JP5426913B2 - Speech recognition dictionary editing device and speech recognition device - Google Patents

Description

  The present invention relates to a technique for editing a speech recognition dictionary used for speech recognition in a speech recognition apparatus.

  For speech recognition, a speech recognition dictionary that stores speech recognition data in which text and pronunciation data of the text are registered is prepared in advance for each text to be recognized, and pronunciation data that matches the speech uttered by the user is registered. Generally, this is performed by using a text registered in the voice recognition data being recognized as a recognized text (for example, Patent Document 1).

  In addition, as a technology for creating such a speech recognition dictionary, for each text included in a specified text list, the corresponding text generated using a text-to-speech (TTS) technology is uttered. There is also known a technique for creating speech recognition data in which pronunciation data is registered together with the text, and creating a speech recognition dictionary storing speech recognition data created for each text (for example, Patent Document 2).

  In addition, as a technique for editing such a speech recognition dictionary, there is also known a technique for correcting pronunciation data registered in speech recognition data of text designated by the user to pronunciation data that matches the voice spoken by the user. (For example, Patent Document 3).

JP 2008-158511 A JP 2004-53979 A JP 2007-248886 A

  Now, if it does not have a function to edit some of the speech recognition data registered in the speech recognition dictionary, editing such as correction of pronunciation data registered in the speech recognition dictionary of text specified by the user, Since it is necessary to recreate the entire speech recognition dictionary, it has been impossible to efficiently edit speech recognition data.

  Therefore, the present invention is a function for editing a part of speech recognition data registered in the speech recognition dictionary stored for each text to be recognized, with the speech recognition data in which the text and the pronunciation data of the text are registered. It is an object of the present invention to make it possible to edit some of the voice recognition data more efficiently even when not provided.

  To achieve the above object, the present invention stores speech recognition data representing correspondence between a text and pronunciation data representing the pronunciation of the text for each text to be recognized, and represents the text represented by speech uttered by a human being. A speech recognition dictionary editing device that edits a speech recognition dictionary used in a speech recognition device that recognizes a speech recognition dictionary storage unit that stores the speech recognition dictionary, and a yomi conversion rule that defines a method for generating pronunciation data from text The stored Yomi conversion rule storage unit and each of the texts to be recognized are grouped into groups of n (where n is an integer of 1 or more) texts, and for each group, each text in the group The speech recognition dictionary storing speech recognition data representing correspondence with pronunciation data generated according to the reading conversion rule of the text A speech recognition dictionary generation unit that is created as a group speech recognition dictionary and stored in the speech recognition dictionary storage unit, and a generation method in which pronunciation data generation method for a specified text that is a text specified by the user is specified by the user The Yomi conversion rule correction unit for correcting the Yomi conversion rule, and the group including the specified text as the correction target group, the voice recognition dictionary of the correction target group is deleted from the voice recognition dictionary storage unit In addition, the speech recognition dictionary storing speech recognition data representing correspondence between each text in the correction target group and pronunciation data generated according to the reading conversion rule corrected by the reading conversion rule correction unit of the text, Created as a new voice recognition dictionary for the group and stored in the voice recognition dictionary storage unit That is obtained by a voice recognition dictionary modification unit.

  Here, in such a speech recognition dictionary editing device, in the speech recognition dictionary correction unit, each group including the same text as the designated text is set as the correction target group, and the correction target is set for each correction target group. Correspondence between each text in the correction target group and pronunciation data generated in accordance with the Yomi conversion rule after correction by the Yomi conversion rule correction section of the text as well as deleting the voice recognition dictionary of the group from the voice recognition dictionary storage unit The voice recognition dictionary storing the voice recognition data representing can be created as a new voice recognition dictionary of the group and stored in the voice recognition dictionary storage unit.

  Further, in such a speech recognition dictionary editing device, when the text to be recognized is added, the added text to be recognized is grouped into groups of n texts, and each group The speech recognition dictionary storing speech recognition data representing correspondence between each text in the group and pronunciation data generated according to the reading conversion rule of the text is created as the speech recognition dictionary of the group, and the speech A speech recognition dictionary adding unit for additionally storing in the recognition dictionary storage unit may be provided.

  In addition, when the speech recognition dictionary correction unit causes the text to be excluded from the text to be recognized, the speech recognition dictionary editing device corrects the group including the excluded text. As a group, the speech recognition dictionary of the correction target group is deleted from the speech recognition dictionary storage unit, and after the excluded text is excluded from the correction target group, each text of the correction target group and the text The speech recognition dictionary storing speech recognition data representing correspondence with pronunciation data generated according to the reading conversion rule is created as a new speech recognition dictionary of the group, and stored in the speech recognition dictionary storage unit. Also good.

Here, the text to be recognized may be, for example, any one of a song title, artist name, album name, and genre name.
According to these speech recognition dictionary editing devices, each text to be recognized is grouped into groups of n (where n is an integer of 1 or more) texts, and a speech recognition dictionary is created for each group. In addition, the voice recognition data that represents the correspondence between the text and the pronunciation data representing the pronunciation of the text can be corrected or deleted by re-creating only the voice recognition dictionary that contains the voice recognition data that needs to be corrected or deleted Therefore, a single speech recognition dictionary storing speech recognition data for all texts to be recognized is provided, and without a function of editing some speech recognition data registered in the speech recognition dictionary, By re-creating the single speech recognition dictionary, it is possible to edit speech recognition data more efficiently than when editing some speech recognition data. That.

The speech recognition dictionary editing apparatus and the speech recognition unit for recognizing text represented by speech uttered by a person using the speech recognition dictionary of each group stored in the speech recognition dictionary storage unit. A speech recognition apparatus may be configured.
Further, an audio playback device is configured by such a voice recognition device, a music data storage unit that stores music data representing music, and a music playback unit that plays back the music data stored in the music data storage unit. You may make it do. In this case, the text to be recognized is the song name of the song represented by the song data stored in the song data storage unit, and the song reproduction unit recognizes the song with the song name recognized by the voice recognition unit. You may make it reproduce | regenerate music data.

