JP2008191581A - Voice input support method and device, and navigation system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve operability and user-friendliness further by constructing a system for supporting a speech content. <P>SOLUTION: A voice input support method comprises: a first step (ST23) for recognizing voice by searching a second dictionary (a sound model DB14) in which a speech content is obtained and a sound model corresponding to a word is stored; and a second step (ST31) which outputs the voice recognizable word by searching a first dictionary (a word dictionary DB13) from a character string of the word which is input and obtained when the voice recognition fails. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention particularly relates to a voice input support method and apparatus, and a navigation system, which are suitable for use in an in-vehicle apparatus that performs facility search and destination setting.

  A speech recognition apparatus that recognizes speech uttered by a user by dictionary search and outputs synthesized speech has been widely used. Also in the field of car navigation, the above-described voice recognition device is usually used to improve operability and convenience. Speech recognition is usually realized by a mechanism that converts spoken speech into a feature value and converts it into a character string by a probabilistic method based on this feature value.

  FIG. 6 shows a basic flow of voice recognition processing in a conventional voice recognition system. In FIG. 6, reference numeral 60 denotes a voice recognition device. Reference numeral 61 denotes a word database (hereinafter simply referred to as a word dictionary DB 61) to be subjected to speech recognition, in which character string data is stored. Reference numeral 62 denotes an acoustic model database (hereinafter simply referred to as acoustic model DB 62) corresponding to the word dictionary DB 61, in which waveform feature amount data is stored.

  The speech recognition device 60 can recognize words, phrases, and sentences by searching the word dictionary DB 61 and the acoustic model DB 62 in combination. Specifically, the speech recognition device 60 performs a waveform feature amount matching process for speech recognition. This is a process of determining the similarity between the waveform feature value data of the user's utterance and the waveform feature value data of the recognized word, and there are methods such as DP (Dynamic Programming) and HMM (Hidden Marcov Model). As a result of similarity determination, the maximum likelihood waveform feature data is output as a speech recognition result, and the speech recognition result is delivered to the application. Examples of applications include control of air conditioners (temperature setting, air volume, wind direction, etc.), operation of recording media such as CD (Compact Disc) and DVD (Digital Versatile Disc) (next track, previous track, etc.) Car navigation system facility search and destination setting.

  By the way, in order to make up for incomplete speech recognition when setting a destination in an in-vehicle device, speech recognition is performed on a phoneme basis, the recognition result is input as a character string, and a database is searched (predetermined distance scale A technique is known in which the top M closest matching results are output and input by allowing a speaker to select one of them (see, for example, Patent Document 1).

JP 2006-39954 A

However, according to the technique disclosed in Patent Document 1 described above, since speech recognition is performed in units of phonemes and a database search is performed using a pseudonym as a key, a destination search for special reading such as a place name is performed. Is not valid. For example, if you want to search for “Hirakata Station”, if you say “Maki Kataeki” that people who do not live in the area will read, then the top M that is the result of matching the database search. “Hirakata Station” is not output individually, and as a result, the user who is the speaker cannot input by voice recognition.
Similarly, “Mita” may be read as “Mita” or “Sanda” depending on the region, and there is a problem that the expected destination search cannot be performed depending on the reading.

In addition, according to the technique disclosed in the above-mentioned Patent Document 1, considering the trouble of correcting an incomplete kana input (the trouble of specifying an incomplete position and replacing it), the reading is started from the beginning. It is likely that typing is faster. Furthermore, when performing a full text search, it is not necessary to input a full text string (full spelling) such as a destination. In this case, since reliable text input is performed from the beginning, the existence of an “error susceptibility database” essential as a configuration is meaningless.
Further, according to the technique disclosed in Patent Document 1, since speech recognition is basically collated with an acoustic model using a distance scale, there are always a plurality of recognition candidates. Therefore, it is necessary to select an optimum destination as the destination. On the other hand, when searching by character string input from the beginning, correct input can be performed from the beginning, so that the user is not required to select when the search result is uniquely determined.

  SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problem, and a voice input support method and apparatus that further improves operability and convenience by constructing a mechanism for supporting identification of utterance contents. And to obtain a navigation system.

