JP5216337B2 - How to create update data for voice recognition dictionary - Google Patents

Description

  The present invention relates to a technique for updating a speech recognition dictionary.

As a technology for updating the speech recognition dictionary, the speech recognition dictionary for address recognition is composed of multiple partial versioned dictionaries that are hierarchized according to the address hierarchy such as address-prefecture-city-town-chome. Compare the latest voice recognition dictionary with the latest voice recognition dictionary for each partial dictionary, and the previous voice recognition for each partial dictionary so that the version of each partial dictionary and each partial dictionary matches the latest voice recognition dictionary A technique for updating, deleting, or adding the contents of a dictionary is known (for example, Patent Document 1).
Japanese Patent Laid-Open No. 2004-13108

  According to the technology described in Patent Document 1, update, deletion, addition necessity determination for each partial dictionary and update, deletion, addition processing according to the necessity determination result for each partial dictionary at the time of updating the conventional speech recognition dictionary Therefore, the process becomes complicated. Even if it is necessary to update only a part of the words in the partial dictionary, it is necessary to update all of the partial dictionaries. Therefore, it is not always necessary to reduce the amount of data necessary for updating the previous speech recognition dictionary. It cannot be reduced sufficiently.

  Therefore, an object of the present invention is to make it possible to easily update the speech recognition dictionary while further reducing the amount of data necessary for updating the speech recognition dictionary.

  In order to achieve the above object, the present invention provides an object information table in which object information, which is information on an object represented by a phrase that is a target of speech recognition, and a reading of each phrase that is a target of speech recognition. As a method for creating update data for updating a speech recognition dictionary composed of a reading dictionary registered with identification information for identifying the object information on the object information table of the object represented by the word or phrase When updating the speech recognition dictionary, the object information table in which the object information of the object represented by the phrase newly added to the object of speech recognition this time is registered is created, and the update used for the current update The step of including in the data, and when updating the speech recognition dictionary, the reading of each word or phrase that is subject to speech recognition in the updated speech recognition dictionary this time, A method for creating update data for a speech recognition dictionary, comprising: creating a smear dictionary registered together with identification information for identifying object information on the object information table formed, and including it in the update data used for the current update I will provide a.

Here, in such an update data creation method of the speech recognition dictionary, the word / phrase to be voice-recognized may be a place name or a facility name, and the object information is at least a word / phrase to be voice-recognized. The location name represented by or the location information of the facility may be included.
According to such a speech recognition dictionary update data creation method, the update object used for updating the speech recognition dictionary includes, as object information, a target object represented by a phrase newly added to the target of speech recognition this time. It only needs to include information. Therefore, the amount of update data can be kept small.

  Further, according to such a method for creating update data of a speech recognition dictionary, the speech recognition dictionary stored in the storage device in the speech recognition device that performs speech recognition using the speech recognition dictionary stored in the storage device The update is performed by replacing the reading dictionary stored in the storage device with the reading dictionary included in the update data, and the object information table included in the update data is stored in the storage device. It is possible to carry out by a simple process of simply executing the step of additionally storing in the storage device while maintaining the target object information table.

  Note that the voice recognition in the voice recognition apparatus is performed by, for example, identifying a reading represented by the input voice using a reading dictionary of the voice recognition dictionary stored in the storage device, and the reading dictionary together with the identified reading. Is performed by outputting the object information of any object information table stored in the storage device indicated by the identification information registered in

  As described above, according to the present invention, it is possible to easily update the voice recognition dictionary while further reducing the amount of data necessary for updating the voice recognition dictionary.

Hereinafter, an embodiment of the present invention will be described by taking application to point recognition by voice recognition in a navigation system mounted on an automobile as an example.
FIG. 1 shows a configuration of a speech recognition dictionary update data creation system according to the present embodiment.
As shown in the figure, the speech recognition dictionary update data creation system includes an update data creation device 1 and a recording device 2.
The update data creation device 1 is a device that creates update data for a speech recognition dictionary that includes a reading dictionary and one or more key data.
Then, the update data creation device 1 creates the update data for the current {n-th time} as follows.
That is, the speech recognition source data (n-1) input to the speech recognition dictionary update data creation system and recorded in the recording device 2 at the time of the previous {(n-1) th} update data creation, and the previous {(n-1) )}} Update data created at the time of creation of the update data and recorded in the recording device 2, and this time the {n-th}, voice input to the speech recognition dictionary update data creation system and recorded in the recording device 2 From the recognition source data (n), a reading dictionary (n), key data (n), and an assignment table (n) are created and recorded in the recording device 2. The recorded reading dictionary (n) and key data (n) are used as update data (n), and are provided to the navigation system via an appropriate communication medium or storage medium.