  As described above, according to the present invention, a part of the speech recognition data registered in the speech recognition dictionary stored for each text to be recognized is the speech recognition data in which the text and the pronunciation data of the text are registered. When a function for editing is not provided, it is possible to efficiently edit some of the voice recognition data.

It is a block diagram which shows the structure of the audio reproduction apparatus which concerns on embodiment of this invention. It is a figure which shows the recorded music information which concerns on embodiment of this invention. It is a flowchart which shows the music DB creation process for HDD recorded music and the voice recognition dictionary creation process for HDD recorded music which concern on embodiment of this invention. It is a figure which shows the process example of the music DB creation process for HDD recorded music which concerns on embodiment of this invention, and the voice recognition dictionary creation process for HDD recorded music. It is a flowchart which shows the music DB creation process for connection device recording music and the speech recognition dictionary creation process for connection device recording music which concern on embodiment of this invention. It is a figure which shows the process example of the music DB creation process for connection device recording music which concerns on embodiment of this invention, and the speech recognition dictionary creation process for connection device recording music. It is a flowchart which shows the speech recognition dictionary edit process and speech recognition dictionary correction process which concern on embodiment of this invention. It is a figure which shows the example of a display of the audio reproduction apparatus which concerns on embodiment of this invention. It is a figure which shows the process example of the speech recognition dictionary edit process and speech recognition dictionary correction process which concern on embodiment of this invention.

Hereinafter, embodiments of the present invention will be described by taking application to an audio playback device as an example.
FIG. 1 shows the configuration of an audio playback apparatus according to this embodiment.
As shown in the figure, the audio playback device includes a microphone 1, a speaker 2, an input device 3, a display device 4, an HDD 5, a CD drive 6, a portable audio player interface 7, a speech synthesis engine 8, a speech recognition engine 9, and an audio output unit 10. , A voice recognition dictionary editing unit 11, a ripping processing unit 12, a music management unit 13, and a reproduction control unit 14.

  However, such an audio playback device may be configured using a computer having a general configuration including a microprocessor, a memory, and other peripheral devices in terms of hardware. In this case, each part of the speech synthesis engine 8, speech recognition engine 9, audio output unit 10, speech recognition dictionary editing unit 11, ripping processing unit 12, music management unit 13, and playback control unit 14 of the audio playback device described above or A part of the process may be realized as a process realized by a microprocessor executing a program stored in the HDD 5 in advance.

Now, the CD drive 6 reads out the music data recorded on the loaded CD-DA disc 20.
A portable audio player 21 is selectively connected to the portable audio player interface 7. The portable audio player 21 is an apparatus that records an audio file of a music piece and has a function of playing back the music piece of the recorded audio file. The portable audio player 21 also records the names of albums to which the music belongs, the genre name of the music, the artist name of the music, the music name (title name) of the music, etc. Attribute information, and when connected to the portable audio player interface 7, in response to a request for transfer of music management information from the audio playback device, the managed attribute information and the audio of each music It has a function of outputting the file identifier to the audio playback device as music management information. In addition, when connected to the portable audio player interface 7, in response to a playback request specifying an audio file identifier from the audio playback device, the audio file specified by the audio file identifier is played back, and the played back signal / Has the function of outputting data to an audio playback device.

The HDD 5 stores an audio file storing CDDB, reading conversion rules, HDD recorded music information, connected device recorded music information, and audio data of the music.
In the CDDB, the name of the album to which the music belongs, the genre name of the music, the artist name of the music, and the music of each music recorded on the CD-DA disc 20 distributed in the market are stored. Attribute information such as the song name (title name) is stored.
Further, the rule for converting the reading is stored in the speech synthesis engine 8 so as to define how to generate reading data representing the pronunciation of the text read out from the various texts.
Further, the HDD recorded music information and the connected device recorded music information have the same configuration, and each includes a music DB and a voice recognition dictionary. However, the music DB of the HDD recorded music information is created with the music recorded in the audio file in the HDD 5 as the target music, and the music DB of the connected device recorded music information is the portable audio player connected to the portable audio player interface 7. The music recorded in the player 21 is created as the target music. The voice recognition dictionary of HDD recorded music information is provided corresponding to the music DB of HDD recorded music information, and the voice recognition dictionary of connected device recorded music information is provided corresponding to the music DB of connected device recorded music information. It is done.

Hereinafter, the configuration of the music DB and the speech recognition dictionary will be described.
First, as shown in FIG. 2a, the music DB has an entry (each row in the figure) provided corresponding to the music for each target music, and each entry in the music table of the corresponding target music. The song ID that is the identifier of the song, the song name of the corresponding target song, the artist name of the artist of the corresponding target song, the album name of the album of the corresponding target song, the genre name of the genre of the corresponding target song, and the target A file identifier representing the identifier of the audio file of the music is registered. Here, each entry in the music DB is arranged so that the one with the smaller music ID comes in front.

Next, as shown in FIG. 2b, a voice recognition dictionary is created corresponding to each of n consecutive music IDs (five in the figure), and different dictionary IDs are assigned consecutive numbers to each voice recognition dictionary. Has been.
Each voice recognition dictionary has an entry corresponding to each of the corresponding n music IDs, and each entry includes an entry of a corresponding music ID and a corresponding music ID registered in the corresponding music DB. Yomi data representing the pronunciation of the song name of the song registered in is registered. Here, the format of the reading data may be arbitrary, but basically has contents equivalent to a phonetic symbol string.

Hereinafter, an operation of creating such HDD recorded music information and connected device recorded music information will be described.
First, an operation for creating HDD recorded music information will be described.
When the ripping processing unit 12 receives a user's ripping instruction via the input device 3, the ripping processing unit 12 reads and encodes the music data of each music recorded on the CD-DA disc 20 mounted on the CD drive 6, Every time, it is recorded in the HDD 5 as an audio file. At this time, the CDDB is referred to, and the song name, artist name, album name, genre name, etc. of the audio file recorded in the HDD 5 are calculated and stored in the audio file as the attribute information of the audio file song. To do. Then, the music management unit 13 is notified of the occurrence of ripping of a new music.

