WO2010086928A1 - 音声認識装置 - Google Patents
音声認識装置 Download PDFInfo
- Publication number
- WO2010086928A1 WO2010086928A1 PCT/JP2009/005488 JP2009005488W WO2010086928A1 WO 2010086928 A1 WO2010086928 A1 WO 2010086928A1 JP 2009005488 W JP2009005488 W JP 2009005488W WO 2010086928 A1 WO2010086928 A1 WO 2010086928A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- voice
- language
- recognition
- speech
- switching means
- Prior art date
Links
- 230000005477 standard model Effects 0.000 claims abstract description 58
- 238000000034 method Methods 0.000 claims description 31
- 238000010586 diagram Methods 0.000 description 8
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 230000005236 sound signal Effects 0.000 description 2
- 101000710013 Homo sapiens Reversion-inducing cysteine-rich protein with Kazal motifs Proteins 0.000 description 1
- 235000016496 Panda oleosa Nutrition 0.000 description 1
- 240000000220 Panda oleosa Species 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/005—Language recognition
Definitions
- the present invention relates to a speech recognition device, and more particularly to a speech recognition device that enables speech recognition using registered speech uttered by a user even when a language used for speech recognition is switched.
- a radio station name corresponding to a radio frequency is registered by voice, or a person name or place name corresponding to a telephone number is registered by voice.
- speech recognition collects speech data for each language and recognizes speech uttered by the user using a recognition algorithm or speech standard model constructed using that data. It is necessary to switch the recognition means itself and the voice standard model.
- the generated user dictionary depends on the speech recognition algorithm and the speech standard model. Once switched, the generated user dictionary cannot be used, and the user has to redo the voice registration operation.
- Patent Literature 1 discloses an electronic dictionary device that determines the language that the user wants to use by switching the language used by the electronic dictionary by collating the speech uttered by the user with the words stored in the system.
- a language switching method and a storage medium for an electronic dictionary device are disclosed.
- Patent Document 2 discloses a portable terminal device and a computer program that determine a language that a user wants to use by switching a language used in an electronic dictionary by comparing a voice uttered by the user with a word stored in the system. It is disclosed.
- the electronic dictionary device disclosed in the above-mentioned Patent Document 1 the language switching method of the electronic dictionary device, and the storage medium collate the speech uttered by the user with the words stored in the system for switching the language used in the electronic dictionary.
- the language that the user wants to use is determined, there is a problem in that a dictionary corresponding to each language cannot be created from the voice spoken by the user.
- Patent Document 2 the portable terminal device and the computer program disclosed in Patent Document 2 are intended to download appropriate dictionary data when entering a different language sphere, and the user corresponding to each language from the voice spoken by the user There was a problem that a dictionary could not be created.
- a voice label sequence corresponding to a registered voice uttered by a user is generated for each language and stored as a user dictionary.
- An object of the present invention is to provide a speech recognition device that can perform speech recognition using registered speech uttered by a user even if the language to be used is switched.
- a speech recognition device includes a speech standard model memory storing speech standard models corresponding to a plurality of languages to be recognized, a speech label memory storing user dictionaries corresponding to a plurality of languages, and a speech standard model.
- a first language switching means for switching and selecting a plurality of languages in the memory; a second language switching means for switching and selecting a plurality of languages in the voice label memory; a first language switching means and a second language switching means; Control means for switching in conjunction with each other, acoustic analysis means for performing acoustic analysis processing on the input user utterance voice, and outputting a feature vector time series of the user utterance voice, and a feature vector time series from the acoustic analysis means and the first And the voice standard model data corresponding to the language of the voice standard model memory selected while switching the language switching means of It generates a voice label string, in which a voice recognition unit for registering the voice label string in the first language switching means and conjunction with the second language language voice label memory which is switched
- the speech recognition apparatus includes a speech standard model memory storing a plurality of languages to be recognized and a speech label memory storing a plurality of languages by the first language switching means and the second language switching means which are linked.
- the feature vector time series of the input user uttered speech and the speech standard model data corresponding to the language of the speech standard model memory selected by the first language switching means while switching the language A voice label sequence for the user uttered voice is generated for each language, and this voice label string is registered in the language of the voice label memory switched and selected by the second language switching means.
- a speech label string to be generated can be generated for each language and stored as a user dictionary. As a result, by using this user dictionary, even if the language used for speech recognition is switched, there is an effect that the user can perform speech recognition using the uttered speech.
- FIG. 1 is a block diagram showing a configuration of a voice recognition device according to Embodiment 1.
- FIG. 6 is a flowchart for explaining processing at the time of user dictionary generation using the operation at the time of voice registration in the first embodiment.
- 5 is a flowchart for explaining an operation when performing speech recognition using processing at the time of speech recognition in the first embodiment. It is a block diagram which shows the structure of the speech recognition apparatus by Embodiment 2.
- FIG. 10 is a flowchart for explaining processing at the time of user dictionary generation using the operation at the time of voice registration in the second embodiment.
- 5 is a flowchart for explaining an operation when performing speech recognition using processing at the time of speech recognition in the first embodiment.
- FIG. 10 is a block diagram illustrating a configuration of a speech recognition apparatus according to Embodiment 3.
- FIG. 15 is a flowchart for explaining processing at the time of generating a user dictionary by using an operation in the case where the first user dictionary is generated in language n in the third embodiment.
- 14 is a flowchart for explaining an operation when performing speech recognition in a state where a user dictionary corresponding to language n is registered on the speech label memory in the third embodiment.
- It is a block diagram which shows the structure of the speech recognition apparatus by Embodiment 4.
- FIG. 15 is a flowchart for describing processing when a user dictionary corresponding to language n is generated using user registration voice in the fourth embodiment.