In addition, the speech recognition source data (n) and the assignment table (n) recorded in the recording device 2 this time {n-th} are set to n = n + 1 to create update data for the next {n-th}. The voice recognition source data (n-1) and the assignment table (n-1) are used.
However, the first update data is created as follows.
That is, this time {first time}, from the voice recognition source data (1) input to the recording device 2 and recorded in the recording device 2 and input to the voice recognition dictionary update data creation system, Key data (1) and assignment table (1) are created and recorded in the recording device 2. The recorded reading dictionary (1) and key data (1) are output as update data (1).

Hereinafter, the details of the update data creation processing of the update data creation device 1 performed in this way will be described.
FIG. 2 shows a procedure of update data creation processing performed by the update data creation device 1.
As shown in the figure, in this process, first, it is determined whether or not the current (n-th) update data creation is the first update data creation (step 202). For example, this time (first time), key data (1) is created from the voice recognition source data (1) input to the voice recognition dictionary update data creation system and recorded in the recording device 2 (step 204).

  Here, as shown in FIG. 3a, the voice recognition source data has an entry provided for each of all points targeted for voice recognition, and each entry includes a point such as “Iwaki City Hall”. A display character string indicating the name of the character, a reading indicating the name of the point, for example, a reading such as “Iwaki Yakusho”, and point information of the point are registered. Further, in the point information of the point, for example, the coordinates of the point and an address code representing the address of the point are registered.

Next, the key data (1) is created from the voice recognition source data (1) as follows.
That is, as shown in FIG. 3c, an entry corresponding to each entry of the voice recognition source data (1) is created in the key data (1), and a voice recognition corresponding to each entry of the created key data (1). The point information of the entry of the original data (1) and the pronunciation information representing the pronunciation of the reading of the entry of the corresponding voice recognition source data (1) are registered.

Returning to FIG. 2, if the key data (1) is created in this way (step 204), the reading dictionary (1) is created as follows (step 206).
That is, as shown in FIG. 3b, an entry corresponding to each entry of the speech recognition source data (1) is created in the Yomi dictionary (1), and a word that is a unique identifier is created in the created Yomi dictionary (1) entry. ID, reading of the entry of the corresponding voice recognition source data (1), and a key pointer indicating the position of the entry of the key data (1) created corresponding to the entry of the corresponding voice recognition source data (1) sign up.

  Then, when the reading dictionary (1) is created, an assignment table (1) having entries corresponding to the respective entries of the created reading dictionary (1) is created. As shown in FIG. 3d, the assignment table (1) is created. In each entry, the reading and key pointer registered in the entry in the corresponding reading dictionary (1) are registered (step 208).

Then, the first update data creation process ends.
On the other hand, when it is determined in step 202 that the current (n-th) update data creation is not the first update data creation, first, the current {n-th} update data creation system is updated. The voice recognition source data (n) input and recorded in the recording device 2 and the voice recognition source data previously input to the voice recognition dictionary update data creation system and recorded in the recording device 2 ({(n-1))}. Difference data from (n-1) is created (step 210).

  Here, the voice recognition source data (n) is generated after the change of the target point of the voice recognition after the voice recognition source data (n-1) is created or after the voice recognition source data (n-1) is created. The name of the point and the change of the point information are reflected, and the voice recognition source data (n) is a display of the name of the corresponding point in the entry of the voice recognition source data (n-1). Only entries that have no change in character string, reading, or point information are included as they are. That is, the voice recognition source data (n) includes a display character string of the name of the corresponding point after the voice recognition source data (n-1) of the entries of the voice recognition source data (n-1) is created. Entry with a change in reading or spot information, entry with no corresponding point after creation of voice recognition source data (n-1) (for example, entry of a closed store), voice recognition source data After the creation of (n-1), an entry for a point excluded from the target of speech recognition is not included.

  In addition, the voice recognition source data (n) is the name of the point after the change for the point where the display character string of the name of the point, the reading or the point information is changed after the voice recognition source data (n-1) is created. An entry for registering the display character string, reading, and spot information, and a new entry for a spot added as a voice recognition target after the voice recognition source data (n-1) is created. ing.