  On the other hand, if the music management unit 13 detects a notification of ripping occurrence of a new music from the ripping processing unit 12 in the HDD recorded music DB creation process shown in FIG. The music ID of the last music registered in the DB is acquired and is set to m (step 304). If no music is registered in the music DB of the HDD recorded music information, m = 0 is set.

Then, let “floor” be a floor function that returns the largest integer less than or equal to x for a real number x.
k = floor ((m-1) / n)
p = 1 + n + (k × n)
p is obtained (step 306). However, n is the number of music IDs corresponding to one voice recognition dictionary.

  Then, a new entry corresponding to each audio file newly recorded in the HDD 5 by the ripping processor 12 is created in the song DB of the HDD recorded song information, and a song ID that increases by 1 from p is added to each entry. At the same time as registration, the song name, artist name, album name, genre name, and audio file identifier represented by the attribute information of the corresponding audio file are registered (step 308). That is, the r-th entry newly created in the song DB of the HDD-recorded song information includes the song ID of p + r−1 and the r-th audio file newly recorded on the HDD 5 by the ripping processor 12. Register attribute information and audio file identifier.

Then, the voice recognition dictionary editing unit 11 is requested to create a voice recognition dictionary of HDD recorded music information (step 310).
Next, the voice recognition dictionary editing unit 11 requested to create the voice recognition dictionary of the HDD recorded music information starts the HDD recorded music voice recognition dictionary creation process shown in FIG. Then, a dictionary source list is created in which pairs of song IDs and song names extracted from the entries whose song IDs are p and later are registered in the order of song IDs (step 352).

  Then, a divided dictionary source list is created by dividing the dictionary source list into n pieces of music in the order of the music IDs, and each of the created divided dictionary source lists starts from 1 in ascending order of the range of registered music IDs. A division number that is incremented by 1 is given (step 354). That is, t is given as the division number to the divided dictionary source list having the t-th largest music ID registered at the top.

Then, for each of the divided dictionary source lists, a voice recognition dictionary is created and stored in the HDD recorded music information (step 356). In addition, the speech recognition dictionary created from the divided dictionary original list with the division number t includes
t-1 + floor ((p + n-1) / n) is assigned as a dictionary ID.

  Here, the creation of the speech recognition dictionary from the divided dictionary source list is performed by requesting the speech synthesis engine 8 to create a speech recognition dictionary based on the divided dictionary source list. When the speech synthesis engine 8 is requested to create a speech recognition dictionary based on the divided dictionary source list, the pronunciation of the song name generated in accordance with the yomi conversion rule stored in the HDD 5 is generated for each song name in the divided dictionary source list. A speech recognition dictionary replaced with reading data representing “” is created and stored in the HDD 5.

Here, a processing example of such an HDD recorded music DB creation process and an HDD recorded music voice recognition dictionary creation process will be described.
Now, assuming that the number of songs n corresponding to one voice recognition dictionary is 5, as shown by 401 in FIG. 4, 12 songs with song IDs 1 to 12 assigned to the song DB of the HDD recorded song information. The voice recognition dictionary of the dictionary ID1 in which the Yomi data of the song names of the songs whose song IDs are 1 to 5 are registered in association with the song IDs, and the Yomi data of the song names of the songs whose song IDs are 6 to 10 are the songs. The HDD-recognized music information includes the voice recognition dictionary of the dictionary ID2 registered in association with the ID, and the voice recognition dictionary of the dictionary ID3 in which the Yomi data of the music titles of the music IDs 11 to 12 are registered in association with the music ID. It shall be included in

In this state, it is assumed that audio files of seven music pieces are newly recorded on the HDD 5 by the ripping processing unit 12.
In this case, first, the music ID of the final music registered in the music DB by the music DB creation process for HDD recording music is 12, so m = 12.
k = floor ((m-1) / n)), p = 1 + n + (k × n)
p = 16 is obtained (steps 304 and 306), and as indicated by the arrow 402, seven entries having the song IDs 16 to 22 are newly created in the song DB, and 7 newly recorded in the HDD 5 in each entry. Each piece of audio file attribute information and audio file identifier is registered (step 308).

  Then, by the HDD recorded music voice recognition dictionary creation process, as shown by an arrow 403, a set of the music ID and the music title extracted from each entry of the music ID 16 or later in the music DB of the HDD recorded music information is shown. Then, a dictionary source list registered in the order of the music ID is created (step 352).

  Then, as indicated by an arrow 404, the dictionary source list is divided into five music pieces in the order of the music IDs, and the divided dictionary source list of division number 1 in which the music titles 16 to 20 are registered, A division dictionary source list of division number 2 in which the song names with IDs 21 to 22 are registered is created (step 354).

  Then, as shown by an arrow 405, a voice recognition dictionary is created from the divided dictionary original list of division number 1 and the melody data of the song names 16 to 20 are registered in association with the song ID, and the division of division number 2 is created. From the dictionary source list, a speech recognition dictionary is created in which the reading data of song names 21 to 22 are registered in association with the song IDs (step 356).

  Here, since p = 16, the speech recognition dictionary created from the divided dictionary source list with the division number t is 1 has a dictionary ID according to t-1 + floor ((p + n-1) / n). 4 is assigned to the speech recognition dictionary created from the divided dictionary source list having the division number t of 2, and 5 is assigned as the dictionary ID according to t-1 + floor ((p + n-1) / n). Is done.