- 15 is a flowchart for explaining an operation when performing speech recognition in a state where a user dictionary corresponding to a language n is registered on the speech label memory according to the fourth embodiment.
- FIG. 1 is a block diagram showing the configuration of the speech recognition apparatus according to the first embodiment.
- the speech analysis model 4 and speech recognition unit 5 are connected to the speech standard model memory 1, speech label memory 2, and speech signal input terminal 3. , Control means 6, language switching means SW1 to SW3, and recognition / registration switching means SW4.
- the voice standard model memory 1 stores a voice standard model corresponding to a plurality of languages (language 1 to language N) to be recognized.
- the voice label memory 2 is a memory for storing a user dictionary corresponding to a plurality of languages (language 1 to language N).
- the user dictionary is stored on the voice label memory corresponding to a plurality of languages (language 1 to language N). Retained.
- the acoustic analysis unit 4 performs an acoustic analysis process on the user utterance voice input from the voice signal input terminal 3, and uses the power spectrum of the input user utterance voice or a time series of feature quantities similar thereto as the feature vector time. Output as a series.
- the speech recognition unit 5 uses the feature vector time series output from the acoustic analysis unit 4 as an input, and uses the data in the speech standard model memory 1 selected by the language switching unit SW1, and uses the feature vector time series.
- a voice label string corresponding to a time series is obtained, and the voice label string is stored as a user dictionary in languages 1 to N which are user dictionary storage destinations selected by the language switching means SW2.
- the speech recognition means 5 receives the feature vector time series output from the acoustic analysis means 4 as input and the data in the speech standard model memory 1 selected by the language switching means SW1 and the language switching means. Using the user dictionary corresponding to the language selected by SW3, speech recognition processing is performed on the feature vector time series, and the recognition result is output through the recognition / registration switching means SW4.
- the control means 6 controls the recognition / registration switching means SW4 to be switched to the recognition side based on a recognition switching instruction inputted from the outside, and when performing registration, that is, user dictionary generation, from the acoustic analysis means 4.
- the language switching means SW2 is switched by receiving a switching request signal from the voice recognition means 5 due to the interruption of the feature vector time series. At this time, the language switching unit SW1 that is linked to the language switching unit SW2 is also switched.
- the recognition / registration switching means SW4 is controlled to switch to the registration side based on a registration switching instruction input from the outside, and the language switching is performed so as to select a language input from the outside.
- the means SW2 is switched.
- the language switching unit SW1 that is linked to the language switching unit SW2 is also switched.
- the language switching means SW1 selects the voice standard model data corresponding to one of the voice standard models of a plurality of languages (language 1 to language N) stored in the voice standard model memory 1. To output automatically.
- the language switching means SW2 switches between the languages 1 to N, which are the storage destinations of the user dictionary in the voice label memory, in conjunction with the language switching means SW1 when the user dictionary is generated.
- the language switching means SW3 switches language 1 to language N, which is a reading destination of the user dictionary in the voice label memory, in conjunction with the language switching means SW1 when performing speech recognition.
- the recognition / registration switching means SW4 is switched to the side connected to the language switching means SW2 when a user dictionary generation instruction is given from the outside, and is switched to the output terminal side when there is a voice recognition instruction.
- FIG. 2 is a flowchart for explaining processing at the time of user dictionary generation using the operation at the time of voice registration in the first embodiment.
- the control means 6 switches the recognition / registration switching means SW4 to the registration side connected to the language switching means SW2 based on a user dictionary generation instruction inputted from the outside (step ST1).
- the output destination of the voice label string output by the voice recognition means 5 is set in the voice label memory 2.
- the registration voice spoken by the user is input from the voice signal input terminal 3 (step ST2).
- the input speech signal is acoustically analyzed by the acoustic analysis means 4, and a feature vector time series is output (step ST3).
- the initial value 1 is substituted into the variable n (step ST4).
- the language switching means SW1 and the language switching means SW2 are switched in conjunction with n.
- the speech standard model corresponding to the language represented by the variable n on the speech standard model memory 1 is set, and at the same time, the storage destination of the user dictionary corresponding to the language represented by the variable n is set on the speech label memory 2. (Step ST5).
- the speech recognition unit 5 receives the feature vector time series from the acoustic analysis unit 4 and generates a speech label sequence on the speech standard model memory 1 using data in the speech standard model memory 1 corresponding to the language represented by the variable n. (Step ST6). Next, the voice label sequence generated by the voice recognition unit 5 is stored in the user dictionary storage destination corresponding to the language represented by the variable n on the voice label memory 2 via the recognition / registration switching unit SW4 and the language switching unit SW2. Registered (step ST7).
- step ST8 It is determined whether the variable n matches the language number N (step ST8). If not, n + 1 is substituted for n and the process returns to step ST5 (step ST9). If the variable n matches the number of languages N, the process proceeds to step ST10. In step ST10, it is determined whether or not there is a next registration voice input. If the user speaks the next registration voice, the process returns to step ST2. When there is no utterance, the process at the time of generating the user dictionary ends.
- the registration voice uttered by the user to be registered automatically uses the data of the voice standard model memory 1 corresponding to the language 1 to language N and automatically uses the language 1 to language.
- the user dictionary corresponding to N is registered in the voice label memory 2.
- FIG. 3 is a flowchart for explaining the operation when speech recognition is performed using the speech recognition processing in the first embodiment.
- the control means 6 switches the language switching means SW1 and the language switching means SW3 so that a predetermined language can be recognized based on a voice recognition instruction inputted from the outside (step ST11). In the above example, one of the languages 1 to N is selected.
- the recognition / registration switching means SW4 is switched to the side for outputting the recognition result (step ST12).
- the voice uttered by the user is input from the voice signal input terminal 3 as an unknown input voice (step ST13).
