JP2010085536A - Voice recognition system, voice recognition method, voice recognition client, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice recognition system which achieves both large vocabulary continuous speech recognition and recognition of local individual words. <P>SOLUTION: A client 1 includes: a voice input part 11 for inputting a voice signal; and a transmission part 12 for transmitting data representing the voice signal to a server 3. The server 3 includes: a reception part 32 for receiving data representing the voice signal from the client 1; a voice recognition part 33 for performing voice recognition in accordance with data received by the reception part 32; and a transmission part 34 for transmitting a result of voice recognition in the voice recognition part 33 to the client 1. Furthermore, the client 1 includes: a reception part 13 for receiving the result of voice recognition from the server 3; a voice recognition part 14 for performing voice recognition of data representing the voice signal corresponding to the received voice recognition result by referring to the voice recognition result received from the server 3 by the reception part 13; and a selection part 17 for selecting a recognition result both from the voice recognition result received from the server 3 and from the result of voice recognition in the voice recognition part 14. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a speech recognition system. More specifically, the present invention relates to a speech recognition system composed of a server and a client.

  There is a voice recognition system including a server and a client in order to improve voice data voice recognition performance between apparatuses on a communication network. For example, the voice recognition communication system of Patent Document 1 receives a voice input device that inputs a voice signal and outputs voice information indicating the voice signal as a packet to a network line, and a packet transmitted through the network line. A voice recognition device that performs voice recognition. The voice input device transmits a packet using one of at least two types of communication protocols having different degrees of reliability, and the voice recognition device receives the transmitted packet according to the one communication protocol. To do.

  Patent Document 2 describes a technique for performing speech recognition while suppressing a data transfer amount between a client and a server. In the technique of Patent Document 2, the client compresses and encodes the feature parameter by the voice processing unit, and transmits the feature parameter compressed and encoded by the interface. The server receives the feature parameter compressed and encoded by the interface, and recognizes the feature parameter compressed and encoded by the speech processing unit.

  There is a technology of a voice recognition system that solves a problem of a load increase of a server-side CPU and a load increase of a network bandwidth between a client and a server, which are problems in a voice recognition system including a server and a client. For example, in the speech recognition system of Patent Document 3, the client attempts primary speech recognition on input speech data, and if speech recognition is possible by primary speech recognition, transmits the speech recognition result to a server, If the voice data cannot be recognized, the voice data is transmitted to the server, and the server performs secondary voice recognition on the voice data.

JP 2001-142488 A JP 2001-337695 A JP 2004-12653 A

  In speech recognition in a call or the like, it is desired to recognize speech at a normal speaking speed (continuous) for general conversation (large vocabulary). When inputting data by voice for general use, large vocabulary continuous voice recognition is necessary, for example, for dictation writing, spoken word analysis or voice interpretation. Large vocabulary continuous speech recognition requires high computing power and is difficult for clients with limited computing power and storage capacity.

  On the other hand, vocabularies used locally and individually such as proper nouns are context-dependent and have many words specific to the client, and there is a limit to how they can be recognized by the server. Therefore, it is required to suppress degradation of speech recognition accuracy with unknown words that are not registered in the server, while corresponding to the limited arithmetic processing capability of the client that requires the speech recognition result.

  The present invention has been made in view of the above situation, and an object of the present invention is to provide a speech recognition system that achieves both large vocabulary continuous speech recognition and local individual word recognition.

  In order to achieve the above object, a speech recognition system according to a first aspect of the present invention is a speech recognition system including a server and a client, wherein the client includes speech input means for inputting a speech signal, Transmitting means for transmitting data indicating an audio signal to a server, wherein the server receives a server receiving means for receiving data indicating the audio signal from the client, and a server for recognizing voice from the data received by the server receiving means Voice recognition means, and server transmission means for sending the result of voice recognition by the server voice recognition means to the client, the client further receiving means for receiving the voice recognition result from the server; The received voice recognition result with reference to the voice recognition result received from the server by the receiving means A voice recognition means for recognizing the corresponding data indicating the transmitted voice signal; a selection means for selecting a recognition result from a voice recognition result received from the server; and a result of voice recognition by the voice recognition means; , Comprising.

