JP2009204872A - Creation system of dictionary for speech recognition - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a creation system of a dictionary for speech recognition, capable of creating a correct dictionary for speech recognition in a short period of time, and accurately and speedily performing speech recognition by using this. <P>SOLUTION: The creation system of the dictionary for speech recognition, which is used for performing device operation by recognizing user's utterance speech is created by attaching reading to an original data for creating the dictionary for speech recognition, which is collected by a speech recognition object data collection section. In the speech recognition object data collection section, the original data for creating the dictionary for speech recognition is separately collected for each data classification, and a dictionary for speech recognition device operation is created by attaching reading to the original data for creating the dictionary for speech recognition, by using a basic dictionary for creating the dictionary for speech recognition, which is separately created. The basic dictionary for creating the dictionary for speech recognition is separately created in the same classification as that in which the original data for creating the dictionary for speech recognition is separated, by collecting the original data for reading conversion which is presumed to be used in the speech recognition device, and by attaching reading beforehand. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a speech recognition dictionary generation system capable of recognizing speech in a language that does not have reading kana information such as English as accurately as possible when creating a dictionary used for speech recognition.

  Conventionally, for example, in a vehicle navigation device, when a driver searches for a destination, etc., it is possible to input even during driving of the vehicle, or without performing manual work when setting a normal destination. For example, a voice recognition technique for recognizing a voice uttered by a driver or the like and obtaining input data has been developed and widely used. In addition to such navigation devices, audio devices, air conditioners, etc. recognize voices uttered by users and perform various operations so that the driver can safely operate various devices mounted on the vehicle. Some have been proposed that can be done.

  In the above-mentioned audio apparatus for vehicles, an audio device storing a large amount of audio data on a hard disk or the like has recently been used, and a desired artist, album, or song is selected and listened to from the audio data. Therefore, a search is performed, and reproduction is performed according to a playlist obtained as a result of the search. At that time, it is also proposed to use a voice recognition device for operating by voice so that the driver can easily operate.

  Particularly in recent years, portable audio players have built-in large-capacity memory chips, small hard disks, etc. as data recording media, and a large amount of audio data compressed with MP3 etc. can be recorded on this data recording medium and carried freely. You can listen to your favorite songs. Portable audio players that incorporate a data recording medium that records such a large amount of audio data are often brought into vehicles, in which case they are reproduced and output by a high-performance audio device installed in the vehicle. It is desirable to listen. Therefore, the vehicular audio apparatus is provided with means for connecting these portable audio apparatuses, reading out the built-in audio data, reproducing them, and outputting them from the vehicular audio apparatus.

  When the portable audio player is connected to the vehicle audio device as described above and the audio data of the data recording medium built in the portable audio device is input, reproduced, and output, the portable audio player carries the portable audio player. The audio data recorded on the data recording medium of the type audio apparatus is searched, and an arbitrary artist, album, or song is selected to create and play a playlist. Also at this time, it is desired that the selection operation is performed by voice so that the driver can easily select a song.

  As described above, various devices used in a vehicle are often operated by a driver, and therefore, it is required that the driver can operate as much as possible so as not to prevent confirmation of safety ahead. It is desirable to operate by. In such a voice operation, the voice data uttered by the user is recognized. Therefore, the voice data uttered by the user is compared with the voice data registered in advance as a recognition dictionary, and the voice data most suitable is compared. Is output as an operation signal instructed by the user by speaking. The recognition dictionary used at that time can also be created by causing a user to utter words used in speech recognition in advance and storing the speech data as a dictionary.

  However, this method can be used when the number of records is relatively small, such as a personal address book or a telephone directory. For example, a song recorded on the hard disk is recorded on the hard disk in order to search and play it back by voice recognition. For example, registering an artist name, an album name, or a song name of 10,000 songs in advance takes too much time for registration and is almost impossible. Especially when audio data from different audio recording media is used, such as when a portable audio player is connected to a vehicle audio device, the recorded songs are different, and those songs are searched by voice recognition. Sometimes it is virtually impossible to use.