- the input unknown input speech is acoustically analyzed by the acoustic analysis means 4, and a feature vector time series is output (step ST14).
- the speech recognition means 5 receives the feature vector time series obtained in step ST14 as input, and based on this feature vector time series, data in the speech standard model memory 1 and speech label memory 2 corresponding to the language set in step ST11.
- Recognition processing is performed using the user dictionary registered in (Step ST15), the recognition result is output via the recognition / registration switching means SW4, and the speech recognition processing is terminated.
- the language of the speech standard model memory used for creating the speech label sequence and the generated speech by the first language switching unit and the second language switching unit are generated for each language and stored as a user dictionary. As a result, even if the language used for speech recognition is switched, speech recognition using registered speech uttered by the user becomes possible.
- FIG. FIG. 4 is a block diagram showing the configuration of the speech recognition apparatus according to the second embodiment, in which language selection means 7 is added to the configuration of FIG. 1 showing the first embodiment, and other configurations are shown in FIG. Since the configuration is the same as that of the first embodiment, duplicate description is omitted.
- the language selection unit 7 selects a language based on the language designation data input by the user and gives it to the control unit 6. Based on the given language, the control means 6 controls the language switching means SW1 and the language switching means SW2 in conjunction with each other or the language switching means SW1 and the language switching means SW3 in conjunction with each other.
- FIG. 5 is a flowchart for explaining a process at the time of generating a user dictionary in the second embodiment. Since the operation from step ST21 to step ST24 is the same as the operation from step ST1 to step ST4 in the first embodiment, a duplicate description is omitted.
- the operation from step ST26 to step ST31 is the same as the operation from step ST5 to step ST10 in the first embodiment, and therefore a duplicate description is omitted.
- the language n is automatically registered in the speech label memory 2 as a user dictionary corresponding to the language n using the speech standard model memory 1 corresponding to the language n.
- FIG. 6 is a flowchart for explaining the operation when speech recognition is performed using the user dictionary generated as described above.
- the control means 6 interlocks and sets the language switching means SW1 and the language switching means SW3 to a predetermined language (step ST41).
- the recognition / registration switching means SW4 is switched to the recognition side (step ST42).
- the operation from step ST43 to step 45 is the same as the operation from step ST13 to step 15 in the first embodiment, and therefore a duplicate description is omitted.
- the language switching unit SW1 and the language switching unit SW2 that are linked to each other are switched based on the language designation data given from the outside.
- the language designation data given from the outside.
- FIG. 7 is a block diagram showing the configuration of the third embodiment of the present invention.
- the audio data memory 8 is a memory for recording and saving a registration audio input from the audio signal input terminal 3.
- the other voice standard model memory 1, voice signal input terminal 3, acoustic analysis means 4, voice recognition means 5, control means 6, language switching means SW1, and recognition / registration switching means SW4 are the same as those in the first embodiment shown in FIG. Because of this, duplicate explanation is omitted.
- FIG. 8 is a flowchart for explaining processing at the time of user dictionary generation, using the operation when the first user dictionary is generated in language n in the third embodiment.
- the control means 6 sets the language switching means SW1 at a position for selecting the language n based on the language n for generating the user dictionary inputted from the outside (step ST51).
- the recognition / registration switching means SW4 is switched to registration (step ST52).
- the output destination of the speech label string output by the speech recognition means 5 is set in the speech label memory 2a.
- the registration voice spoken by the user is input from the voice signal input terminal 3 (step ST53).
- the input registration voice is stored in the voice data memory 8 (step ST54).
- the input registration speech signal is acoustically analyzed by the acoustic analysis means 4, and the feature vector time series is output to the speech recognition means 5 (step ST55).
- the speech recognition means 5 inputs the feature vector time series and obtains a speech label string using the speech standard model data corresponding to the language represented by the variable n in the speech standard model memory 1 (step ST56).
- the voice label sequence generated by the voice recognition means 5 is registered as a user dictionary on the voice label memory 2a via the recognition / registration switching means SW4 (step ST57). Then, it is determined whether or not there is a next registration voice (step ST58). If the user speaks the next registration voice, the process returns to step ST53, and the operations of steps ST53 to ST57 are repeated. . If there is no utterance, the user dictionary generation process ends.
- the above is the processing when the first voice registration is performed in language n.
- the registration voice uttered by the user is stored in the voice data memory 8 and the user dictionary corresponding to the language n is registered in the voice label memory 2a.
- FIG. 9 shows a state where the voice data itself is stored in the voice data memory 8 and the user dictionary corresponding to the language n is registered in the voice label memory 2a for the registration voice spoken by the user as described above. It is a flowchart explaining the operation
- the control means 6 determines whether to change the recognition target language by switching the language switching means SW1 based on a voice recognition instruction inputted from the outside (step ST61). If not changed, that is, if the language switching means SW1 remains in the language n, the process proceeds to step ST62. In step ST62, the recognition / registration switching means SW4 is switched to recognition.
- the voice uttered by the user is input from the voice signal input terminal 3 as an unknown input voice (step ST63).
- the input unknown input speech is acoustically analyzed by the acoustic analysis means 4, and the feature vector time series is output to the speech recognition means 5 (step ST64).
- the speech recognition means 5 receives the feature vector time series obtained in step ST64 and performs recognition processing using the data in the speech standard model memory 1 corresponding to the language n and the user dictionary in the speech label memory 2a (step (ST65)
- the recognition result is output via the recognition / registration switching means SW4, and the voice recognition process is terminated.
- step ST61 the processing when the language switching means SW1 is set to a language other than n in step ST61 will be described. In this case, the process proceeds to step ST66, and the language switching means SW1 is set to a language n 'other than the language n. Then, the recognition / registration switching means SW4 is switched to registration (step ST67).
