voice recognition dictionary generation What task der belonging to the search category in the priority processing, there is no remaining task that belongs to the search category in the priority processing in the dictionary generation in the list of tasks , the search categories can be step of releasing the (S47) and further including configuration to that those in the priority processing.

The dynamic speech recognition dictionary generation apparatus according to the present invention includes a thread creation means (S11) for creating a new dictionary creation thread when a speech recognition dictionary creation request is received from a user, and the dictionary creation thread. and a list addition means (S12) to add tasks to the list of dictionary generation in task, a generator of dynamic speech recognition dictionary to generate a speech recognition dictionary required by the task set in said list, Preferential processing search category presence / absence determining means (S14) for determining whether or not there is a search category for classifying commands corresponding to tasks that require speech recognition dictionary generation for each purpose; when it is determined that there is the search categories being processed, the speech recognition dictionary required by the task belonging to the search category, belonging to the search category not being priority processing And means (S19) for generating in preference speech recognition dictionary required by disk, when it recognizes a command that requires a dynamic generation of speech recognition dictionary when there is speech of the user, corresponding to the command the search category that task belongs becomes configured to have a means (S30) for setting a search category in priority processing.

In generating apparatus for dynamic speech recognition dictionary according to the present invention, when the voice recognition dictionary generation is completed, the dictionary generation task deletion to delete completed tasks of said voice recognition dictionary generated from a list of dictionary generation in task (S41 a), I task der that the completed tasks of speech recognition dictionary generation belongs to the search category in the priority processing, it remains a task that belongs to the search category in the priority processing in the dictionary generation in the list of tasks Tei If not, the search categories can be configured to have a means for releasing (S47) that is intended in the priority processing.

  According to the method and apparatus for generating a dynamic speech recognition dictionary according to the present invention, when a command that requires a dynamically generated speech recognition dictionary is recognized due to a user's utterance, the task during dictionary generation is performed. The processing priority of the dictionary generation thread of the task is changed depending on whether or not the task belongs to the search category corresponding to the command. It is possible to shorten the time until the corresponding function becomes effective, that is, the request that the user wants immediately is processed preferentially. In addition, it is possible to shorten the time until the function similar to the function responding to the user's utterance becomes effective.

It is a block diagram which shows the structure of the production | generation apparatus of the dynamic speech recognition dictionary which concerns on one Embodiment of this invention. It is a figure which shows the example of the process performed as a result of carrying out the speech recognition using them, and the recognition dictionary which extracted and created the telephone directory data of the speech recognition dictionary from the dictionary production | generation task. It is a figure which shows the example of the process performed as a result of carrying out the speech recognition using them using the recognition dictionary which extracted and created the music data of the speech recognition dictionary from the dictionary production | generation task. It is a figure which shows the example of the process performed as a result of carrying out the speech recognition using them using the recognition dictionary which extracted and created the POI list | wrist of the speech recognition dictionary from the dictionary production | generation task. It is a flowchart which shows the process at the time of receiving the request | requirement which concerns on speech recognition dictionary production | generation . It is a flowchart illustrating the first half of the processes performed when there is speech of a Subscriber. Is a flow chart showing the second half of the process when there is speech of a Subscriber. It is a figure which shows the correspondence of a user's utterance content, a dictionary production | generation task, and a search category. It is a flowchart which shows the process at the time of completion of dictionary production | generation. It is a figure which shows the example of the time flow of the process performed without a priority being set when there is a user's utterance . It is a figure which shows the example of the time flow of the process performed according to the priority set when there is a user's utterance .

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  A dynamic speech recognition dictionary generating apparatus according to an embodiment of the present invention is configured as shown in FIG. In FIG. 1, a dynamic speech recognition dictionary generation apparatus 100 has a processing unit 11 composed of a computer unit (including a CPU). The processing unit 11 is connected to a navigation device 12 and a storage unit 13 (for example, a hard disk drive (HDD) unit) that stores map information and various information necessary for the navigation device 12. Further, the processing unit 11 is provided with a display unit 14 that is provided in the vehicle interior and configured by an LCD or the like, an output circuit 16 that supplies an audio signal to a speaker 15 provided in the vehicle interior, and a user's utterance. The voice input circuit 17 is connected to a voice recognition engine 20 which is an internal driver that recognizes a user's voice (word) and converts it into text. Connected to the voice input circuit 17 are a switch 18 operated by the user and a microphone 19 for converting a voice signal generated by the user's speech into an electrical signal. Further, a communication device (Bluetooth) 21, a CD / DVD player 22, and a connector 25 are connected to the processing unit 11. The communication device 21 is wirelessly connected to a mobile phone 22 brought into the vehicle. The CD / DVD player 23 is built in an instrument panel in the vehicle, and the CD / DVD player 23 can be inserted into the CD / DVD player 23. Further, the connector 25 is provided so that its insertion port is exposed to the instrument panel, and a portable player 26 can be connected to the connector 25. Note that music is also stored in the storage unit 13.