  In addition to the creation of a dictionary for speech recognition, recognition is performed by previously creating reading data from a character string using a speech synthesis (TTS: text-to-speech) technique and comparing it with speech uttered by the user. Is also possible. That is, for example, in a navigation apparatus, guidance to a left or right turn is given by voice when guidance is given to a user, and at that time, a voice synthesis technique for converting text data into voice is used. Using this technology, text data consisting of character strings such as artist names, album names, or song names recorded in each song can be used as read data, which can be registered as voice data to create a voice recognition dictionary. Conceivable.

  This method has the advantage that the user can save time and effort compared to the previous method. However, when these data are not marked with a reading pseudonym, it is up to the TTS to assign the reading character, and the user intends. Reading data may not be added. That is, for example, when “110” is originally read as “one ten” and no reading pseudonym is given, it is given as “one hundred ten”. May not be granted. This is especially true for names related to songs, which are often used in a unique way of reading to increase the appeal of sales, whereas users often remember only by pronunciation, so In many cases, an appropriate song corresponding to is not searched.

In addition, in order to create an appropriate speech recognition dictionary from data including unreadable symbols such as “−”, “?”, “+”, “/”, Etc., speech recognition in which pronunciation data is accumulated so as to match the user's utterance A technique for generating a dictionary is disclosed in Japanese Patent Application Laid-Open No. 2004-53978, and a technique for detecting the occurrence status of a paraphrase vocabulary and registering and using the spoken paraphrase vocabulary is disclosed in Japanese Patent Application Laid-Open No. 2004-53978. This is disclosed in Japanese Patent Publication No. 2007-213005.
JP 2004-53978 A JP 2007-213055 A

  As described above, in particular, when operating a vehicle device, it is desired to perform various operations by recognizing a voice uttered by a user, and it is also desirable to perform various operations by voice recognition in an audio device. On the other hand, if a voice recognition dictionary is created by TTS based on artist information, album name, or song information recorded along with audio data on an audio recording medium such as a hard disk, it cannot be read as originally read, Since different speech recognition dictionaries are created, there are many cases where the speech related to the song uttered by the user cannot be correctly recognized.

  In some cases, audio device users may input “Kana” in advance for these song data. In this case, the data of the Kana can be used. In this case, the above-described problem occurs.

  Furthermore, when handling a huge amount of data, such as when recording audio data on a large-capacity data storage medium such as the hard disk, it takes a long time to create a speech recognition dictionary using TTS. Therefore, for example, when a portable audio player that records a large amount of audio data is connected to a vehicle audio device and an arbitrary song is to be selected and listened to by voice recognition, the portable audio player is connected to the vehicle audio device. When a process for automatically creating a dictionary for speech recognition is performed immediately, a long time is required for the process, and device operation with voice is performed until the speech recognition dictionary is created. Therefore, it becomes a device with poor usability, which may give the user discomfort and distrust.

  This is not limited to the audio device mounted on the vehicle as described above. For example, a new place name provided when the map device is updated by downloading new map data or new map difference data in the navigation device. This is the same when searching by voice recognition, and when creating a dictionary for voice recognition from text data of place names, the original reading cannot be given, There is a problem that it is not possible to cope with a specific original reading uttered by the user, and appropriate recognition cannot be performed.

  Furthermore, in recent years, when a mobile phone is brought into a vehicle, it can be connected to a navigation device, connected to the Internet network using the mobile phone, and various information can be captured and displayed and used. In addition, it has also been proposed to operate a mobile phone using a voice recognition function. When making a call using the phone book of the mobile phone at that time, it is also possible to search by name for the name, company name, etc. registered in the phone book, output the phone number, and have the function to make a call Proposed. Even in such a case, when creating a speech recognition dictionary considering that the phone book is searched by voice recognition immediately when a mobile phone is connected, text data existing in the phone book is unique. Therefore, the created speech recognition dictionary is not necessarily an appropriate dictionary, and therefore, a search by appropriate speech recognition cannot be performed and a call cannot be made.

  These problems are not necessarily limited to devices for vehicles, but the same applies when a speech recognition dictionary is created using data that is stored in a device to which a reading device is not attached and speech recognition is performed using the dictionary. Cause problems.

  Accordingly, the present invention creates a speech recognition dictionary using data that is stored in a device to which reading is not assigned, and performs speech recognition using the dictionary to perform various device operations. A main object is to provide a speech recognition dictionary generation system that creates a correct speech recognition dictionary in time and can perform speech recognition accurately and at high speed using the dictionary.