- the acoustic analysis means 4 reads out the voice data stored in the voice label memory 2a (step ST68). Then, the acoustic analysis means 4 obtains the feature vector time series for the read voice data and outputs it to the voice recognition means 5 (step ST69).
- the speech recognition means 5 receives the feature vector time series obtained in step ST69 as input, and uses the speech standard model data in the speech standard model memory 1 corresponding to the language n ′ set in step ST66, to produce a speech label sequence. Is generated (step ST70).
- step ST71 the generated voice label string is registered as a user dictionary in the voice label memory 2a (step ST71).
- step ST72 it is checked whether or not the audio data that has not been subjected to the processing from step ST69 to step ST71 remains (step ST72). If it remains, the process returns to step ST68, and the remaining audio data is subjected to steps ST69 to ST69. The process of step ST71 is repeated. If no audio data remains, the process moves to step ST62. Since the processing after step ST62 is the same as the above-described processing in which the language switching means SW1 is not switched while maintaining the language n, redundant description is omitted.
- a user dictionary corresponding to the language n ′ is immediately generated using the voice data recorded in the voice data memory 8 and stored in the voice label memory 2a. Therefore, even when the language n ′ is set, speech recognition using the user dictionary is possible. Further, in the third embodiment, since the voice label memory 2a only needs to be secured for one language, the memory can be reduced.
- the voice label for each language can be obtained by holding the registered voice uttered by the user and the acoustic analysis means 4 having the voice data memory 8 for acoustic analysis using the registered voice. Even when the language used by the speech recognition system is switched without having to secure the memory 2a, the registered speech uttered by the user recorded in the speech data memory 8 is used to correspond to the selected language. A user dictionary is created, and speech recognition using the user dictionary becomes possible.
- FIG. 10 is a block diagram showing the configuration of the fourth embodiment of the present invention, where the acoustic analysis means 4 outputs the registration voice uttered by the user instead of the voice data memory 8 in the third embodiment. Is provided with an acoustic data memory 9 for recording and storing the feature vector time series. Since other configurations are the same as the configurations of the third embodiment, the redundant description is omitted.
- FIG. 11 is a flowchart for explaining processing at the time of generating a user dictionary corresponding to language n using the user's registration voice, in the case of Embodiment 4 (when the first user dictionary is generated in language n). .
- the language switching means SW1 is set to the language n for generating the user dictionary (step ST81).
- the recognition / registration switching means SW4 is switched to registration (step ST82).
- the output destination of the speech label string output by the speech recognition means 5 is set in the speech label memory 2a.
- the registration voice spoken by the user is input from the voice signal input terminal 3 (step ST83).
- the acoustic analysis means 4 performs an acoustic analysis of the input registration audio signal and outputs a feature vector time series (step ST84).
- the feature vector time series obtained in step ST84 is recorded and saved on the acoustic data memory 9 (step ST85).
- the voice recognition means 5 receives the feature vector time series as an input, and generates a voice label sequence using the voice standard model data corresponding to the language represented by the variable n on the voice standard model memory 1 (step ST86).
- the voice label sequence generated by the voice recognition means 5 is registered as a user dictionary on the voice label memory 2a via the recognition / registration switching means SW4 (step ST87). Then, it is determined whether or not there is a next registration voice. If the user speaks the next registration voice, the process returns to step ST83, and the processes of steps ST83 to ST88 are repeated. If there is no utterance, the user dictionary generation process is terminated (step ST88).
- the above is the processing when the first voice registration is performed in language n.
- the feature vector time series of the registration speech uttered by the user is stored in the acoustic data memory 9, and the user dictionary corresponding to the language n is registered in the speech label memory 2a.
- the feature vector time series corresponding to the speech is stored in the acoustic data memory 9, and the user dictionary corresponding to the language n is stored on the speech label memory 2a. It is a flowchart explaining the operation
- control means 6 determines whether to change the language to be recognized by switching the language switching means SW1 based on a voice recognition instruction inputted from the outside (step ST91), and when the language switching SW1 remains in the language n. Then, the process proceeds to step ST92.
- step ST92 the recognition / registration switching means SW4 is switched to recognition.
- step ST93 the voice uttered by the user is input from the voice signal input terminal 3 as an unknown input voice.
- the input unknown input speech is acoustically analyzed by the acoustic analysis means 4, and the feature vector time series is output to the speech recognition means 5 (step ST94).
- the speech recognition means 5 receives the feature vector time series obtained in step ST94 and performs recognition processing using the data in the speech standard model memory 1 corresponding to the language n and the user dictionary in the speech label memory 2 (step ST95), the recognition result is output through the recognition / registration switching means SW4.
- step ST91 when the language switching means SW1 is selected and changed to other than language n, the process proceeds to step ST96, the language switching means SW1 is set to a language n ′ other than language n, and the recognition / registration switching means SW4 is registered. (Step ST97).
- the voice recognition means 5 reads the feature vector time series corresponding to the registration voice stored in the acoustic data memory 9 (step ST98).
- the speech recognition means 5 receives the read feature vector time series as an input, and generates a speech label sequence using the speech standard model data corresponding to the language n ′ set in step ST96 (step ST99).
- the obtained voice label string is registered as a user dictionary on the voice label memory 2a via the recognition / registration switching means SW4 (step ST100).
- step ST101 it is determined whether or not a feature vector time series that has not been subjected to the processing from step ST99 to step ST100 remains on the acoustic data memory 9 (step ST101). If so, the process returns to step ST98. Then, the processes of step ST99 to step ST101 are repeated for the remaining audio data. If no feature vector time series remains, the process proceeds to step ST92. Since the processing after step ST92 is the same as the above-described processing in which the language switching means SW1 is not switched while maintaining the language n, duplicate explanation is omitted.