In step 210, the voice recognition source data (n) is included in the voice recognition source data (n) and is not included in the voice recognition source data (n-1) (added by the voice recognition source data (n). B) Extract entries to make difference data.
That is, for example, as shown in FIG. 4a, the speech recognition source data (n-1) has entries in which display character strings a11, a12, a13, and a14 are registered, and the speech recognition source data as shown in FIG. 4b. When (n) has an entry in which display character strings of a11, a13, a21, and a22 are registered, it is included in the speech recognition source data (n) and not included in the speech recognition source data (n-1). Then, the entry in which the display character strings a21 and a22 of the voice recognition source data (n) are registered is extracted and used as difference data as shown in FIG. 4c.

Returning to FIG. 2, if difference data is created in step 210, next, key data (n) is created from the difference data (step 212).
Here, the key data (n) in step 212 is created by creating an entry corresponding to each entry of the difference data in the key data (n) as shown in FIG. Each entry is registered by registering point information of the entry of the corresponding difference data and pronunciation information representing the pronunciation of the reading of the entry of the corresponding difference data.

Returning to FIG. 2, if the key data (n) is created in this way (step 212), the reading dictionary (n) is created as follows (step 214).
That is, as shown in FIG. 4e, an entry corresponding to each entry of the voice recognition source data (n) is created in the yomi dictionary (n), and a word that is a unique identifier is created in the entry of the created yomi dictionary (n). The ID and the entry of the corresponding voice recognition source data (n) are registered.
Then, for each entry in the reading dictionary (n), a key pointer is registered as follows. That is, if the entry of the speech recognition source data (n) corresponding to the entry in the reading dictionary (n) is an entry that has not been extracted into the difference data, the reading registered in the entry in the reading dictionary (n) The key pointer registered in the entry of the assignment table (n-1) in which the matching reading is registered is registered as the key pointer in the entry of the reading dictionary (n).

  On the other hand, if the entry of the speech recognition source data (n) corresponding to the entry in the reading dictionary (n) is an entry extracted as difference data, the difference data corresponding to the entry of the corresponding speech recognition source data (n) The key pointer indicating the position of the entry of the key data (n) created corresponding to the entry is registered.

  As a result, for example, when a reading dictionary (n) having entries in which readings of b11, b13, b21, and b22 are registered is created from the voice recognition source data (n) shown in FIG. As shown in FIG. 4c, the assignment table (n-1) has entries in which readings b11, b12, b13, and b14 are registered. As shown in FIG. As shown in FIG. 4e, the entry in which b11 and b13 in the reading dictionary (n) are registered is registered in the entry in which b11 and b13 in the assignment table (n-1) are registered. The registered key pointers p11 and p13 are registered.

  Here, the key pointers p11 and p13 are keys having entries in which point information and pronunciation information of points having b11 and b13 are created in any of the previous times ((n−x) times). This indicates the position of the entry in which the point information and pronunciation information of the points having b11 and b13 in the data (nx) are registered.

  The entry in which the readings b21 and b22 in the reading dictionary (n) are registered corresponds to the entry of the difference data corresponding to the entry in which the readings b21 and b22 in the speech recognition source data (n) are registered. A key pointer indicating the position of the entry of the created key data (n) is registered.

  Returning to FIG. 2, if the reading dictionary (n) is created in this way, an assignment table (n) having entries corresponding to the entries of the created reading dictionary (n) is created, and is shown in FIG. 4g. As described above, the reading and key pointer registered in the entry of the corresponding reading dictionary (n) are registered in each entry of the assignment table (n).

Then, the nth update data creation process is terminated.
The update data creation process performed by the update data creation device 1 has been described above.
As described above, according to the present embodiment, only the point information and pronunciation information of the point corresponding to the newly added voice recognition target this time are registered as the key data in the update data used for updating the voice recognition dictionary. The key data should be included. Here, since the key data has a larger data amount per point than the Yomi dictionary, the data amount of the update data can be effectively suppressed small in this way.

Next, a point recognition operation by voice recognition in the navigation system using the update data created in this way will be described.
FIG. 5 shows a configuration of the navigation system according to the present embodiment.
As illustrated, the navigation system includes a voice recognition dictionary update processing unit 11, a voice recognition dictionary storage unit 12, a microphone 13, a voice recognition unit 14, a voice generation unit 15, a speaker 16, a navigation device 17, a display device 18, and an input device. 19, an address database 20 and a map database 21 are provided.