  In order to solve the above-described problem, the speech recognition dictionary generation system according to the present invention uses a speech recognition target data collection unit to create a speech recognition device operation dictionary used for device operation by recognizing a user's speech. In the speech recognition dictionary generation system that generates by adding reading kana to the collected speech recognition dictionary generation original data, the speech recognition target data collection unit divides the speech recognition dictionary generation source data for each type of data. Using the basic dictionary for speech recognition dictionary generation that is collected separately, and generating the speech recognition device operation dictionary by assigning a reading pseudonym to the original data for speech recognition dictionary generation, and generating the speech recognition dictionary The basic dictionary is used to collect the reading conversion original data predicted to be used in the voice recognition device in advance and give a reading pseudonym. In the generation of the voice recognition device operation dictionary, the type of data in the voice recognition dictionary generation basic dictionary corresponding to the type of the voice recognition target data is generated. It is characterized in that it is generated by assigning a reading pseudonym using.

  Further, in another speech recognition dictionary generation system according to the present invention, in the speech recognition dictionary generation system, the device operated by the speech recognition is an audio device, and the speech recognition target data collection unit includes: Data necessary for playback operation is collected for each type of data, and the basic dictionary for speech recognition dictionary generation predicts data collected by the speech recognition target data collection unit and collects basic reading data for each type of data However, it is characterized by being given a reading.

  In another speech recognition dictionary generation system according to the present invention, when the speech recognition target data collection unit of the speech recognition dictionary generation system is connected to another audio player in the audio device, the audio player It collects by taking in the music information of the music currently recorded on this data recording medium.

  Further, in another dictionary generation system for speech recognition according to the present invention, in the dictionary generation system for speech recognition, a word that cannot be given a reading pseudonym to the original data for speech recognition dictionary generation by the basic dictionary for speech recognition dictionary generation, A voice recognition device operation dictionary is generated by adding a reading pseudonym by voice synthesis means.

  In another speech recognition dictionary generation system according to the present invention, the separately created basic dictionary for speech recognition dictionary generation is binary processed in the dictionary generation system for speech recognition.

  In another speech recognition dictionary generation system according to the present invention, in the speech recognition dictionary generation system, the basic dictionary for speech recognition dictionary generation is not read by data collected by the speech recognition target data collection unit. When the processing excluding the symbols is performed, the same processing is performed when the reading is given to the reading conversion original data.

  Since the present invention is configured as described above, a dictionary for operating a voice recognition device is created using data that is stored in a device to which a reading device is not attached and voice recognition is performed using the dictionary. When a device is operated, a correct speech recognition dictionary can be easily created in a short time, and a speech recognition dictionary generation system can be used to accurately and quickly perform speech recognition using the dictionary. Can do.

  The present invention creates a speech recognition dictionary using data that has not been given a reading pseudonym stored in a device to be used, and performs voice recognition using the dictionary for correct operation in a short time. Voice recognition used to create a voice recognition dictionary and use it to recognize voices of users and perform device operations for accurate and high-speed voice recognition. In the speech recognition dictionary generation system that generates the device operation dictionary by reading and adding kana to the voice recognition dictionary generation original data collected by the voice recognition target data collection unit, the voice recognition target data collection unit The recognition dictionary generation original data is collected separately for each type of data, and a reading pseudonym is given to the voice recognition dictionary generation original data using a separately created basic dictionary for speech recognition dictionary generation. The voice recognition device operation dictionary is generated by the above, the basic dictionary for voice recognition dictionary generation collects the reading conversion original data that is predicted to be used in the voice recognition device in advance and gives a reading pseudonym, The voice recognition dictionary is created by dividing the original data for generating the voice recognition dictionary into the same type, and when generating the voice recognition device operation dictionary, the voice recognition dictionary corresponding to the type of the voice recognition target data This was realized by using a type of data in the basic dictionary for generation and assigning it with a reading pseudonym.

  Embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a functional block diagram of an embodiment in which the present invention is applied to an audio apparatus. As shown in the figure, the present invention first includes a basic dictionary 6 for generating a speech recognition dictionary on a PC, which is first performed by the manufacturer of the audio apparatus. And secondly, a process for generating a dictionary for operating a portable audio player voice recognition device using the basic dictionary 6 for voice recognition dictionary generation in the audio device.