  On the other hand, the processing unit 11 is configured to newly create a dictionary generation thread when a voice recognition dictionary generation request is received from the user, and add the dictionary generation thread task to the list of dictionary generation tasks. Is done. The processing unit 11 is configured to determine whether or not there is a search category in priority processing in which commands corresponding to tasks requiring dictionary generation are classified for each purpose. Further, when the processing unit 11 recognizes a command that requires a dynamically generated speech recognition dictionary when a user utters, the processing unit 11 determines whether or not there is a task that is generating a dictionary and the task corresponds to the command. The processing priority of the dictionary generation thread of the task is changed depending on whether the task belongs to the category.

  In FIG. 6C, since the purpose of “Favorite Place Search” and “POI Search” is to set the destination of the navigation device, these commands belong to the search category “set destination”. In addition, since the purpose of “Phone book” and “Recent Calls” is to make a call, it belongs to the search category of “calling a phone”. Furthermore, since the purpose of “DAP Music Search”, “HDD Music search”, and “Bt Audio search” is to search for music in each device, these commands belong to the search category “music search”.

  The processing unit 11 is configured to determine whether or not dictionary generation has been completed and to delete a task for which dictionary generation has been completed from the list of tasks that are being generated by the dictionary. Further, the processing unit 11 determines whether or not the task for which dictionary generation has been completed is a task belonging to the search category being preferentially processed, and the task belonging to the search category being preferentially processed remains in the list of tasks being generated. It is configured to change the processing priority of a thread according to whether or not it exists.

The operation of the dynamic speech recognition dictionary generation apparatus 100 configured as described above will be described. First, a dictionary generation process as shown in FIG. 5 will be described. When a recognition dictionary generation request is received by a user's operation such as editing a song stored in the storage unit 13 or connecting the portable player 26 to the connector 25, the processing unit 11 newly generates a dictionary generation thread. (S11), the list is added to the dictionary generation task list (S12), and the dictionary generation task list is updated (S13). Then the processing unit 11 determine the constant whether search categories in priority process is set as described below (S14), if has not been set (at S14 NO), the threads created new The processing priority is set to the standard value of the dictionary generation processing thread (S17).

If the search category in the priority process is set (YES in S14), the processing unit 11 determines whether the task that subsequently generated the dictionary generation thread belongs to the search category in the priority process ( S15). If the processing unit 11 is a task belonging to the search category being preferentially processed (YES in S15), the processing priority of the newly created thread is set to the standard value of the dictionary generation processing thread (S17). If the task does not belong to the search category (NO in S15), the thread processing priority is set to (standard value -1) and the thread processing priority is lowered by one level (S18). Thereafter, generation of a voice recognition dictionary in response to the voice recognition dictionary generation request is started (S19).

Specifically, connection to the connector 25 of the portable for the player 26, the editing operation of the music data stored in the storage unit 13, and a voice recognition dictionary due to the radio connection of the mobile telephone 22 to the communicator 21 Is generated. For example, as shown in FIG. 3, the music data of the portable player 26, or from the music data stored in the storage unit 13, the speech recognition dictionary Ru is generated. For example, as shown in FIG. 2, the speech recognition dictionary from the phone book data of the cellular phone 22 is Ru is generated. For example, a voice recognition dictionary related to a telephone call (see FIG. 6C) is generated from the outgoing / incoming history data of the mobile phone 22 in the same manner as the voice recognition dictionary related to the telephone directory data . When the generation of the voice recognition dictionary is completed, for example, when the user turns on the switch 18 and utters a desired album name toward the microphone 19, the voice is recognized by the voice recognition dictionary generated from the music data of the portable player 26. Recognized and played from the beginning of the album ( see FIG. 3). When the user turns on the switch 18 and speaks the name of the other party who wants to make a call toward the microphone 19, the voice is recognized by the voice recognition dictionary generated from the phone book data, and a call is made to the other party ( ( See FIG. 2). Note that a speech recognition dictionary may be generated from the POI list shown in FIG. The next dictionary generation process is started when one dictionary generation process is completed, but each dictionary generation process may be performed in a time-sharing manner.