  In such a configuration, when the speech recognition dictionary update processing unit 11 is provided with update data via an appropriate communication medium or storage medium from the speech recognition dictionary update data creation system shown in FIG. The included dictionary is recorded in the speech recognition dictionary storage unit 12. At this time, if a reading dictionary is already stored in the speech recognition dictionary storage unit 12, it is replaced with the reading dictionary included in the update data.

  When the update data is provided, the voice recognition dictionary update processing unit 11 records key data included in the update data in the voice recognition dictionary storage unit 12. At this time, if the key data is already stored in the speech recognition dictionary storage unit 12, the already stored key data is maintained as it is, and the key data included in the update data is additionally stored.

  As a result, at the point of time when the provided n-th update data reading dictionary (n) and key data (n) are stored in the speech recognition dictionary storage unit 12, the speech recognition dictionary storage unit 12 stores the data shown in FIG. Thus, one reading dictionary (n) and n key data from key data (1) to key data (n) are stored as a voice recognition dictionary.

As described above, according to the present embodiment, the update of the speech recognition dictionary using the update data is included in the update data and the replacement with the read dictionary included in the update data of the read dictionary stored in the storage device. It is realized by a simple process of only storing additional key data in the storage device.
Note that the voice recognition dictionary update processing unit 11 is configured such that when the reading dictionary storage unit 12 has already stored a reading dictionary, the provided update data is already stored in the reading recognition dictionary storage unit 12. If it does not include a newer reading dictionary, the provided update data is discarded, and the reading dictionary and the key data included in the updating data are not stored in the voice recognition dictionary storage unit 12, and the voice recognition dictionary is stored. The storage state of the reading dictionary and key data in the storage unit 12 is maintained as it is.

Next, when the navigation device 17 is requested to input a place name by the user via the input device 19, the navigation device 17 requests the voice recognition unit 14 to execute the place name recognition process.
The speech recognition unit 14 requested to execute the place name recognition process identifies the entry of the reading dictionary (n) in which the reading that best matches the voice input from the microphone 13 is registered with reference to the reading dictionary (n). Then, the key pointer registered in the entry of the specified reading dictionary (n) is read out. Then, the point information and the pronunciation information registered in the key data entry of any one of the key data (1) to the key data (n) in which the read key pointer indicates the position are read, and the read pronunciation information is The voice generation unit 15 is caused to generate the voice to be shown and output to the speaker 16. As a result, the speaker 16 outputs (talks back) the voice that reads out the reading that best matches the voice input from the microphone 13 recognized by the voice recognition unit 14.

After that, when an approval of recognition content is input from the user via the input device 19, the navigation device 17 notifies the approval to the voice recognition unit 14, and the voice recognition unit 14 that has received the notification The point information read out is output to the navigation device 17.
The navigation device 17 that has received the spot information identifies the coordinates and address of the spot indicated by the received spot information while referring to the address database 20 and the map database 21, and uses the identified coordinates and address as the coordinates of the user-designated spot. As the address, an image representing the designated point on the map is displayed on the display device 18, or the designated point is set as the destination.

The embodiment of the present invention has been described above.
In the above embodiment, as the point information to be included in the voice recognition source data and key data, the coordinates of the point and the address code representing the address of the point are registered, but as the point information, in addition to the coordinates of the point, The navigation device 17 may include a mark identifier for representing the point on the map, coordinates of the arrangement position of the mark, and the like. In this case, the navigation device 17 uses the coordinates and marks of the points indicated by the received point information as the coordinates of the user-designated points and marks representing the designated points, and sets the designated points as the destination, or the voice recognition unit 14. The mark specified by the point information received from the display device is highlighted on the map displayed on the display device 18.

  Moreover, since all the information of the assignment table (i) used in the above embodiment is included in the Yomi dictionary (i), the Yomi dictionary (i) is directly used without creating the assignment table (i). Update data may be created. In other words, in the above update data creation process, the assignment table (n) in steps 208 and 216 is not created, and in step 214, the reading dictionary (n-1) is substituted for the assignment table (n-1). ) May be used to create the reading dictionary (n).