  In order to solve the above-described problem, the speech input support method according to the present invention is a first step of acquiring speech content, searching a second dictionary in which an acoustic model corresponding to a word is stored, and performing speech recognition. And a second step of searching the first dictionary from the character string of the word that is input and acquired when the speech recognition fails and outputting a speech-recognizable word.

  In addition, the speech input support device according to the present invention includes a first dictionary composed of words capable of speech recognition, a storage unit storing a second dictionary composed of acoustic models corresponding to the words, and utterance contents. A speech recognition unit that acquires and performs speech recognition by searching the second dictionary stored in the storage unit, and a character string of the word that is input and acquired when the speech recognition fails, to the storage unit And a full-text search processing unit that searches the stored first dictionary and outputs a speech-recognizable word.

  In addition, the navigation system according to the present invention includes a storage unit storing a first dictionary composed of facility names or destination names capable of voice recognition, and a second dictionary composed of acoustic models corresponding to the facility names or destination names. , The name of the facility that is input and acquired when the speech content consisting of the facility name or the destination name is acquired, the second dictionary stored in the storage unit is searched for speech recognition, and the speech recognition fails. Alternatively, the first dictionary stored in the storage unit is searched from the character string of the destination name to output a facility name or destination name that can be recognized by voice, and the facility name search is performed based on the output facility name or destination name. Alternatively, a control unit that performs navigation by destination setting is provided.

  According to the present invention, it is possible to further improve operability and convenience by constructing a mechanism for supporting identification of utterance contents.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing an internal configuration of a voice input support apparatus according to Embodiment 1 of the present invention.
A speech input support device 10 according to Embodiment 1 of the present invention includes a speech input acquisition unit 11, a speech recognition engine unit 12, a word dictionary DB (database) 13, an acoustic model DB 14, a grammar dictionary DB 15, and an application. The program execution control unit 16 (hereinafter referred to as the application execution control unit 16), a key input acquisition unit 17, a full-text search processing unit 18, and a search result output unit 19 are configured.

The voice input acquisition unit 11 acquires a voice input by a user's utterance collected via a microphone (not shown), converts it into waveform feature data, and supplies the waveform feature data to the voice recognition engine unit 12. The speech recognition engine unit 12 combines a word dictionary DB 13 in which text data is stored and an acoustic model DB 14 in which waveform feature data necessary for comparing the similarity with the feature extraction result of the speech waveform is stored. To recognize phrases and sentences. The speech recognition engine unit 12 may further check a set of words, phrases, and sentences generated by a combination of the word dictionary DB 13 and the acoustic model DB 14 from the grammar data stored in the grammar dictionary DB 15.
That is, the speech recognition engine unit 12 performs a matching process between the waveform feature value data of the user's utterance and the waveform feature value data of the recognized word using a well-known DP or HMM technique, and the waveform feature of the maximum likelihood. A word, phrase, or sentence corresponding to the quantity data is delivered to the application execution control unit 16 as a voice recognition result. The application execution control unit 16 performs control of an air conditioner, operation of a medium such as a CD or a DVD, facility search or destination setting of a car navigation system.

  On the other hand, the key input acquisition unit 17 acquires a character string input from a soft keyboard (not shown) and supplies it to the full-text search processing unit 18. The full-text search processing unit 18 performs a full-text search of the word dictionary DB 13 using the character string as a key, outputs the search result to the search result output unit 19, and displays the search result on a display (not shown) by the search result output unit 19. Or, the user can hear the pronunciation through a speaker (not shown). As a result, the user can know the words, phrases or sentences that can be recognized by the speech recognition apparatus and the pronunciation method.