- a user dictionary corresponding to the language n ′ is immediately generated using the feature vector time series in the acoustic data memory 9 and stored in the voice label memory 2a. Therefore, even when the language n ′ is set, speech recognition using the user dictionary is possible.
- the voice label memory 2a only needs to be secured for one language, so that the voice label memory can be reduced.
- the acoustic data memory 9 for storing the feature vector time series for which acoustic analysis has been completed is provided, the calculation amount can be reduced as compared with the case of the third embodiment, and user dictionary generation processing can be performed. Speed can be achieved.
- the feature vector time series obtained by the acoustic analysis by the acoustic analysis means 4 is held in the acoustic data memory 9, and the feature vector time series held in the acoustic data memory 9 is stored.
- the voice recognition means 5 generates a voice label string.
- the speech recognition apparatus can generate and maintain a user dictionary for each language from user uttered speech, the speech used when the user wants to convert a desired word into a foreign language and save it. Suitable for use in recognition devices and the like.
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Navigation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
Description
実施の形態1.
図1は、実施の形態1による音声認識装置の構成を示すブロック図であり、音声標準モデルメモリ1、音声ラベルメモリ2、音声信号入力端3に接続された音響分析手段4、音声認識手段5、制御手段6、言語切換手段SW1~SW3、認識/登録切換手段SW4を備えている。
図4は実施の形態2による音声認識装置の構成を示すブロック図であり、実施の形態1を示す図1の構成に言語選択手段7を付加したもので、他の構成は図1に示す実施の形態1の構成と同じであるから、重複説明は省略する。
次いで、言語指定データFIg(m)(m=1・・・Nについて定義されており、0もしくは1の値をとる)を用いて、言語選択手段7はFIg(n)=1であるか否かを判定し(ステップST25)、FIg(n)=1であればステップST26へ、FIg(n)=0であればステップST29の処理へ移行する。以下、ステップST26~ステップST31までの動作は実施の形態1におけるステップST5~ステップST10の動作と同じであるから、重複説明は省略する。
図7はこの発明の実施の形態3の構成を示すブロック図であり、音声ラベルメモリ2aは、ある言語n(n=1~N)に対応したユーザ辞書を1言語分だけ保存するためのメモリである。音声データメモリ8は、音声信号入力端3から入力されるところの登録用音声を記録保存しておくメモリである。他の音声標準モデルメモリ1、音声信号入力端3、音響分析手段4、音声認識手段5、制御手段6、言語切換手段SW1、認識/登録切換手段SW4は図1に示す実施の形態1の構成と同じであるから、重複説明は省略する。
まず、制御手段6は外部から入力されたユーザ辞書生成を行う言語nに基づいて、言語nを選択する位置に言語切換手段SW1を設定する(ステップST51)。次に、認識/登録切換手段SW4を登録に切換える(ステップST52)。これにより、音声認識手段5が出力する音声ラベル列の出力先が音声ラベルメモリ2aに設定される。
まず、制御手段6は外部から入力された音声認識指示に基づいて、言語切換手段SW1を切換えて認識対象言語を変更するかを判断する(ステップST61)。変更しない、つまり、言語切換手段SW1が言語nのままの場合、ステップST62の処理に移る。ステップST62では、認識/登録切換手段SW4を認識に切換える。次いで、ユーザが発話した音声を、未知入力音声として音声信号入力端3から入力する(ステップST63)。入力された未知入力音声を音響分析手段4において音響分析し、特徴ベクトル時系列を音声認識手段5に出力する(ステップST64)。音声認識手段5は、ステップST64で得られた特徴ベクトル時系列を入力し、言語nに対応する音声標準モデルメモリ1のデータと音声ラベルメモリ2a上のユーザ辞書を用いて認識処理を行い(ステップST65)、認識結果を認識/登録切換手段SW4を介して出力し、音声認識処理を終了する。
実施の形態4.
図11は実施の形態4において(言語nで最初のユーザ辞書生成を行う場合)を用いて、ユーザの登録用音声を用いて言語nに対応するユーザ辞書生成時の処理を説明するフローチャートである。まず、言語切換手段SW1をユーザ辞書生成を行う言語nに設定する(ステップST81)。
Claims (7)
- 認識対象とする複数の言語に対応した音声標準モデルを格納した音声標準モデルメモリと、
複数の言語に対応したユーザ辞書を保存する音声ラベルメモリと、
前記音声標準モデルメモリの複数の言語を切換選択する第1の言語切換手段と、
前記音声ラベルメモリの複数の言語を切換選択する第2の言語切換手段と、
前記第1の言語切換手段と第2の言語切換手段を連動して切換える制御手段と、
入力されたユーザ発話音声に対し音響分析処理を施し、前記ユーザ発話音声の特徴ベクトル時系列を出力する音響分析手段と、
前記音響分析手段からの特徴ベクトル時系列と前記第1の言語切換手段を切換えながら選択された前記音声標準モデルメモリの言語に対応する音声標準モデルのデータとにより前記入力されたユーザ発話音声に対する音声ラベル列を生成し、この音声ラベル列を前記第1の言語切換手段と連動して前記第2の言語切換手段で切換えられる前記音声ラベルメモリの言語に登録する音声認識手段と、
を備えた音声認識装置。 - 音声ラベルメモリの複数の言語を切換選択する第3の言語切換手段と、
音声認識手段の認識結果を外部へ出力するか音声ラベルメモリに登録するかを選択する認識/登録切換手段を備え、
制御手段は出力指示に基づいて前記認識/登録切換手段を認識に切換えるとともに、第1の言語切換手段と前記第3の言語切換手段を連動して切換え、音声認識手段は入力されたユーザ発話音声の特徴ベクトル時系列と前記第1の言語切換手段を切換えながら選択された前記音声標準モデルメモリの言語に対応する音声標準モデルのデータとにより前記入力されたユーザ発話音声に対する音声ラベルを前記音声ラベルメモリから読み出し、前記認識/登録切換手段を介して出力することを特徴とする請求項1記載の音声認識装置。 - 外部から与えられる言語指定データに基づいて言語を選択する言語選択手段を備え、
制御手段は、前記言語選択手段で選択された言語に基づいて該言語に対応する言語を選択するように、第1の言語切換手段と第2の言語切換手段を連動して切換えることを特徴とする請求項1記載の音声認識装置。 - 認識対象とする複数の言語に対応した音声標準モデルを格納した音声標準モデルメモリと、
ユーザ辞書を保存する音声ラベルメモリと、
前記音声標準モデルメモリの複数の言語を切換選択する第1の言語切換手段と、