As a result, for every five songs of the audio file newly recorded by the ripping process, the voice recognition dictionary of the song title of the song is given a dictionary ID following the existing voice recognition dictionary in the HDD recorded song information. It has been added.
Next, the operation of creating connected device recorded music information will be described.
The music management unit 13 monitors the connection of the portable audio player 21 to the portable audio player interface 7 in the connection device recording music DB creation process shown in FIG. 5A (step 502), and the connection of the portable audio player 21 occurs. If so, whether or not the connected portable audio player 21 is the same portable audio player 21 as the portable audio player 21 that detected the previous connection is checked based on the identification information of the portable audio player 21 (step 504). If it is not the same as that of the portable audio player 21 that detected, the process proceeds to step 508.

  On the other hand, if the connected portable audio player 21 is the same player as the portable audio player 21 that detected the previous connection, it is determined whether the audio file recorded in the portable audio player 21 has been changed since the previous connection. Inspect (step 506), if not changed, the process returns to step 502, and if changed, the process proceeds to step 508.

  Here, whether or not the audio file recorded in the portable audio player 21 has been changed since the last connection is determined by requesting the portable audio player 21 to transfer music management information, so that the music from the portable audio player 21 can be transmitted. The management information may be acquired, and the acquired music management information may be compared with the music DB of the connected device recorded music information. In addition, when information about the date and time when the audio file recorded in the portable audio player 21 was last updated can be acquired from the portable audio player 21, the date and time when the audio file was last updated and the previous portable audio player played the audio. It may be determined whether or not the audio file recorded in the portable audio player 21 has been changed since the previous connection by comparing with the date and time connected to the portable audio player interface 7 of the apparatus.

When the process proceeds from step 504 or step 506 to step 508, the music DB of the connected device recording music information is deleted.
Then, the music management information is acquired from the portable audio player 21, and the new music DB including entries corresponding to the music pieces for which the music management information represents the attribute information and the identifier of the audio file is used as the connection device recording music information. Store (step 512). Here, in each newly created song DB entry, a song ID that is incremented by 1 is registered, and the song name, artist name, album name, genre name, and audio file indicated by the attribute information of the corresponding song are recorded. Register the identifier. That is, the r-th entry newly created in the song DB of the connected device recorded song information is recorded in the portable audio player 21 indicated by the song ID of r and the song management information acquired from the portable audio player 21. The attribute information of the r-th music piece and the audio file identifier are registered.

Then, the voice recognition dictionary editing unit 11 is requested to create a voice recognition dictionary of the connected device recorded music information.
Next, the voice recognition dictionary editing unit 11 requested to create the voice recognition dictionary of the connected device recorded music information starts the connected device recorded music voice recognition dictionary creating process shown in FIG. All the speech recognition dictionaries of information are deleted (step 552).
Then, a dictionary source list is created in which pairs of song IDs and song names extracted from each entry in the song DB of the connected device recorded song information are registered in the order of song IDs (step 554).
Then, a divided dictionary source list is created by dividing the dictionary source list into n pieces of music in the order of the music IDs, and each divided dictionary source list is created in ascending order of the music IDs registered at the beginning. A division number increasing by 1 is given (step 556). That is, t is given as the division number to the divided dictionary source list having the t-th largest music ID registered at the top.

Then, for each of the divided dictionary source lists, a voice recognition dictionary is created and stored in the HDD recorded music information (step 558). Further, t is assigned as a dictionary ID to the speech recognition dictionary created from the divided dictionary original list with the division number t.
Here, the speech recognition dictionary is created from the divided dictionary source list by requesting the speech synthesis engine 8 to create a speech recognition dictionary based on the divided dictionary source list as described above. When the speech synthesis engine 8 is requested to create a speech recognition dictionary based on the divided dictionary source list, the pronunciation of the song name generated in accordance with the yomi conversion rule stored in the HDD 5 is generated for each song name in the divided dictionary source list. A speech recognition dictionary replaced with reading data representing “” is created and stored in the HDD 5.

Here, a processing example of such a connected device recording music tune DB creation process and a connected device recorded music speech recognition dictionary creation process will be shown.
Now, assuming that the number of songs n corresponding to one voice recognition dictionary is 5, as shown in FIG. 6, a portable audio player 21 in which 12 audio files are recorded is connected to an audio playback device, and the portable audio player 21 is connected to the portable audio player 21. It is assumed that the audio player 21 is different from the portable audio player 21 connected to the previous audio playback device or is the portable audio player 21 in which the recorded audio file has been updated since the previous connection.

  In this case, first, after the existing music DB is deleted by the HDD recorded music DB creation processing, a music DB having entries with music IDs 1 to 12 is newly created as indicated by an arrow 601. The attribute information of the 12 audio files and the identifier of the audio file newly recorded in the portable audio player 21 are registered in each entry (step 510).

  Then, after the existing voice recognition dictionary is deleted by the voice recognition dictionary creation process for connected device recorded music, the song ID extracted from each entry in the music DB of the connected device recorded music information as indicated by an arrow 602 A dictionary source list in which pairs of song names and song names are registered in the order of song IDs is created (step 554).

  Then, as indicated by an arrow 603, the dictionary source list is divided into five music pieces in the order of the music IDs, and the divided dictionary source list of division number 1 in which the music titles of music IDs 1 to 5 are registered, A division dictionary source list with division number 2 in which the song names with IDs 6 to 10 are registered and a division dictionary source list with division number 3 in which the song names with song IDs 11 to 12 are registered are created (step 556).

  Then, as shown by the arrow 604, a voice recognition dictionary with a dictionary ID of 1 is created from the divided dictionary source list with a division number of 1, in which Yomi data of the song names of the song IDs 1 to 5 are registered in association with the song ID. A voice recognition dictionary with a dictionary ID of 2 is created from the divided dictionary source list with a division number of 2 and the Yomi data with the song names of 6 to 10 are registered in association with the song ID. From the original list, a voice recognition dictionary with a dictionary ID of 3 is created, in which Yomi data with song names of 11 to 12 are registered in association with the song ID (step 558).