FIG. 2 is a diagram showing a flow of operation of the voice input support apparatus according to the first embodiment of the present invention. FIG. 2 also shows each procedure of the voice input support method according to the first embodiment of the present invention.
In FIG. 2, first, the user utters a reading (voice) necessary for executing the application, for example, necessary for facility name search and destination setting (step ST20). Receiving this, the speech input support device 10 acquires the user's utterance content as waveform feature amount data by the speech input acquisition unit 11, and supplies it to the speech recognition engine unit 12 (step ST21). The speech recognition engine unit 12 refers to the acoustic model DB 14 (step ST22), performs a matching process with the waveform feature value data by the user's utterance acquired using the DP or HMM technique, and performs the maximum likelihood waveform feature value. The data is delivered to the application execution control unit 16 as a voice recognition result (synthesized voice) (steps ST23 and ST24). At this time, the speech recognition engine unit 12 may check a set of words, phrases, and sentences generated by a combination of the word dictionary DB 13 and the acoustic model DB 14 from the grammar data stored in the grammar dictionary DB 15 (step ST25).

The application execution control unit 16 executes applications such as facility name search and destination setting based on the received recognition result, and outputs the result to a display or a speaker (not shown). When it is confirmed that the user who has viewed this has failed in voice recognition, the user inputs characters related to the utterance content (name of facility, destination, etc.) by operating a soft keyboard (not shown) (step ST26).
Upon receiving this, the voice input support device 10 acquires the character input described above by the key input acquisition unit 17 and supplies it to the full-text search processing unit 18 (step ST27). As a result, the full-text search unit 18 performs a full-text search with reference to the word dictionary DB 13 using the input and acquired character string as a key, and outputs the search result to the search result output unit 19 (step ST29). At this time, the full-text search processing unit 18 may output a search result by combining the grammar check with reference to the grammar dictionary DB 15 (step ST30). The grammar search result output unit 19 displays the search result on the display, or outputs the pronunciation of the search to a speaker or the like (step ST31). As a result, the user can know the words, phrases or sentences that can be recognized by the speech recognition engine unit 12 and the pronunciation method (step ST32).

As described above, according to the speech input support device according to the first embodiment of the present invention, speech recognition is possible by searching the word dictionary DB 13 from the character string of words that are input and acquired when speech recognition fails. By outputting words, the user knows a recognizable word or a set of phrases or sentences assumed by the voice recognition system by a character string search by key input of a soft keyboard or the like with higher certainty than voice. be able to.
As a result, a recognizable word, phrase or sentence can be spoken in the used speech recognition system, and as a result, obstacles to using the speech recognition system can be reduced. Also, an input object such as a set of words, phrases, and sentences that can be recognized by the speech recognition system being used is reproduced using an output device such as a display or a speaker, or TTS (Text To Speech). By presenting it to the user who is a speaker, etc., the user can learn an utterance method with a high recognition rate with respect to an input target capable of voice recognition, and the recognition rate can be improved by imitating the utterance method from the next time.

  According to the first embodiment described above, when speech recognition fails, a character string formed by reading a word is acquired based on a key input operation using a soft keyboard or the like. You may substitute by acquiring based on voice input operation. Moreover, according to Embodiment 1 described above, in addition to the word dictionary DB 13 and the acoustic model DB 14, the grammar dictionary DB 15 is also searched to recognize input objects such as words. Use is not essential, and only the word dictionary DB 13 and the acoustic model DB 14 are essential.

Embodiment 2. FIG.
FIG. 3 is a block diagram showing a configuration of a navigation system incorporating a voice input support device according to Embodiment 2 of the present invention.
As shown in FIG. 3, the navigation system 30 according to the second embodiment of the present invention includes a control unit 31, an input unit 32, a microphone 33, a position information acquisition unit 34, a map data storage unit 35, An audio output unit 36 and a display unit 37 are included.

The control unit 31 is realized by a microcomputer, a memory, and an application program that operates using the microcomputer. Specifically, the microcomputer acquires the utterance contents including the facility name or the destination name, searches the second dictionary stored in the storage unit, performs voice recognition, and is input and acquired when the voice recognition fails. The first dictionary stored in the storage unit is searched from the character string of the facility name or destination name to be output, and the facility name or destination name capable of voice recognition is output. Based on the output facility name or destination name, Navigation by facility name search or destination setting.
That is, the control unit 31 has each functional block as the voice input support device shown in FIG. 1, and the built-in memory as the storage unit has a first dictionary consisting of facility names or destination names that can be recognized by voice. In addition to the second dictionary consisting of acoustic models corresponding to facility names or destination names, a facility name or destination name database essential for navigation is added and stored (hereinafter collectively referred to as Called database).