  In the process of creating a basic dictionary for generating a speech recognition dictionary on the first PC, in the illustrated example, a device operation database (DB) 1 and a basic song information database (DB) 2 for operating devices are used. The basic reading data collecting unit 3 collects original data for reading conversion. In principle, this work is performed by a manufacturer of an audio device as a service that increases the added value of the audio device. When collecting the data, it is collected for each type of data, and in order to select and play a song with an audio device, as shown in the figure, basic device operation data for operating the device, song name data, artist name data, album name Data, for example, genre name data added as necessary, is collected for each type.

  In collecting the device operation basic data in the basic reading data collecting unit 3 as described above, it is possible to use this when the words necessary for operating the device are stored in advance as the device operation database 1. When such data does not exist, a word estimated to be necessary for device operation is input by using a personal computer.

  For the other song name data, artist name data, and album name data, data stored in advance in the basic song information database 2 is used. Various kinds of basic music information database 2 can be used. For example, CDDB (CD database) which collects and discloses CD music information can be used. Since this data includes the TOC data of the CD, it is possible to easily collect data such as the song name, artist name, album name, genre name, and release date recorded on the CD.

  The original data for reading conversion in the basic reading data collecting unit 3 in FIG. 1 is, for example, data as shown in FIG. That is, in the reading conversion original data collection example shown in FIG. 2, “Play”, “Vol”, “Vol.”, “Artist”, “Song”, “By”, etc. are extracted as device operation basic data from the device operation database. Is shown. Here, a user who performs an operation based on speech recognition can perform processing for creating a speech recognition dictionary by a program as described later so that the speech uttered to the audio device can be recognized. In order to be able to do so, the words used are pre-examined and further estimated and collected.

  For example, FIG. 2 shows an example in which “Anything for you”, “Black & Blue”, “Crazy 4 U”, “DANCE 2”, and the like are extracted as the reading data of the song titles collected from the basic song information database. Also, “Bonnie Pink”, “Cocco”, “hide”, “Mr. Children”, etc. are extracted as the reading data of the artist name, and “Best-first-things [disc1]”, “Best-first-things” are extracted as the album name reading data. In this example, [disc2], “Crispy!”, “ULTRA BURE” and the like are extracted, and “AC / DC” and the like are extracted as other data.

  In the reading conversion processing unit 4 in FIG. 1, the reading conversion original data collected by the basic reading data collecting unit 3 is subjected to a reading conversion process by assigning a reading kana to the reading converted original data. Get the original data. Here, it is also created for each data type. When the kana is given, it can be used if the kana information is given by the CDDB, and when it is not given, this voice recognition dictionary is generated. A person in the department that creates the basic dictionary inputs appropriate data. The data to be entered at this time does not necessarily need to be assigned to all currently provided songs. Even if it is given for a song name, artist name, album name, etc. that are specially read by famous artist songs, this book The invention can be implemented.

  As a result of this processing, for example, read-converted original data as shown in FIG. 3 is created. That is, in the example shown in FIG. 3, the character whose character string is “Play” in the device operation basic data is read as “play”, and hereinafter “Vol” and “Vol.” Are similarly referred to as “volume”, In this example, “Artist” is read as “Artist”, “Song” is read as “Song”, and “By” is read as “Bi”. For song title reading data, “Anything for you” is “Anything For You”, “Black & Blue” is “Black and Blue”, “Crazy 4 U” is “Crazy For You”, “DANCE2” is “Dance Dance” Shows an example to read.

  In this way, “DANCE 2” is usually read as “Dance Two”, but by inputting “Dance Dance” formally, it is difficult with conventional audio devices. Even for English characters that are rarely used for reading kana, correct kana can be given in advance, and when the user generates “dance dance” in the subsequent speech recognition process, the correct “DANCE 2” It is possible to immediately reproduce the song.

  As the reading data of the artist name, “Bonnie Pink” should be read as “Bonnie Pink”, “Cocco” as “Koko”, “hide” as “Hide”, and “Mr. Children” as “Mr. Children”. “Mr. Children” is also input separately in consideration of the fact that it is often referred to as “mystil” by an abbreviation or nickname. By inputting such abbreviations and nicknames, it becomes possible to flexibly cope with the user uttering the same artist in various modes during voice recognition.