As a result, for every five songs of the audio file recorded in the portable audio player 21, the connected device recording song information in a form in which the voice recognition dictionary of the song name is assigned with the consecutive dictionary ID from 1. It will be created.
The operation for creating HDD recorded music information and connected device recorded music information has been described above.
Next, the reproduction control process performed by the reproduction control unit 14 will be described.
When the user gives an instruction to play back the music recorded on the HDD via the input device 3, the playback control unit 14 sets the HDD playback mode and controls the playback of the audio file recorded on the HDD 5 according to the user's operation. To do. Here, the playback of the audio file is performed by instructing the audio output unit 10 to play the audio file to be played back as an audio file to be played back. The audio output unit 10 reads out the audio file designated as the reproduction target audio file from the HDD 5, decodes it, and outputs it to the speaker 2.

  Further, when the HDD playback mode is set, the playback control unit 14 sets all the voice recognition dictionaries included in the HDD recorded music information to the voice recognition engine 9 as used voice recognition dictionaries. Then, the voice recognition engine 9 performs voice recognition of the voice input from the microphone 1 using each voice recognition dictionary set as the use voice recognition dictionary, and is associated with the reading data that matches the input voice. The reproduction control unit 14 is notified of the music ID registered in the used voice recognition dictionary. Then, when the music ID is notified from the voice recognition engine 9, the playback control unit 14 plays the audio file indicated by the file identifier of the entry in which the notified music ID is registered in the music DB of the HDD recorded music information. By setting the audio output unit 10 as the target audio file, the audio file having the music ID, that is, the music having the song name spoken by the user is reproduced and output to the speaker 2.

  Next, when the portable audio player 21 is connected, the playback control unit 14 sets the connected device playback mode when instructed to play the connected device recorded music by the user via the input device 3. The playback of the audio file recorded in the portable audio player 21 is controlled in accordance with the operation. Here, the reproduction of the portable audio player 21 issues a reproduction request designating the identifier of the audio file to be reproduced to the portable audio player 21, and the portable audio player 21 reproduces the audio file and the reproduced signal / data. This is performed by causing the audio output unit 10 to output the sound represented by the signal / data output from the portable audio player 21 to the speaker 2.

  In addition, when the connected device playback mode is set, the playback control unit 14 sets all the voice recognition dictionaries included in the connected device recorded music information as the use voice recognition dictionary in the voice recognition engine 9. Then, the voice recognition engine 9 performs voice recognition of the voice input from the microphone 1 using each voice recognition dictionary set as the use voice recognition dictionary, and is associated with the reading data that matches the input voice. The reproduction control unit 14 is notified of the music ID registered in the used voice recognition dictionary. Then, when the music ID is notified from the voice recognition engine 9, the reproduction control unit 14 selects the audio file indicated by the file identifier of the entry in which the notified music ID is registered in the music DB of the connected device recording music information. By issuing a playback request designating the identifier of the audio file to be played back to the portable audio player 21 as a playback audio file, playback of the audio file with the song ID, that is, the song with the song name spoken by the user, and the speaker 2 is performed. Is output.

Next, the speech recognition dictionary editing operation performed in the audio playback device will be described.
When the speech recognition dictionary editing unit 11 receives a speech recognition dictionary editing request from the user via the input device 3, the speech recognition dictionary editing unit 11 executes the speech recognition dictionary editing process illustrated in FIG.
As shown in the drawing, in this speech recognition dictionary editing process, first, designation of a song name for correcting the song data and input of new song data for the song name are accepted, and the entered song data is given priority as the reading data for the designated song name. The reading conversion rule of the HDD 5 is corrected so as to be used for the purpose (step 702).

Here, this step 702 is performed as follows, for example.
That is, the speech recognition dictionary editing unit 11 displays a reading change song name selection window as shown in FIG. As shown in the drawing, the Yomi change song name selection window displays a list 801 of song names registered in either the song DB of HDD recorded song information or the song DB of connected device recorded song information, and the song name on the list 801 is displayed. Accept the selection. Then, if the “Change Yomi” button 802 is operated, a Yomi change window as shown in FIG. 8B is displayed, and the song names selected on the Yomi change song name selection window list 801 are displayed 811 and an input box is displayed. The input of reading data for designating a new pronunciation of the song name to 812 is accepted. When the “confirmation button” 813 of the reading change window is operated, the voice synthesis engine 8 is instructed to output voice accompanied with reading data input to the input box 812. When the voice synthesis engine 8 is instructed to output the voice, the voice synthesis engine 8 generates a synthesized voice represented by the reading data according to the instruction and outputs the synthesized voice to the speaker 2 via the audio output unit 10.

  If the “change” button 814 of the reading change window is operated, the speech recognition dictionary editing unit 11 inputs the reading data for the song name selected on the list 701 of the reading change song name selection window 701 into the input box 812. Change to the Yomi data. That is, the reading conversion rule of the HDD 5 is corrected so that the reading data input to the input box 812 is preferentially used as reading data for the song name selected on the list 701 of the reading change song name selection window.

  Now, returning to FIG. 7a, if the reading conversion rule is corrected in this way, the HDD-recorded music information is set as the target music information (step 704), and the speech recognition dictionary correcting process is executed (step 706). Then, the connected device recording music information is set as the target music information (step 708), and the speech recognition dictionary correction process is executed (step 710).

FIG. 7 b shows the procedure of the speech recognition dictionary correction process performed in steps 706 and 710.
As shown in the figure, in this speech recognition dictionary correction process, first, the same song name as the song name whose Yomi data was changed in step 702 of the speech recognition dictionary editing process is registered in the entry of the song DB of the target song information. All music IDs registered in the entry are extracted (step 752).

Then, for each extracted music ID (steps 754, 762, 764), the processing of steps 756-760 is performed.
Here, in steps 756-760, the extracted music ID value is set to q (step 756), and the speech recognition dictionary whose dictionary ID is floor ((q + n-1) / n) is already set as the speech recognition dictionary. If not set (step 758), the re-created speech recognition dictionary is set (step 760).