It is a block diagram which shows the structure of the speech recognition dictionary update data creation system which concerns on embodiment of this invention. It is a flowchart which shows the update data creation process which concerns on embodiment of this invention. It is a figure which shows the process example of the update data creation process which concerns on embodiment of this invention. It is a figure which shows the process example of the update data creation process which concerns on embodiment of this invention. It is a block diagram which shows the structure of the navigation system which concerns on embodiment of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Update data preparation apparatus, 2 ... Recording apparatus, 11 ... Voice recognition dictionary update process part, 12 ... Voice recognition dictionary memory | storage part, 13 ... Microphone, 14 ... Voice recognition part, 15 ... Voice generation part, 16 ... Speaker, 17 ... navigation device, 18 ... display device, 19 ... input device, 20 ... address database, 21 ... map database.

Claims (5)

An object information table in which object information that is information on an object represented by a word to be recognized by speech recognition is registered;
Speech recognition comprising the reading of each word or phrase that is subject to speech recognition and a reading dictionary registered with identification information for identifying the object information on the object information table of the object represented by the word or phrase An update data creation method for updating a dictionary,
When updating the voice recognition dictionary, the update data used for the current update is created by creating a target information table in which the target information of the target represented by the words newly added to the target of voice recognition is registered. Steps to include,
When updating the speech recognition dictionary, the phrase represents a reading of each word including a word newly added to the speech recognition target this time as a target of speech recognition in the updated speech recognition dictionary this time For readings in which the object information of the object is registered in the object information table created this time, for identifying the object information of the object represented by the phrase on the object information table created this time For the readings that are registered with the identification information and that are registered in the object information table in which the object information of the object represented by the word is created in the past, the word on the object information table created in the past is voice create a readings dictionary registered together with the identification information for identifying the object information of the object, characterized in that a step to be included in the update data to be used for this update representing Update data creation method of 識辞 statement. An update data creation method for a speech recognition dictionary according to claim 1,
The speech recognition target word or phrase is a place name or facility name, and the object information includes at least a place name or facility location information represented by the speech recognition target word or phrase. Dictionary update data creation method. The speech recognition dictionary which performs speech recognition using the speech recognition dictionary memorize | stored in the memory | storage device, The update data creation method of the speech recognition dictionary of Claim 1 or 2 as the speech recognition dictionary memorize | stored in the said memory | storage device A speech recognition dictionary update method for updating using update data created by
Replacing the pre-update Yomi dictionary stored in the storage device with a Yomi dictionary included in the update data;
And further storing the object information table included in the update data in the storage device while maintaining the object information table stored in the storage device. Update method. A method for updating a speech recognition dictionary according to claim 3,
The voice recognition device uses the reading dictionary of the voice recognition dictionary stored in the storage device to identify the reading represented by the input voice, and the identification information registered in the reading dictionary together with the identified reading indicates, A method for updating a speech recognition dictionary, comprising: outputting object information stored in any of the object information tables stored in the storage device. A speech recognition dictionary update system for updating a speech recognition dictionary used by a speech recognition apparatus for speech recognition,
The speech recognition dictionary includes an object information table in which object information that is information on an object represented by a word that is a target of speech recognition is registered, and a reading of each word that is a target of the speech recognition. The object consists of a Yomi dictionary registered with identification information for identifying the object information on the object information table,
The speech recognition dictionary update system
A speech recognition device that performs speech recognition using the speech recognition dictionary;
An update data creation device for creating update data used for updating the speech recognition dictionary;
The update data creation device includes:
When updating the speech recognition dictionary, the object information table in which the object information that is the object information represented by the words newly added to the object of speech recognition is registered is created, and the updated information The target object information of the target object represented by the phrase is read in the target object information table created this time , as the target of voice recognition in the voice recognition dictionary, and the reading of each word including the phrase newly added to the target of voice recognition this time. The registered reading is registered together with identification information for identifying the object information of the object represented by the word on the object information table created this time, and the object information of the object represented by the word Is read in the object information table created in the past, the identification for identifying the object information of the object represented by the word on the object information table created in the past. Create a readings dictionary registered together with information, having the updated data creation means for creating the update data comprising more the object information table and readings dictionary created,
The voice recognition device
A storage device storing the speech recognition dictionary;
The yomi dictionary included in the update data replaces the pre-update yomi dictionary stored in the storage device, and the object information table included in the update data is replaced with the object information stored in the storage device. A voice dictionary updating means for additionally storing in the storage device while maintaining the table;
Using the reading dictionary of the voice recognition dictionary stored in the storage device, the reading represented by the input voice is identified, and the identification information registered in the reading dictionary together with the identified reading is stored in the storage device. A speech recognition dictionary updating system, comprising: speech recognition means for outputting the object information of any object information table.