  As the album name reading data, “Best-first-things [disc1]” is “Best First Things Disc One”, “Best-first-things [disc2]” is “Best First Things Disc Two”, “Crispy!” In this example, “crispy” and “ULTRA BURE” are read as “ultra blue”. In addition, as other data, “AC / DC” indicates an example in which “AC” is read as “AC”.

  In the binarization processing unit 5, the read-converted original data created by the read conversion processing unit 4 is processed into binary data for each type. Here, the binarization is performed so that this data can be used directly, whereby the basic dictionary for speech recognition dictionary generation created here is an audio device or a navigation device connected to this audio device. When performing various other processes, it is possible to prevent a general compilation process from being performed, and to prevent any process other than reading from occurring in the audio apparatus or the navigation apparatus connected thereto. It should be noted that, when performing binary processing, more accurately, as is well known, compilation is performed and binarization is performed.

  The basic dictionary 6 for generating a speech recognition dictionary for each type thus binarized is the vehicle audio device shown as the audio device 11 or a navigation device connected thereto (hereinafter abbreviated as an audio device). The voice recognition processing unit 21 in FIG. In that case, the data can be input to a memory provided in an audio device or the like, downloaded to a data recording medium such as an HDD, or moved to a memory in which data is recorded, and further supplied as a circuit chip.

  In the example of the creation process of the portable audio player voice recognition device operation dictionary 29 in the audio device shown in FIG. 1, the portable audio player 12 is connected to the audio device 11 for vehicles and the like, and the external device of the portable audio player is connected. Unit 13 and the external device connection unit 14 of the audio device 11 are connected by wire or wirelessly, so that the portable audio player 12 can be played back by a signal from the external device operation signal output unit 16 in the audio device 11. An operation is performed, and arbitrary data of the portable audio player can be captured from the data capturing unit 14 of the audio device 11 in accordance with the operation instruction.

  The speech recognition processing unit 21 of the audio device 11 shown in FIG. 1 includes a speech recognition target data collection unit 23. Here, the portable audio player is connected to the audio device 11 as described above, and the audio device 11 and the audio device 11 are portable. When the above-described communication is possible between the type audio players, the voice recognition target data collection unit 23 automatically moves from a data recording medium such as a memory chip or a hard disk incorporated in the portable audio player 12 Import song information for audio data recorded in The music information does not need to include audio data recording the music itself. These pieces of music information can be extracted and collected from the tag portion in which the music information in the data is recorded when the audio data is recorded in the MP3 format on the data recording medium of the portable audio player.

  The song information collected here is fetched by dividing it into types such as a song name, artist name, album name, and genre name, which are instructed when the user selects a song. All of these types can be taken in from the music information data recorded on the data recording medium by MP3 or the like. However, when the data does not exist, the audio device 11 has CDDB data in advance. The data can be retrieved and imported. Further, when the audio device 11 has a communication function such as the Internet, the data can be directly connected to the CDDB data providing site. Note that the type of song information collected when creating the basic dictionary 6 for generating a speech recognition dictionary that works on the PC as described above is the speech recognition dictionary that is collected for each type of the data collection unit 23 for the speech recognition dictionary. The same classification as the original data for generation is used to prevent mismatch between the two.

  The voice recognition processing unit 21 downloads the data of the voice recognition dictionary generation basic dictionary 6 created on the PC to a data recording medium accessed by the voice recognition processing unit of the audio device 11 or in advance in the form of a chip or the like. The data of the basic dictionary 6 for voice recognition dictionary generation can be used by installing or inserting it as a memory chip. When the voice recognition dictionary generating basic dictionary 6 exists in the voice recognition processing unit 21 in this way, the data can be updated thereafter, and the manufacturer of the audio device 11 provides update data. Therefore, it is more preferable to cope with the production of a large number of songs year by year.

  The reading data generation post-processing unit 24 of the voice recognition processing unit 21 creates reading data for the music information collected by the voice recognition target data collection unit 23 using the basic dictionary 6 for voice recognition dictionary generation. In this case, when the type of song information collected by the voice recognition target data collection unit 23 is, for example, artist name data, the voice recognition dictionary generation basic dictionary 6 used for generating the reading data is also the same type of artist. Search the name data part. By performing such processing, accurate reading data can be given from a small amount of data at high speed.