  Next, when the processing in steps 756-760 is completed for each music ID extracted in step 752 as described above, in step 760, each of the speech recognition dictionaries that are re-created dictionaries (steps 766, 776). 778) and steps 768-774.

  Here, in Steps 768-774, the dictionary ID of the speech recognition dictionary as the re-creation target dictionary is set as g, and the speech recognition dictionary whose dictionary ID of the target music information is g is deleted (Step 770). Then, a dictionary source list in which the song ID and the song name of each entry in which the song IDs from 1-n + (g × n) to g × n are registered in the song DB of the target song information is created (step 772). ), A speech recognition dictionary with a dictionary ID g is created from the created dictionary source list and stored in the target music information (step 774). Here, the speech recognition dictionary is created from the dictionary source list by requesting the speech synthesis engine 8 to create a speech recognition dictionary based on the dictionary source list. When the speech synthesis engine 8 is requested to create a speech recognition dictionary based on the dictionary source list, the speech synthesis engine 8 represents the pronunciation of the song name generated in accordance with the yomi conversion rule stored in the HDD 5 for each song name in the dictionary source list. A voice recognition dictionary replaced with reading data is created and stored in the target music information of the HDD 5.

Here, a processing example of the speech recognition dictionary editing process as described above is shown.
Now, assuming that the number of songs n corresponding to one voice recognition dictionary is 5, as shown by 901 in FIG. 9A, 12 songs with 1 to 12 song IDs assigned to the song DB of the HDD recorded song information. The voice recognition dictionary of the dictionary ID1 in which the Yomi data of the song names of the songs whose song IDs are 1 to 5 are registered in association with the song IDs, and the Yomi data of the song names of the songs whose song IDs are 6 to 10 are the songs. The HDD-recognized music information includes the voice recognition dictionary of the dictionary ID2 registered in association with the ID, and the voice recognition dictionary of the dictionary ID3 in which the Yomi data of the music titles of the music IDs 11 to 12 are registered in association with the music ID. It shall be included in

At this time, it is assumed that the reading of the song name “sade” is changed from “seydo” to “shaaday” by the user.
Then, the reading conversion rule is modified so that “shaaday” is preferentially used as reading data for “sade” (step 702).
Then, 2 and 11 are extracted as the song IDs of the entries in which the song name “sade” in the song DB is registered (step 752). A voice recognition dictionary with a dictionary ID of 1 is set as the re-created voice recognition dictionary using floor ((q + n-1) / n) where 2 is q for the extracted music ID of 2, and the extracted music ID is 11 is set to 11 as q and floor ((q + n-1) / n) sets the speech recognition dictionary with dictionary ID 3 to the recreated speech recognition dictionary (step 760).

Then, the speech recognition dictionaries with dictionary IDs 1 and 3 set in the re-created speech recognition dictionary are deleted (step 770).
Next, the dictionary ID 1 of the re-created speech recognition dictionary is set as g, and a dictionary in which song IDs and song names of entries from 1-n + (g × n) = 1 to g × n = 5 in the song DB are registered. An original list is created as indicated by an arrow 902 (step 772), and a new speech recognition dictionary having a dictionary ID of 1 is created from the dictionary original list as indicated by an arrow 904 (step 774). At this time, the new recognition dictionary is created in accordance with the Yomi conversion rule after being corrected in Step 702, so “shaaday” is registered as the Yomi data for the song ID “2” of the song name “sade”. It will be a thing.

  The same processing is performed for the dictionary ID 3 of another re-created speech recognition dictionary, and the dictionary ID 3 is g, and 1-n + (g × n) = 11 to g × n = of the music DB. A dictionary source list in which song IDs and song names of up to 15 entries are registered is created as indicated by an arrow 903, and a new speech recognition dictionary with a dictionary ID of 3 is created from this dictionary source list as indicated by an arrow 905. Is done. At this time, the new recognition dictionary is created according to the Yomi conversion rule after being corrected in Step 702. Therefore, “shaaday” is registered as the Yomi data for the song ID 11 of the song name “sade”. Will be.

As a result, by updating only the voice recognition dictionary including the song ID of the song having the song name “sade” in the voice recognition dictionary of the HDD recorded song information, all the reading data of “sade” in each voice recognition dictionary is obtained. It has been corrected to “shaaday”.
Although FIG. 9 shows the case of editing the voice recognition dictionary of the HDD recorded music information, the dictionary correction of the voice recognition of the connected device recorded music information is performed in the same manner, and the voice recognition dictionary of the connected device recorded music information is changed. By updating only the voice recognition dictionary including the song ID of the song with the song name “sade”, all the “sade” reading data is corrected to “shaaday”.

The embodiment of the present invention has been described above.
By the way, in the above embodiment, the voice recognition dictionary editing process accepts the change of the reading data of the specific music title at step 702, and the voice recognition dictionary correction process corrects the voice recognition dictionary among the entries in the music DB of each recorded music information. All the song IDs registered in the entry in which the same song name as the song name whose Yomi data was changed in step 702 of the editing process are extracted (step 752), and the voice recognition dictionary in which the extracted song ID is registered is extracted. The process of replacing the newly created voice recognition dictionary (steps 756-760) has been performed, but the following process may be performed instead of such a process.

  That is, in step 702 of the voice recognition process, a change in the reading of the song name of the song with the specific song ID of the specific recorded song information is accepted, and in the voice recognition dictionary correction process, the specific song ID of the particular recorded song information is received. Only the speech recognition dictionary in which is registered may be replaced with a newly created speech recognition dictionary.