Next, as shown in FIGS. 6A and 6B, a dictionary generation process in the case where there is a user's utterance will be described. When a command that requires a dynamically generated speech recognition dictionary is recognized from the user's utterance (YES in S21 and S22), the processing unit 11 sets a search category corresponding to the command as shown in FIG. 6C. It is determined whether the dictionary of the task to which it belongs has already been generated or is being generated (S23). Then, if there is no dictionary belonging to the search category corresponding to the command and the dictionary generation has not started (NO in S23), the processing unit 11 follows the processing procedure of FIG. 5 (S24) and sets a dictionary generation thread. Create a new speech recognition dictionary necessary for the task corresponding to the command by creating a new one (S11), adding the task to the list (S12), and updating the list of tasks being generated (S13). The process which concerns on is performed.

Processing unit 11 refers to the list of dictionary generation during a task (S26), if it is determined that there is a task in the dictionary generating (S25 in YES), N-th of the list (initial value = 1 (S27)) The task A is defined as A (S28), and it is determined whether or not the task A is included in the search category to which the recognition command belongs (S29). Then, the processing unit 11, the case where the task A is determined to be included in the affiliation search category recognition command (YES in S 29), set this search category to search categories in priority processing (S30, S31) The thread priority of the task A is set to a standard value (S32). If there is at least one task included in the search category to which the recognition command belongs, the processing unit 11 sets that search category as the search category being preferentially processed. On the other hand, when the processing unit 11 determines that the task A is not included in the search category to which the recognition command belongs (NO in S29), the thread priority of the task A is set to (standard value -1), and the processing is performed. The priority is lowered by one level (S33). These processes are repeated until all the tasks during dictionary generation are completed (S34, S35). As a result, the task corresponding to the recognition command recognized from the user's utterance belongs to the search category being preferentially processed, and the speech recognition dictionary necessary for the task is generated preferentially.

Specifically, as shown in FIG. 8B, the connection to the connector 25 of the portable player 26, the editing operation of the music data stored in the storage unit 13, and the wireless connection of the mobile phone 22 to the communication device 21 are performed. When performed in order, generation of a voice recognition dictionary from the music data of the portable player 26 (FIG. 3), generation of a voice recognition dictionary of the music data stored in the storage unit 13, and voice from the phone book data of the mobile phone 22 Generation of a recognition dictionary (FIG. 2) and generation of a voice recognition dictionary from outgoing / incoming history data of the mobile phone 22 are performed in a time-sharing manner. Then, the generated voice recognition dictionary from each and utterance by the user as "Recently calls" during the generation of speech recognition dictionary, the music data of the portable player 26 belonging to the search category of non-search category "Phone Call" (Figure 3) , And the generation priority of the voice recognition dictionary of the music data stored in the storage unit 13 is lowered (standard value −1), the generation of the voice recognition dictionary is temporarily interrupted. Generation of the voice recognition dictionary (FIG. 2) and generation of the voice recognition dictionary from the outgoing / incoming history data of the mobile phone 22 (priority = standard value) are continued in a time-sharing manner. As a result, the generation of the voice recognition dictionary from the phone book data of the mobile phone 22 (FIG. 2) and the generation of the voice recognition dictionary from the outgoing / incoming history data of the mobile phone 22 are relatively prioritized and completed quickly. A list of recent outgoing / incoming histories is displayed on the display unit. As a result, the user can quickly make a call to the other party by speaking the other party who wants to call from the list. As described above, it is possible to shorten the time until the function corresponding to the user's utterance becomes effective among the search functions by voice.
In addition, when the user utters “Recently calls” during the generation of each speech recognition dictionary, if the processing related to the priority setting as shown in FIG. 6B is not performed, specifically, FIG. Processing is performed as shown. When the connection of the portable player 26 to the connector 25, the editing operation of the music data stored in the storage unit 13, and the wireless connection of the mobile phone 22 to the communication device 21 are performed in this order, the music of the portable player 26 first. A speech recognition dictionary is generated from the data (FIG. 3). When this dictionary generation is completed, a voice recognition dictionary of music data stored in the storage unit 13 is generated. When this dictionary generation is completed, a voice recognition dictionary is generated from the telephone directory data of the mobile phone 22 (FIG. 2). When this dictionary generation is completed, a voice recognition dictionary is generated from the outgoing / incoming history data of the mobile phone 22. In this way, the voice recognition dictionary is sequentially generated according to the operation order of the user.

  In addition, it is possible to shorten the time until the function similar to the function responding to the user's utterance becomes effective. In other words, when the user tries to search for music in the portable player 26 but the desired music does not exist, it is assumed that the user searches for music in the storage unit or the CD / DVD player 24 continuously. In addition, when attempting to make a call, first, a case may be assumed where a call history is searched first, and if it is not higher in the list, a search is made in the telephone directory. Even in such a case, the time until the voice search function is activated can be shortened.