指示言語に基づいて前記第1の言語切換手段を指示言語選択位置に切換える制御手段と、
ユーザが発話した音声を保持する音声データメモリと、
入力されたユーザ発話音声または前記音声データメモリから読み出したユーザ発話音声に対し音響分析処理を施し、前記ユーザ発話音声の特徴ベクトル時系列を出力する音響分析手段と、
前記第1の言語切換手段で選択された言語に対応する音声標準モデルのデータと前記特徴ベクトル時系列に基づいて、ユーザ発話音声に対する音声ラベルを生成し、この音声ラベルを前記音声ラベルメモリに登録する音声認識手段と、
を備えた音声認識装置。 - 認識対象とする複数の言語に対応した音声標準モデルを格納した音声標準モデルメモリと、
ユーザ辞書を保存する音声ラベルメモリと、
前記音声標準モデルメモリの複数の言語を切換選択する第1の言語切換手段と、
指示言語に基づいて前記第1の言語切換手段を指示言語選択位置に切換える制御手段と、
入力されたユーザ発話音声に対し音響分析処理を施し、前記ユーザ発話音声の特徴ベクトル時系列を出力する音響分析手段と、
前記音響分析手段で音響分析して得られた特徴ベクトル時系列を保持する音響データメモリと、
前記第1の言語切換手段で選択された言語に対応する音声標準モデルのデータと前記音響データメモリから読み出した特徴ベクトル時系列に基づいて、ユーザ発話音声に対する音声ラベルを生成し、この音声ラベルを前記音声ラベルメモリに登録する音声認識手段と、
を備えた音声認識装置。 - 音声認識手段の認識結果を音声ラベルメモリに登録するか外部へ出力するかを選択する認識/登録切換手段を備え、
制御手段は認識指示に基づいて前記認識/登録切換手段を認識に切換えるとともに、第1の言語切換手段を切換え、音声認識手段は入力されたユーザ発話音声の特徴ベクトル時系列と前記第1の言語切換手段を切換えながら選択された音声標準モデルメモリの言語に対応する音声標準モデルのデータとにより前記入力されたユーザ発話音声に対する音声ラベルを前記音声ラベルメモリから読み出し、前記認識/登録切換手段を介して出力することを特徴とする請求項4記載の音声認識装置。 - 音声認識手段の認識結果を音声ラベルメモリに登録するか外部へ出力するかを選択する認識/登録切換手段を備え、
制御手段は認識指示に基づいて前記認識/登録切換手段を認識に切換えるとともに、第1の言語切換手段を切換え、音声認識手段は入力されたユーザ発話音声の特徴ベクトル時系列と前記第1の言語切換手段を切換えながら選択された音声標準モデルメモリの言語に対応する音声標準モデルのデータとにより前記入力されたユーザ発話音声に対する音声ラベルを前記音声ラベルメモリから読み出し、前記認識/登録切換手段を介して出力することを特徴とする請求項5記載の音声認識装置。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
DE112009004313.5T DE112009004313B4 (de) | 2009-01-28 | 2009-10-20 | Stimmerkennungseinrichtung |
US13/119,445 US8099290B2 (en) | 2009-01-28 | 2009-10-20 | Voice recognition device |
CN2009801490344A CN102239517B (zh) | 2009-01-28 | 2009-10-20 | 声音识别装置 |
JP2010546177A JP4703787B2 (ja) | 2009-01-28 | 2009-10-20 | 音声認識装置 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2009-016783 | 2009-01-28 | ||
JP2009016783 | 2009-01-28 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2010086928A1 true WO2010086928A1 (ja) | 2010-08-05 |
Family
ID=42395198
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2009/005488 WO2010086928A1 (ja) | 2009-01-28 | 2009-10-20 | 音声認識装置 |
Country Status (5)
Country | Link |
---|---|
US (1) | US8099290B2 (ja) |
JP (1) | JP4703787B2 (ja) |
CN (1) | CN102239517B (ja) |
DE (1) | DE112009004313B4 (ja) |
WO (1) | WO2010086928A1 (ja) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2022182038A1 (ko) * | 2021-02-24 | 2022-09-01 | 삼성전자주식회사 | 음성 명령 처리 장치 및 방법 |
Families Citing this family (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8392189B2 (en) * | 2009-09-28 | 2013-03-05 | Broadcom Corporation | Speech recognition using speech characteristic probabilities |
US9786281B1 (en) * | 2012-08-02 | 2017-10-10 | Amazon Technologies, Inc. | Household agent learning |
US20140067366A1 (en) * | 2012-08-30 | 2014-03-06 | Google Inc. | Techniques for selecting languages for automatic speech recognition |
US9953630B1 (en) * | 2013-05-31 | 2018-04-24 | Amazon Technologies, Inc. | Language recognition for device settings |
JP6085538B2 (ja) * | 2013-09-02 | 2017-02-22 | 本田技研工業株式会社 | 音響認識装置、音響認識方法、及び音響認識プログラム |
WO2015075789A1 (ja) * | 2013-11-20 | 2015-05-28 | 三菱電機株式会社 | 音声認識装置および音声認識方法 |
CN104978015B (zh) * | 2014-04-14 | 2018-09-18 | 博世汽车部件(苏州)有限公司 | 具有语种自适用功能的导航***及其控制方法 |
US10515151B2 (en) * | 2014-08-18 | 2019-12-24 | Nuance Communications, Inc. | Concept identification and capture |
CN105957516B (zh) * | 2016-06-16 | 2019-03-08 | 百度在线网络技术(北京)有限公司 | 多语音识别模型切换方法及装置 |
CN106710586B (zh) * | 2016-12-27 | 2020-06-30 | 北京儒博科技有限公司 | 一种语音识别引擎自动切换方法和装置 |
WO2019049364A1 (ja) * | 2017-09-11 | 2019-03-14 | 三菱電機株式会社 | 音声認識辞書データ作成装置、音声認識装置、音声認識辞書データ作成方法 |
JP7117970B2 (ja) * | 2018-10-17 | 2022-08-15 | 株式会社日立ビルシステム | 案内ロボットシステム及び案内方法 |
US10885912B2 (en) * | 2018-11-13 | 2021-01-05 | Motorola Solutions, Inc. | Methods and systems for providing a corrected voice command |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2002304190A (ja) * | 2001-02-14 | 2002-10-18 | Sony Internatl Europ Gmbh | 発音変化形生成方法及び音声認識方法 |