  Further, in the above embodiment, deletion of a song having a specific song ID in the HDD-recorded song information is accepted from the user, and the audio file of the song with the song ID that has been accepted to be erased is deleted from the HDD 5. You may make it delete the entry of the said specific music ID from music DB. In this case, the voice recognition dictionary having the entry of the specific music ID is replaced with the voice recognition dictionary created based on the music DB after deleting the entry of the specific music ID. That is, if the music ID of the deleted music is q, the entry with the music ID q is deleted from the music DB, and the voice recognition dictionary of g = floor ((q + n-1) / n) is deleted. Create a dictionary source list in which the song ID and song name of each entry in which the song IDs from 1-n + (g × n) to g × n are registered in the information song DB, and from the created dictionary source list A speech recognition dictionary with dictionary ID g is created and stored in the HDD recorded music information.

  In the above embodiment, the case of recognizing the song name of the song has been described. However, the technique of creating or editing / modifying the voice recognition dictionary in the present embodiment is not limited to the artist name, album name, genre name, or The present invention can be similarly applied to the case of creating or editing / correcting a speech recognition dictionary for other arbitrary texts.

  DESCRIPTION OF SYMBOLS 1 ... Microphone, 2 ... Speaker, 3 ... Input device, 4 ... Display device, 5 ... HDD, 6 ... CD drive, 7 ... Portable audio player interface, 8 ... Speech synthesis engine, 9 ... Speech recognition engine, 10 ... Audio output 11: Voice recognition dictionary editing section, 12 ... Ripping processing section, 13 ... Music management section, 14 ... Playback control section, 20 ... CD-DA disc, 21 ... Portable audio player.

Claims (20)