  Further, the operation when the dictionary generation is completed will be described with reference to FIG. First, the processing unit 11 deletes the task for which dictionary generation has been completed from the list of tasks for which dictionary generation is in progress (S41). Next, the processing unit 11 determines whether or not the generated task is a task belonging to the search category being preferentially processed (S43). If the task whose generation has been completed is a task belonging to the search category being preferentially processed (YES in S43), the task belonging to the search category being preferentially processed is listed in the list of tasks being generated in the dictionary. It is determined whether it remains (S45). When all the tasks in the dictionary generation are other than the search category being preferentially processed (NO in S45), the processing unit 11 sets the processing priority of each thread to the standard value (S46), and the preferential processing is in progress. The search category is cleared (S47).

  As described above, the dynamic speech recognition dictionary generation method according to the present invention can shorten the time until the function that responds to the user's utterance out of the search function by voice becomes effective, and the user This has the effect of shortening the time until the function similar to the function that responds to the utterance becomes effective, and is useful for generating a dynamic dictionary for speech recognition.

DESCRIPTION OF SYMBOLS 100 Dynamic speech recognition dictionary creation device 11 Processing unit 12 Navigation device 13 Storage unit 14 Display unit 15 Output circuit 16 Speaker 17 Speech input circuit 18 Switch 19 Microphone 20 Speech recognition engine 21 Communication device 22 Mobile phone 23 CD / DVD player 24 CD / DVD
25 Connector 26 Portable player

Claims (5)

Creating a new dictionary generation thread when receiving a voice recognition dictionary generation request from a user;
Adding the task of the dictionary generation thread to a list of tasks during dictionary generation ,
A method for generating a dynamic speech recognition dictionary for generating a speech recognition dictionary necessary for tasks set in the list,
A search categories for classifying the command corresponding to the speech recognition dictionary generation necessary tasks for each object, determining whether there is in the priority processing,
When it is determined that there is the search category being preferentially processed, the speech recognition dictionary necessary for the task belonging to the search category is prioritized over the speech recognition dictionary necessary for the task belonging to the search category not preferentially processing. Generating steps,
When it recognizes a command that requires a dynamic generation of speech recognition dictionary when there is speech of the user, and a step of setting a search category task corresponding to the command belongs to the search category in priority processing A method for generating a dynamic speech recognition dictionary. Deleting the task of generating the speech recognition dictionary from the list of tasks during dictionary generation when the speech recognition dictionary generation is completed;
If the task has been completed of the voice recognition dictionary generation What task der belonging to the search category in the priority processing, there is no remaining task that belongs to the search category in the priority processing in the dictionary generation in the list of tasks The method for generating a dynamic speech recognition dictionary according to claim 1, further comprising the step of canceling that the search category is in a priority process . Thread creation means for creating a new dictionary generation thread when a voice recognition dictionary generation request is received from a user;
And a list addition means for adding the task of the dictionary generation thread to the list of dictionary generation in task,
A dynamic speech recognition dictionary generation device that generates a speech recognition dictionary necessary for tasks set in the list,
A search categories for classifying the command corresponding to the speech recognition dictionary generation necessary tasks for each object, and determining priority processing search categories presence determining means for determining whether or not there is in the priority processing,
When it is determined that there is the search category being preferentially processed, the speech recognition dictionary necessary for the task belonging to the search category is prioritized over the speech recognition dictionary necessary for the task belonging to the search category not preferentially processing. Generating means (S19),
When it recognizes a command that requires a dynamic generation of speech recognition dictionary when there is speech of the user, and means for setting a search category task corresponding to the command belongs to the search category in priority processing Dynamic speech recognition dictionary generation device. When the voice recognition dictionary generated is completed, and deletes dictionary generation task to delete completed tasks of said voice recognition dictionary generated from a list of dictionary generation in task,
If the task has been completed of the voice recognition dictionary generation What task der belonging to the search category in the priority processing, there is no remaining task that belongs to the search category in the priority processing in the dictionary generation in the list of tasks The dynamic speech recognition dictionary generating apparatus according to claim 3 , further comprising: means for canceling that the search category is being preferentially processed . Voice recognition dictionary generation means (S19) for creating a plurality of voice recognition dictionaries;
Necessary for a task corresponding to the command in generating the plurality of speech recognition dictionaries by the speech recognition dictionary generating means when a command is recognized for a task that requires a speech recognition dictionary when a user utters An apparatus for generating a dynamic speech recognition dictionary, comprising: priority control means for prioritizing the speech recognition dictionary over other speech recognition dictionaries.

Images