  When the reading data is generated in the voice recognition dictionary generating basic dictionary 6 in the reading data generation post-processing unit 24, the reading data is added to the portable audio player voice recognition device operation dictionary 29 as music information data with reading data 25. Record. When the reading data generation post-processing unit 24 does not have the reading data in the speech recognition dictionary generation basic dictionary 6, the reading data is output to the reading data generation post-processing unit 27 as a post-processing unit to be performed next as no reading data 26. To do. In the post-data generation post-processing unit 27, for example, a speech synthesis dictionary 28 in speech synthesis (TTS: text-to-speech) technology for speech guidance widely used in navigation devices and the like, and speech synthesis processing technology are used. For example, when there is speech synthesis data corresponding to the artist name, the data is directly used, and when there is no speech synthesis data, reading data for reading that is assumed to be normal reading is generated. In this way, it is recorded in the portable audio player speech recognition device operation dictionary 29 as music information information with read data generated by the read data generation post-processing unit 27. In the example of FIG. 1, in order to show an example of operating the portable audio player 12 by voice recognition in the voice recognition dictionary unit 21 of the audio device 11, it is shown as “a portable audio player voice recognition device operation dictionary 29. However, when operating various devices, it can be simply referred to as “speech recognition device operation dictionary 29”.

  In the portable audio player speech recognition device operation dictionary 29, as described above, the reading data generation processing unit 24 uses the speech recognition dictionary generation basic dictionary 6 to generate song information data with reading data, and after generation of the reading data. For the music stored in the portable audio player 12 connected to the audio device 11 based on the song information data with read data generated by the processing unit 27, the voice recognition is immediately performed when the portable audio player is connected to the audio device. A dictionary can be generated.

  Therefore, after that, the user recognizes “voice”, “artist”, and “bonnie pink” in a predetermined order in order to reproduce the music, and the voice recognition processing unit 30 carries these voices with the microphone 17. Type audio player voice recognition device operation dictionary 29 can be searched and recognized in order, and a voice recognition result 31 that “bonnie pink” music is played can be obtained. Based on the voice recognition result 31, the external device operation signal output unit 16 outputs an instruction to select and output a “bonny pink” song to the portable audio player 12 via the external device connection unit of both devices. The audio device 11 takes in the output audio data and performs reproduction processing.

  In the speech recognition dictionary generation system of the present invention configured by the functional blocks as described above, for example, it can be implemented by sequentially operating according to the operation flow shown in FIGS. FIG. 4 shows an operation flow of the creation process of the voice recognition dictionary generation basic dictionary. This operation is performed in the process of creating the voice recognition dictionary generation basic dictionary 6 on the PC of FIG. A word for generating a basic dictionary for generating a recognition dictionary is collected (step S1). The collected words are then A. Device operation basic data; Song name data, C.I. Artist name data, D.C. Album name data, E.I. The original data for reading conversion classified for each type such as other is created (step S2). These operations are performed by collecting words for device operation from the device operation database 1 and various pieces of song information from the basic song information database in the basic reading data collection unit 3 in FIG.

  Thereafter, the word reading of the original data is input (step S3). This processing can be performed by the reading conversion processing unit 4 in FIG. 1 by the method described above. Next, the original data for reading conversion obtained in this way and the reading data given thereto are binarized (step S4), and a basic dictionary for generating a speech recognition dictionary divided for each type of word is created. (Step S5).

  Generation of the voice recognition device operation dictionary using the basic dictionary for voice recognition dictionary generation thus obtained can be performed by the operation flow shown in FIG. In the voice recognition device operation dictionary generation process shown in FIG. 5, the portable audio player is first connected to the audio device (step S11). Next, music information of the portable audio player is acquired for each data type by the audio device (step S12). This processing is performed by the voice recognition target data collection unit 23 of the voice recognition processing unit 21 in the audio apparatus 11 of FIG.