  The input unit 32 is realized by a soft keyboard such as a touch panel, a touch pad, a remote controller, and a commander, and inputs contents operated by the user to the control unit 31. In addition, the microphone 33 collects the content of the user's utterance and inputs it to the control unit 31. The position information acquisition unit 34 is realized by a GPS (Global Positioning System) or a gyro, and acquires the current position data of the own vehicle and inputs it to the control unit 31. The map data storage unit 35 stores road data and facility data necessary for navigation. The voice output unit 36 outputs voices such as guidance guidance voice by navigation, voice recognition result repetition, navigation system state, operation end state, event, and the like. The display unit 37 is realized by a liquid crystal display or the like, on which a map display for navigation, menu information, and the like are displayed.

4 and 5 are flowcharts showing the operation of the navigation system according to Embodiment 2 of the present invention, which shows facility search and destination setting by voice recognition. Here, the destination intended by the user is “National Kyoto International Conference Center”, but the user assumes that the official name of the facility is “Kyoto International Conference Center”.
Hereinafter, the operation of the navigation system according to Embodiment 2 of the present invention will be described in detail with reference to the flow charts of FIGS. 4 and 5.

First, it is assumed that the user speaks “Kyoto International Conference Hall” (step ST400 “YES”), and the sound collected by the microphone 33 is input to the control unit 31 (step ST401). Thus, the control unit 31 recognizes the facility name spoken by the user by searching the database stored in the built-in memory using the speech recognition engine. The voice recognition method is the same as that of the first embodiment, and the description here is omitted to avoid duplication of explanation.
Here, when voice recognition fails (step ST402 “No”), the control unit 31 outputs a message such as “no corresponding facility” to the voice output unit 36 or the display unit 37 to the user. Notify that the voice recognition has failed (step ST403). Subsequently, the user who confirms the voice recognition result by display or the like determines whether or not to retry the voice recognition, and when the voice recognition is performed again, for example, the pressing of the speech key by the input unit 32 is detected ( Step ST404 “YES”), the process returns to step ST400.

If the user believes that the facility name is correct at “Kyoto International Conference Center”, he / she thinks that the speech recognition failure is bad for his / her speech method, and repeats the processes of steps ST400 to ST404. After repeating the retry for voice recognition several times, it is assumed that the user speaks the name of a different reading and the control unit 31 acquires it. For example, if the user utters “Kyoto Kaikan”, in this case, since the facility name “Kyoto Kaikan” exists in the database, the control unit 31 proceeds to the processing after step ST406 after successful speech recognition. it can.
When the voice recognition is successful (step ST402 “YES”), the control unit 31 notifies the user of the facility name recognized by the voice output unit 36 or the display unit 37 (step ST406). Here, the facility name is displayed on the display unit 37. Subsequently, for example, the user operates the input unit 32 or inputs a voice through the microphone 33 to instruct the control unit 31 to display a map (of the facility), and the facility name searched and displayed Confirm.

  Subsequently, the user confirms and determines whether or not it matches the intended facility (step ST407). This confirmation determination is made, for example, when the facility intended by the user is located in the northern part of Kyoto City (Kita Ward), but “Kyoto Kaikan” By visually confirming that there is, the user can confirm that the facility is not intended. As a result of the confirmation determination, if the facility is an intended facility (step ST407 “YES”), the control unit 31 is notified that the destination is set (step ST408). This can be realized by instructing “go there” by voice recognition or by selecting a menu such as “set to destination” displayed on the display unit 37. In this way, the destination setting or facility name search process is terminated, and then the control unit 31 executes the original process as a navigation system such as route search.