JP2003022087A (ja) * | 2001-03-07 | 2003-01-24 | Sony Internatl Europ Gmbh | 音声認識方法 |
JP2004101727A (ja) * | 2002-09-06 | 2004-04-02 | Nippon Telegr & Teleph Corp <Ntt> | 多言語音声認識方法、装置、プログラム、および多言語話者適応方法、装置、プログラム |
Family Cites Families (32)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5640490A (en) * | 1994-11-14 | 1997-06-17 | Fonix Corporation | User independent, real-time speech recognition system and method |
DE19610848A1 (de) * | 1996-03-19 | 1997-09-25 | Siemens Ag | Rechnereinheit zur Spracherkennung und Verfahren zur rechnergestützten Abbildung eines digitalisierten Sprachsignals auf Phoneme |
US6085162A (en) * | 1996-10-18 | 2000-07-04 | Gedanken Corporation | Translation system and method in which words are translated by a specialized dictionary and then a general dictionary |
US6088669A (en) * | 1997-01-28 | 2000-07-11 | International Business Machines, Corporation | Speech recognition with attempted speaker recognition for speaker model prefetching or alternative speech modeling |
US6219641B1 (en) * | 1997-12-09 | 2001-04-17 | Michael V. Socaciu | System and method of transmitting speech at low line rates |
DE19847419A1 (de) * | 1998-10-14 | 2000-04-20 | Philips Corp Intellectual Pty | Verfahren zur automatischen Erkennung einer buchstabierten sprachlichen Äußerung |
JP2001005488A (ja) * | 1999-06-18 | 2001-01-12 | Mitsubishi Electric Corp | 音声対話システム |
JP2001101187A (ja) * | 1999-09-30 | 2001-04-13 | Sony Corp | 翻訳装置および翻訳方法、並びに記録媒体 |
US7725307B2 (en) * | 1999-11-12 | 2010-05-25 | Phoenix Solutions, Inc. | Query engine for processing voice based queries including semantic decoding |
JP2001282788A (ja) | 2000-03-28 | 2001-10-12 | Kyocera Corp | 電子辞書装置及び電子辞書装置の使用言語切替方法、記憶媒体 |
US7035805B1 (en) * | 2000-07-14 | 2006-04-25 | Miller Stephen S | Switching the modes of operation for voice-recognition applications |
JP2002247646A (ja) | 2001-02-19 | 2002-08-30 | Sony Corp | 携帯端末装置およびコンピュータプログラム |
JP3837061B2 (ja) * | 2001-11-28 | 2006-10-25 | 富士通株式会社 | 音信号認識システムおよび音信号認識方法並びに当該音信号認識システムを用いた対話制御システムおよび対話制御方法 |
DE10207895B4 (de) * | 2002-02-23 | 2005-11-03 | Harman Becker Automotive Systems Gmbh | Verfahren zur Spracherkennung und Spracherkennungssystem |
US7143033B2 (en) * | 2002-04-03 | 2006-11-28 | The United States Of America As Represented By The Secretary Of The Navy | Automatic multi-language phonetic transcribing system |
JP2004053742A (ja) * | 2002-07-17 | 2004-02-19 | Matsushita Electric Ind Co Ltd | 音声認識装置 |
US20040044517A1 (en) * | 2002-08-30 | 2004-03-04 | Robert Palmquist | Translation system |
DE10305369B4 (de) * | 2003-02-10 | 2005-05-19 | Siemens Ag | Benutzeradaptives Verfahren zur Geräuschmodellierung |
US7606714B2 (en) * | 2003-02-11 | 2009-10-20 | Microsoft Corporation | Natural language classification within an automated response system |
DE10334400A1 (de) * | 2003-07-28 | 2005-02-24 | Siemens Ag | Verfahren zur Spracherkennung und Kommunikationsgerät |
US20050197837A1 (en) * | 2004-03-08 | 2005-09-08 | Janne Suontausta | Enhanced multilingual speech recognition system |
US8036893B2 (en) * | 2004-07-22 | 2011-10-11 | Nuance Communications, Inc. | Method and system for identifying and correcting accent-induced speech recognition difficulties |
KR100640893B1 (ko) * | 2004-09-07 | 2006-11-02 | 엘지전자 주식회사 | 음성 인식용 베이스밴드 모뎀 및 이동통신용 단말기 |
US7340390B2 (en) * | 2004-10-27 | 2008-03-04 | Nokia Corporation | Mobile communication terminal and method therefore |
CA2597803C (en) * | 2005-02-17 | 2014-05-13 | Loquendo S.P.A. | Method and system for automatically providing linguistic formulations that are outside a recognition domain of an automatic speech recognition system |
ATE385024T1 (de) * | 2005-02-21 | 2008-02-15 | Harman Becker Automotive Sys | Multilinguale spracherkennung |