  By this data collection, original data for generating a speech recognition dictionary is created (step S13), and then words for recognition of the original data are sequentially selected and output (step S14), and at that time, the type of the original data is determined (step S15). ). Thereafter, the type of the basic dictionary for speech recognition dictionary generation corresponding to the type of original data is selected (step S16), and it is determined whether or not there is reading data in the selected type of basic dictionary for speech recognition dictionary generation (step S17). ). As a result, when it is determined that there is no reading data in the voice recognition dictionary generation basic dictionary, reading data is generated by the voice synthesis (TTS) dictionary and its processing technology. In these processes, the reading data generation processing unit 24 in FIG. 1 uses the speech recognition dictionary generation basic dictionary 6 to search for reading data of the same data type, and if there is no reading data, the reading data generation post-processing is performed. In the unit 27, the speech synthesis (TTS) dictionary 28 and its processing technique are used to obtain reading data.

  If it is determined in step S17 that the selected type of speech recognition dictionary generation basic dictionary contains reading data, the reading data is generated by the speech recognition dictionary generation basic dictionary (step S19), and then in step S18, the reading data Along with the generation, it is determined whether or not the reading data of all the voice recognition original data has been generated (step S20), and there is still the voice recognition original data for which no reading data has been generated. When the determination is made, the process returns to step S14, and the same operation is repeated from the operation of selecting and outputting the recognition words of the original data in order. When it is finally determined in step S20 that the reading data of all the voice recognition original data has been generated, the voice recognition device operation dictionary is completed with this data, and the portable audio player voice recognition device operation dictionary of FIG. 29 is completed (step S21).

  The reproduction operation of the audio device by voice recognition performed using the portable audio player voice recognition device operation dictionary obtained as shown in FIG. 5 can be performed by the operation flow shown in FIG. That is, in the audio device playback operation processing by voice recognition shown in FIG. 6, the music playback operation voice is first uttered (step S31), and then the voice recognition device operation dictionary data generated earlier is generated. A search is performed (step S32), and a speech recognition process is performed by this search (step S33).

  In this search and voice recognition processing, as described above, the device operation of the portable audio player is performed by voice recognition in advance, and as a result of generating the voice recognition device operation dictionary, the portable type connected to the audio device is obtained. All the music information stored in the audio player is present in the voice recognition device operation dictionary, and the music existing in the portable audio player is almost certainly recognized in the voice related to the playback of the music uttered by the user. Can do.

  Thereafter, it is determined whether or not the utterance of the music playback operation has ended (step S34). For example, by determining whether or not the utterance has been interrupted for 3 seconds, it is determined that the utterance has not yet ended. That is, when the voice is subsequently uttered, the process returns to step S32 again to search the voice recognition device operation dictionary that previously generated the uttered voice, and the same operation is repeated thereafter. When the next utterance is not made for a predetermined time in step S34 as described above, or when it is determined that the sound is not the playback operation of the portable audio player, it is determined that the utterance of the music playback operation has ended. Then, the device is operated with the recognized words, and the designated music is reproduced (step S35).

  In these processes, the speech recognition processing unit 30 in FIG. 1 inputs the user's utterance voice input from the microphone 17 and searches the portable audio player voice recognition device operation dictionary 29 to perform voice recognition. Is output from the external device operation signal output unit 16 to the portable audio player 12 to search for a predetermined song and output data. In the audio device 11, the data acquisition unit 15 captures the data. It is executed by performing a reproduction process.

  As a result of using the basic dictionary for speech recognition dictionary generation as described above, when the original data of the recognition dictionary is data of a program as shown in FIG. 7A, read-converted data as shown in FIG. 3 is obtained. In this case, this is binary processed to create a basic dictionary for speech recognition dictionary generation, and when used in the speech recognition device 21 in the audio device 11 in FIG. 1, the data in the portion shown in bold in FIG. Reading can be given by this dictionary.

  In the example shown in FIG. 7A, when the song names are sequentially fetched under the device operation instruction of “Play by Song”, that is, the device is operated by inputting the song name, “Play”, “by” "Song" device operation basic data can be given a reading pseudonym by the reading data of FIG. 3A, and "Anything for you", "Black & Blue", "Crazy 4U", " As for “DANCE 2”, a reading pseudonym is given by the reading data of FIG. 3B, and then it can be used as a speech recognition device operation dictionary.