  On the other hand, if it is determined in step ST407 that the facility is not intended by the user ("NO" in step ST407), the process returns to step ST400 when the facility search is continued by voice recognition. If the full text search is given up with the facility name search by voice recognition, the process proceeds to step ST404. In addition, when the user gives up the voice recognition in step ST404 and decides to search by a character string input by operating the input unit 32 (step ST404 “NO”), the user inputs a character string related to the destination. This is notified to the control unit 31 (step ST405). This is realized, for example, by selecting a menu button displayed on the display unit 37 and having a content such as “Search by key input”.

Subsequently, the user operates the input unit 32 to input a character string of an intended facility name, and the control unit 31 acquires the content (step ST409 in FIG. 5). Here, for example, it is assumed that “Kyoto International Conference Hall” is entered. Thereby, the control unit 31 performs a full-text search of the database based on the character string input by the user, and displays the result on the display unit 37 (step ST410). Here, when a plurality of facilities are searched, they are displayed on the display unit 37 by a list display or the like to prompt the user to select. Thus, the user selects a desired facility name, but when a plurality of candidate facilities are displayed in a list, the user selects the target facility by moving the cursor or the like. Here, “National Kyoto International House” is the desired facility name.
Specifically, when a desired facility is selected (step ST411 “YES”), the user confirms the facility by a map display or the like (step ST412), and determines whether the facility is an intended facility (step ST413). If it is an intended facility (step ST413 “YES”), the process proceeds to step ST414. If it is an unintended facility (step ST413 “NO”), the process returns to step ST411. At this time, the user can know that the facility that was supposed to be the “Kyoto International Conference Hall” has the official facility name “National Kyoto International House”.

Subsequently, the user notifies the control unit 31 that the destination is “National Kyoto International House” (step ST414). The destination setting is realized, for example, by selecting a menu button with a content such as “Destination setting” displayed on the display unit 37. Thus, the destination setting or facility name search process is terminated, and the navigation system thereafter executes the original process such as route search.
In step ST414 described above, the user can learn that the facility name is not “Kyoto International Conference Center” but “National Kyoto International Conference Center”. In this case, “Kyoto International Conference Center” can be spoken from the beginning to be recognized, so that the destination can be easily set in a short time by using the steps ST400 to ST402 and ST406 to ST408. is there.

As described above, according to the voice input support device according to the second embodiment of the present invention, the navigation system presents the voice input target to the user via the output device such as a display or a speaker. The facility search or destination setting can be performed without knowing the input target that can be recognized from the beginning, and it is possible to know the speech method that the voice recognition system can easily recognize. Therefore, the operability and convenience in voice recognition can be improved. Can be improved.
In addition, since the input function based on voice recognition and the input function based on a soft keyboard have been designed independently, a set of words, phrases, and sentences to be input by the voice recognition function and a search function are provided. The set of words, phrases, and sentences to be searched may not match, but according to the second embodiment described above, by inputting the facility name as a character string, the input target can be matched, Therefore, the user can know a recognizable word or a set of phrases or sentences assumed by the voice recognition system by a character string search by key input. As a result, the user can utter a recognizable word, phrase or sentence in the voice recognition system being used, and as a result, the obstacles to using the voice recognition system can be reduced. .

  The speech input support method according to the present invention includes a first dictionary (word dictionary DB 13) that stores words that can be recognized and a second dictionary (acoustics) that stores acoustic models corresponding to the words. A speech input support method in a speech recognition system comprising a model DB 14). For example, in the flowchart shown in FIG. 2, speech content is acquired and speech recognition is performed by searching the second dictionary (acoustic model DB 14). The first step (steps ST20 to ST24) for performing the speech recognition, and the first dictionary (word dictionary DB13) can be searched from the character string of the word that is input and acquired when the voice recognition fails, and voice recognition is possible. A second step (steps ST26 to ST32) for outputting a simple word.

According to the voice input support method according to the present invention, the first dictionary (word dictionary DB 13) is searched from the character string of the words that are input and acquired when the voice recognition fails, and the voice-recognizable words are output. By this, the user can know a recognizable word, a set of phrases or sentences assumed by the voice recognition system by a character string search by key input such as a soft keyboard with higher certainty than voice, This makes it possible to utter words, phrases or sentences that are easy to recognize in the used speech recognition system, and as a result, it is possible to reduce obstacles in using the speech recognition system.
Also, an input object such as a set of words, phrases, and sentences that can be recognized by the speech recognition system being used is reproduced using an output device such as a display or a speaker, or TTS (Text To Speech). By presenting to the user who is a speaker, etc., the user can learn an utterance method with a high recognition rate with respect to an input target that can be recognized, and the recognition rate can be improved by imitating the utterance method from the next time It is.