GB2424742A (en) * | 2005-03-31 | 2006-10-04 | Ibm | Automatic speech recognition |
DE102005018174A1 (de) * | 2005-04-19 | 2006-11-02 | Daimlerchrysler Ag | Verfahren zur gezielten Ermittlung eines vollständigen Eingabedatensatzes in einem Sprachdialog 11 |
CN101008942A (zh) * | 2006-01-25 | 2007-08-01 | 北京金远见电脑技术有限公司 | 机器翻译装置和机器翻译方法 |
KR100707970B1 (ko) * | 2006-03-10 | 2007-04-16 | (주)인피니티 텔레콤 | 이동통신단말기를 이용한 통역 서비스 방법 |
JP2007249653A (ja) * | 2006-03-16 | 2007-09-27 | Fujitsu Ltd | マークアップ言語情報の処理装置、情報処理方法、およびプログラム |
DE102006057159A1 (de) * | 2006-12-01 | 2008-06-05 | Deutsche Telekom Ag | Verfahren zur Klassifizierung der gesprochenen Sprache in Sprachdialogsystemen |
-
2009
- 2009-10-20 DE DE112009004313.5T patent/DE112009004313B4/de not_active Expired - Fee Related
- 2009-10-20 CN CN2009801490344A patent/CN102239517B/zh not_active Expired - Fee Related
- 2009-10-20 US US13/119,445 patent/US8099290B2/en not_active Expired - Fee Related
- 2009-10-20 JP JP2010546177A patent/JP4703787B2/ja not_active Expired - Fee Related
- 2009-10-20 WO PCT/JP2009/005488 patent/WO2010086928A1/ja active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2002304190A (ja) * | 2001-02-14 | 2002-10-18 | Sony Internatl Europ Gmbh | 発音変化形生成方法及び音声認識方法 |
JP2003022087A (ja) * | 2001-03-07 | 2003-01-24 | Sony Internatl Europ Gmbh | 音声認識方法 |
JP2004101727A (ja) * | 2002-09-06 | 2004-04-02 | Nippon Telegr & Teleph Corp <Ntt> | 多言語音声認識方法、装置、プログラム、および多言語話者適応方法、装置、プログラム |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2022182038A1 (ko) * | 2021-02-24 | 2022-09-01 | 삼성전자주식회사 | 음성 명령 처리 장치 및 방법 |
Also Published As
Publication number | Publication date |
---|---|
CN102239517A (zh) | 2011-11-09 |
US8099290B2 (en) | 2012-01-17 |
JP4703787B2 (ja) | 2011-06-15 |
US20110166859A1 (en) | 2011-07-07 |
DE112009004313T5 (de) | 2012-05-24 |
CN102239517B (zh) | 2013-05-08 |
JPWO2010086928A1 (ja) | 2012-07-26 |
DE112009004313B4 (de) | 2016-09-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP4703787B2 (ja) | 音声認識装置 | |
JP4942860B2 (ja) | 認識辞書作成装置、音声認識装置及び音声合成装置 | |
JP6686154B2 (ja) | 発話認識方法及び装置 | |
US7869999B2 (en) | Systems and methods for selecting from multiple phonectic transcriptions for text-to-speech synthesis | |
JP4316494B2 (ja) | 音声認識装置 | |
JP2000122691A (ja) | 綴り字読み式音声発話の自動認識方法 | |
JP3803029B2 (ja) | 音声認識装置 | |
JPH10143191A (ja) | 音声認識システム | |
JP2002006878A (ja) | 音声フレーズ認識方法及び音声認識装置 | |
JP3459712B2 (ja) | 音声認識方法及び装置及びコンピュータ制御装置 | |
JP2006048058A (ja) | 多言語による名称の音声認識のための方法とシステム | |
WO2011121649A1 (ja) | 音声認識装置 | |
JP4905522B2 (ja) | 機器制御装置、機器制御方法及びプログラム | |
JP2006058390A (ja) | 音声認識装置 | |
CN108417222B (zh) | 加权有限状态变换器解码***以及语音识别*** | |
JP2002049390A (ja) | 音声認識方法およびサーバならびに音声認識システム | |
JP2001154691A (ja) | 音声認識装置 | |
JP2008152043A (ja) | 音声認識装置及び音声認識方法 | |
US20090254335A1 (en) | Multilingual weighted codebooks | |
JP3478171B2 (ja) | 音声認識装置及び音声認識方法 | |
JP2004037813A (ja) | 車載用音声認識装置およびそれを用いた音声認識システム | |
CN117809621B (zh) | 一种语音合成方法、装置、电子设备及存储介质 | |
JP2006010849A (ja) | 音声合成装置 | |
WO2021019775A1 (ja) | 多言語音声認識装置および多言語音声認識方法 | |
KR20000071365A (ko) | 음성 인식 시스템에서 역추적 매트릭스 저장 방법 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 200980149034.4 Country of ref document: CN |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 09839120 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2010546177 Country of ref document: JP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13119445 Country of ref document: US |
|
WWE | Wipo information: entry into national phase |
Ref document number: 1120090043135 Country of ref document: DE Ref document number: 112009004313 Country of ref document: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 09839120 Country of ref document: EP Kind code of ref document: A1 |