  As a result of such processing, as shown in FIG. 7 (b), all of the device operation basic data parts shown in bold in the figure, and four song names shown in bold in the eight song titles in the song name data part. Reading data can be acquired. As a result, only the remaining four music titles for which reading data could not be acquired need to be processed by the speech synthesis (TTS) dictionary and the TTS processing method to obtain reading data. As a result, for example, when reading data is added to 10,000 pieces of music data, and it takes 5 minutes to create a dictionary by performing speech synthesis (TTS), the method of the present invention is used. When reading data is obtained from the basic dictionary for 100% speech recognition dictionary generation (hit rate 100%), the reading process can be completed in a few dozen seconds, and even when the hit rate is 50%, it is shortened to several minutes. It was confirmed that the speech recognition dictionary generation system according to the present invention is extremely effective.

It is a functional block diagram of the Example of this invention. It is a figure which shows the collection example of the original data for reading conversion in the Example. It is a figure which shows the example of the read converted original data which performed the read conversion about the collected original data for the read conversion. It is an operation | movement flowchart which performs the process which produces the basic dictionary for speech recognition dictionary production | generation in the Example. It is an operation | movement flowchart which performs the dictionary production | generation process for voice recognition apparatus operation in the Example. It is an operation | movement flowchart which performs audio equipment reproduction operation by voice recognition in the Example. It is a figure which shows the example of the original data read with the basic dictionary for speech recognition dictionary production | generation in the Example.

Explanation of symbols

1 Equipment operation database
2 basic song information database 3 basic reading data collection unit 4 reading conversion processing unit 5 binarization processing unit 6 basic dictionary for speech recognition dictionary generation 11 audio device 12 portable audio player 13 external device connection unit 14 external device connection unit 15 data collection Insertion unit 16 External device operation signal output unit 17 Microphone 21 Speech recognition processing unit 23 Speech recognition target data collection unit 24 Reading data generation processing unit 25 Reading data present 26 Reading data not present 27 Reading data generation post-processing unit 28 Speech synthesis (TTS) Dictionary 29 Band type audio player Voice recognition device operation dictionary 30 Voice recognition processing unit 31 Voice recognition result

Claims (6)

A speech recognition device operation dictionary used for device operation by recognizing a user's uttered speech is generated by adding a pseudonym to the original data for speech recognition dictionary generation collected by the speech recognition target data collection unit. In the dictionary generation system for speech recognition,
The voice recognition target data collection unit collects voice recognition dictionary generation original data separately for each type of data,
Using the basic dictionary for voice recognition dictionary generation created separately, generating the voice recognition device operation dictionary by giving a reading pseudonym to the voice recognition dictionary generation original data,
The basic dictionary for speech recognition dictionary generation is the same type as the type that collects the reading conversion original data predicted to be used in the speech recognition apparatus in advance and assigns the reading pseudonym, and separates the speech recognition dictionary generation original data It was created separately,
When generating the dictionary for operating the voice recognition device, it is generated by adding a reading pseudonym using the type of data in the basic dictionary for voice recognition dictionary generation corresponding to the type of the voice recognition target data. Dictionary generation system for voice recognition. The device operated by the voice recognition is an audio device,
The voice recognition target data collection unit collects data necessary for the playback operation of the audio device for each type of data,
The basic dictionary for voice recognition dictionary generation is created by predicting data collected by the voice recognition target data collection unit, collecting basic reading data for each type of data, and adding readings. Item 4. The dictionary generation system for speech recognition according to Item 1.   The voice recognition target data collecting unit collects music information of songs recorded on a data recording medium of the audio player when another audio player is connected to the audio device. Item 3. The dictionary generation system for speech recognition according to Item 2.   A word for which a kana cannot be assigned to the original data for generating a speech recognition dictionary by the basic dictionary for generating a speech recognition dictionary is provided with a reading pseudonym by a speech synthesizer to generate a dictionary for operating a speech recognition device. Item 4. The dictionary generation system for speech recognition according to Item 1.   The speech recognition dictionary generation system according to claim 1, wherein the separately created basic dictionary for speech recognition dictionary generation is subjected to binary processing.   The basic dictionary for speech recognition dictionary generation performs the same processing when processing the data collected by the speech recognition target data collection unit to remove symbols that are not readable, and when adding readings to the original data for reading conversion. The dictionary generation system for speech recognition according to claim 1, wherein the dictionary generation system is created.

Images