A speech recognition dictionary used in a speech recognition device for recognizing text represented by speech uttered by humans, storing speech recognition data representing correspondence between text and pronunciation data representing pronunciation of the text for each text to be recognized. A speech recognition dictionary editing device for editing
A voice recognition dictionary storage unit for storing the voice recognition dictionary;
A Yomi conversion rule storage unit that stores Yomi conversion rules that stipulate how to generate pronunciation data from text;
The texts to be recognized are grouped into groups of n (where n is an integer of 1 or more) texts, and each group is generated according to the text in the group and the text conversion rule for the texts. A speech recognition dictionary generating unit that stores the speech recognition dictionary storing the speech recognition data representing the correspondence with the generated pronunciation data as the speech recognition dictionary of the group, and storing the speech recognition dictionary in the speech recognition dictionary storage unit;
A Yomi conversion rule correction unit for correcting the Yomi conversion rule so that the generation method of the pronunciation data for the specified text that is the text specified by the user is the generation method specified by the user;
The group including the designated text is set as a correction target group, the voice recognition dictionary of the correction target group is deleted from the voice recognition dictionary storage unit, and each text in the correction target group and the text conversion rule of the text The speech recognition dictionary storing the speech recognition data representing the correspondence with the pronunciation data generated according to the corrected Yomi conversion rule by the correcting unit is created as a new speech recognition dictionary of the group, and stored in the speech recognition dictionary storage unit A speech recognition dictionary editing apparatus, comprising: a speech recognition dictionary correction unit for storing. The speech recognition dictionary editing device according to claim 1,
The speech recognition dictionary correcting unit sets each group including the same text as the designated text as the correction target group, and for each correction target group, the speech recognition dictionary of the correction target group is stored in the speech recognition dictionary storage unit. The speech recognition dictionary storing the speech recognition data representing the correspondence between each text in the correction target group and the pronunciation data generated according to the correction conversion rule of the text by the correction conversion rule correction unit. A speech recognition dictionary editing device, which is created as a new speech recognition dictionary of the group and stored in the speech recognition dictionary storage unit. The speech recognition dictionary editing device according to claim 1,
When the text to be recognized is added, the added text to be recognized is grouped into groups of n texts, and for each group, the text in the group and the text of the text A voice recognition dictionary adding unit that creates the voice recognition dictionary storing the voice recognition data representing the correspondence with the pronunciation data generated according to the yomi conversion rule as a voice recognition dictionary of the group and additionally storing the voice recognition dictionary in the voice recognition dictionary storage unit A speech recognition dictionary editing apparatus comprising: The speech recognition dictionary editing device according to claim 1,
The speech recognition dictionary correcting unit, when the text is excluded from the text to be recognized, the group including the excluded text as a correction target group, the speech recognition dictionary of the correction target group Correspondence between each text of the correction target group and the pronunciation data generated according to the reading conversion rule of the text after removing the excluded text from the correction target group and deleting from the speech recognition dictionary storage unit The speech recognition dictionary editing apparatus is characterized in that the speech recognition dictionary storing speech recognition data representing is created as a new speech recognition dictionary of the group and stored in the speech recognition dictionary storage unit. A speech recognition dictionary editing device according to claim 1;
A speech recognition apparatus comprising: a speech recognition unit that recognizes text represented by speech uttered by a person using the speech recognition dictionary of each group stored in the speech recognition dictionary storage unit. A speech recognition dictionary editing apparatus according to claim 3;
A speech recognition apparatus comprising: a speech recognition unit that recognizes text represented by speech uttered by a person using the speech recognition dictionary of each group stored in the speech recognition dictionary storage unit. A speech recognition dictionary editing device according to claim 4,
A speech recognition apparatus comprising: a speech recognition unit that recognizes text represented by speech uttered by a person using the speech recognition dictionary of each group stored in the speech recognition dictionary storage unit. The speech recognition dictionary editing device according to claim 1,
The speech recognition dictionary editing apparatus, wherein the text to be recognized is any one of a song title, artist name, album name, and genre name. The speech recognition device according to claim 5, a music data storage unit that stores music data representing music, and a music playback unit that plays back the music data stored in the music data storage unit,
The text to be recognized is the song name of the song represented by the song data stored in the song data storage unit,
The audio reproducing apparatus, wherein the music reproducing unit reproduces music data of a music having a music name recognized by the voice recognition unit. A computer program that is read and executed by a computer,
The computer program stores the computer,
A speech recognition dictionary used for speech recognition for recognizing text represented by speech uttered by humans, storing speech recognition data representing correspondence between text and pronunciation data representing pronunciation of the text for each text to be recognized A voice recognition dictionary storage unit for storing;
A Yomi conversion rule storage unit that stores Yomi conversion rules that stipulate how to generate pronunciation data from text;
The texts to be recognized are grouped into groups of n (where n is an integer of 1 or more) texts, and each group is generated according to the text in the group and the text conversion rule for the texts. A speech recognition dictionary generating unit that stores the speech recognition dictionary storing the speech recognition data representing the correspondence with the generated pronunciation data as the speech recognition dictionary of the group, and storing the speech recognition dictionary in the speech recognition dictionary storage unit;
A Yomi conversion rule correction unit for correcting the Yomi conversion rule so that the generation method of the pronunciation data for the specified text that is the text specified by the user is the generation method specified by the user;
The group including the designated text is set as a correction target group, the voice recognition dictionary of the correction target group is deleted from the voice recognition dictionary storage unit, and each text in the correction target group and the text conversion rule of the text The speech recognition dictionary storing the speech recognition data representing the correspondence with the pronunciation data generated according to the corrected Yomi conversion rule by the correcting unit is created as a new speech recognition dictionary of the group, and stored in the speech recognition dictionary storage unit A computer program that functions as a stored voice recognition dictionary correction unit. A computer program according to claim 10,
The speech recognition dictionary correcting unit sets each group including the same text as the designated text as the correction target group, and for each correction target group, the speech recognition dictionary of the correction target group is stored in the speech recognition dictionary storage unit. The speech recognition dictionary storing the speech recognition data representing the correspondence between each text in the correction target group and the pronunciation data generated according to the correction conversion rule of the text by the correction conversion rule correction unit. A computer program created as a new speech recognition dictionary of the group and stored in the speech recognition dictionary storage unit. A computer program according to claim 10,
The computer further
When the text to be recognized is added, the added text to be recognized is grouped into groups of n texts, and for each group, the text in the group and the text of the text A voice recognition dictionary adding unit that creates the voice recognition dictionary storing the voice recognition data representing the correspondence with the pronunciation data generated according to the yomi conversion rule as a voice recognition dictionary of the group and additionally storing the voice recognition dictionary in the voice recognition dictionary storage unit A computer program that functions as a computer program. A computer program according to claim 10,
The speech recognition dictionary correcting unit, when the text is excluded from the text to be recognized, the group including the excluded text as a correction target group, the speech recognition dictionary of the correction target group Correspondence between each text of the correction target group and the pronunciation data generated according to the reading conversion rule of the text after removing the excluded text from the correction target group and deleting from the speech recognition dictionary storage unit A computer program for creating the speech recognition dictionary storing speech recognition data representing a new speech recognition dictionary of the group and storing it in the speech recognition dictionary storage unit. A computer program according to claim 10,
The computer program further causes the computer to function as a speech recognition unit that recognizes text represented by speech uttered by a person using the speech recognition dictionary of each group stored in the speech recognition dictionary storage unit. A computer program characterized by the above. A computer program according to claim 14,
The speech recognition apparatus according to claim 1, wherein the text to be recognized is any one of a song title, artist name, album name, and genre name. A computer program according to claim 14,
Further causing the computer to function as a music playback unit that plays back the music data stored in a music data storage unit that stores music data representing music;
The text to be recognized is the song name of the song represented by the song data stored in the song data storage unit,
The computer program according to claim 1, wherein the music playback unit plays back music data of a song having a song name recognized by the voice recognition unit. A speech recognition dictionary used in a speech recognition device for recognizing text represented by speech uttered by humans, storing speech recognition data representing correspondence between text and pronunciation data representing pronunciation of the text for each text to be recognized. A speech recognition dictionary editing method for editing
Each of the texts to be recognized is grouped into groups of n (where n is an integer of 1 or more) texts, and for each group, each text in the group and the pronunciation data from the text of the text. A voice recognition dictionary generating step of creating and storing the voice recognition dictionary storing the voice recognition data representing the correspondence with the pronunciation data generated in accordance with the Yomi conversion rule that defines the generation method of
A Yomi conversion rule correction step for correcting the Yomi conversion rule so that the generation method of the pronunciation data for the specified text which is the text specified by the user is the generation method specified by the user;
The group including the specified text as a correction target group, the speech recognition dictionary of the correction target group is deleted, and each text in the correction target group and the pronunciation generated according to the corrected Yomi conversion rule of the text A speech recognition dictionary editing method, comprising: a speech recognition dictionary correcting step of creating and storing the speech recognition dictionary storing speech recognition data representing correspondence with data as a new speech recognition dictionary of the group. The speech recognition dictionary editing method according to claim 17,
In the speech recognition dictionary correction step, each group including the same text as the designated text is set as the correction target group, and for each correction target group, the voice recognition dictionary of the correction target group is deleted, and the correction target The speech recognition dictionary storing speech recognition data representing correspondence between each text in the group and pronunciation data generated according to the corrected reading conversion rule of the text is created and stored as a new speech recognition dictionary of the group A speech recognition dictionary editing method characterized by: The speech recognition dictionary editing method according to claim 17,
When the text to be recognized is added, the added text to be recognized is grouped into groups of n texts, and for each group, the text in the group and the text of the text A voice recognition dictionary having a voice recognition dictionary adding step of creating and storing the voice recognition dictionary storing the voice recognition data representing the correspondence with the pronunciation data generated according to the reading conversion rule as the voice recognition dictionary of the group Dictionary editing method. The speech recognition dictionary editing method according to claim 17,
In the speech recognition dictionary correction step, when the text is excluded from the text to be recognized, the speech recognition dictionary of the correction target group is set as the correction target group including the group including the excluded text. In addition to deleting the excluded text from the correction target group, voice recognition data representing correspondence between each text of the correction target group and the pronunciation data generated according to the reading conversion rule of the text is stored. The speech recognition dictionary editing method, wherein the speech recognition dictionary is created and stored as a new speech recognition dictionary of the group.