Each of the voice input acquisition unit 11, the voice recognition engine unit 12, the application execution control unit 16, the key input acquisition unit 17, the full text search processing unit 18, and the search result output unit 19 illustrated in FIG. Specifically, the functions possessed are realized by the CPU constituting the voice input support device 10 sequentially reading and executing a program recorded in a built-in memory. At this time, each of the word dictionary DB 13, the acoustic model DB 14, and the grammar dictionary DB 15 is allocated and stored in a built-in memory or an external external memory.
In addition, the functions of the constituent blocks of the voice input support device 10 described above may be realized entirely by software as described above, or at least a part thereof may be realized by hardware. For example, the processing in the speech recognition engine unit 12 and the full-text search processing unit 18 may be realized on a computer by one or a plurality of programs, or at least a part thereof may be realized by hardware.

It is a block diagram which shows the internal structure of the audio | voice input assistance apparatus which concerns on Embodiment 1 of this invention. It is a figure which shows the flow of a process of the audio | voice input assistance method which concerns on Embodiment 1 of this invention. It is a block diagram which shows the internal structure of the navigation system incorporating the audio | voice input assistance apparatus which concerns on Embodiment 2 of this invention. It is a flowchart which shows operation | movement of the navigation system containing the audio | voice input assistance apparatus which concerns on Embodiment 2 of this invention. It is a flowchart which shows operation | movement of the navigation system containing the audio | voice input assistance apparatus which concerns on Embodiment 2 of this invention. It is a figure which shows the flow of the speech recognition process of the conventional speech recognition apparatus.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 Speech input support device, 11 Speech input acquisition part, 12 Speech recognition engine part, 13 Word dictionary DB, 14 Acoustic model DB, 15 Grammar dictionary DB, 16 Application execution control part, 17 Key input acquisition part, 18 Full text search process part , 19 Search result output unit, 30 Navigation system, 31 Control unit, 32 Input unit, 33 Microphone, 34 Location information acquisition unit, 35 Map data storage unit, 36 Audio output unit, 37 Display unit.

Claims (6)

A first dictionary in which words capable of speech recognition are stored, and a second dictionary in which acoustic models corresponding to the words are stored;
A first step of acquiring utterance content, searching the second dictionary and performing speech recognition;
A second step of searching the first dictionary from a character string of the word that is input and acquired when the speech recognition fails, and outputting a speech-recognizable word;
A voice input support method comprising:   2. The speech input support method according to claim 1, wherein the first step further recognizes the word by searching a grammar dictionary to be added. The second step includes
The voice input support method according to claim 1, wherein a character string of the word input based on a key input operation is acquired. The second step includes
The voice input support method according to claim 1, wherein a character string of the word input based on a voice input operation is acquired. A storage unit storing a first dictionary composed of words capable of speech recognition, and a second dictionary composed of acoustic models corresponding to the words;
A speech recognition unit that acquires utterance content, searches the second dictionary stored in the storage unit, and performs speech recognition;
A full-text search processing unit that searches the first dictionary stored in the storage unit from the character string of the word that is input and acquired when the speech recognition fails, and outputs a speech-recognizable word;
A voice input support device comprising: A storage unit for storing a first dictionary composed of facility names or destination names capable of voice recognition, a second dictionary composed of acoustic models corresponding to the facility names or destination names;
Acquires the utterance content consisting of the facility name or destination name, searches the second dictionary stored in the storage unit to perform speech recognition, and the facility name input and acquired when the speech recognition fails or Search the first dictionary stored in the storage unit from the character string of the destination name, output the facility name or destination name that can be recognized by voice, and search the facility name based on the output facility name or destination name or A control unit for performing navigation by setting a destination;
A navigation system characterized by comprising:

Images