  According to the present invention, it is possible to perform speech recognition on local individual words while performing continuous speech recognition by the server.

  Preferably, the server speech recognition means extracts phrase information including data representing a phrase break position and a phrase attribute from data indicating the voice signal, and the server transmission means transmits the phrase information to the client. Then, the voice recognition unit of the client refers to the phrase information received from the server and recognizes the voice of the data indicating the transmitted voice signal.

  As a result, the efficiency of voice recognition on the client side can be improved.

  Preferably, the phrase information includes data representing a part of speech of a word included in the phrase, and the voice recognition unit of the client refers to data representing a part of speech of the phrase information received from the server, and transmits the transmitted voice Voice recognition is performed on data indicating a signal.

  Thereby, the accuracy of voice recognition on the client side can be improved.

  Preferably, the server speech recognition unit estimates an attribute of an unknown word among results of speech recognition from the data indicating the speech signal, and the server transmission unit estimates the attribute of the unknown word estimated by the server speech recognition unit. Is transmitted to the client, and the voice recognition means of the client refers to the attribute of the unknown word received from the server and performs voice recognition on the data indicating the transmitted voice signal.

  Since the client recognizes the voice by referring to the unknown word attribute of the server, the voice recognition processing operation of the client can be speeded up. As a result, the efficiency of speech recognition can be improved for unknown words.

  Preferably, the voice recognition unit of the client performs voice recognition when a voice recognition result received from the server meets a predetermined condition.

  As a result, it is possible to reduce the load on the client for voice recognition.

  Preferably, when the speech recognition result received from the server and the speech recognition result by the speech recognition means have a plurality of candidates for one word, the client preferably has at least two candidates among the plurality of candidates. Display means for displaying and an input means for receiving a command for selecting at least one of the candidates displayed by the display means, wherein the selection means is based on the selection command input by the input means. To select the recognition result.

  Thereby, the voice recognition result can be selected more accurately by the user's judgment.

  Preferably, the client includes an extraction unit that extracts a feature parameter of speech input by the input unit, and the transmission unit transmits the feature parameter extracted by the extraction unit to the server, and the server speech recognition unit Recognizes speech using feature parameters received from the client.

  As a result, the amount of data transferred between the client and the server can be suppressed.

  A speech recognition method according to a second aspect of the present invention is a speech recognition method in a speech recognition system composed of a server and a client, wherein a speech input step of inputting a speech signal at the client; A transmission step of transmitting data to the server, a server reception step of receiving data indicating a voice signal from the client at the server, a server voice recognition step of recognizing voice from the data received at the server reception step, Refer to the server transmission step for transmitting the result of speech recognition in the server speech recognition step to the client, the reception step for receiving the speech recognition result from the server at the client, and the speech recognition result received in the reception step. The data indicating the transmitted audio signal To the speech recognizing speech recognition step, and the speech recognition result received from the server, characterized in that it comprises a selection step of selecting the result of the speech recognition, the recognition result from by the speech recognition step.

  A voice recognition client according to a third aspect of the present invention includes a voice input unit that inputs a voice signal, a transmission unit that transmits data indicating the voice signal to a server, and a reception that receives a result of voice recognition from the server. The speech recognition result received from the server, and the speech recognition means with respect to the data indicating the transmitted speech signal, with reference to the speech recognition result received by the means and the reception means. And a selection means for selecting a recognition result from the result of speech recognition.

  Preferably, the receiving means receives phrase information including data representing a phrase break position and a phrase attribute from the server, and the speech recognition means refers to the phrase information received by the receiving means, and Voice recognition is performed on data indicating the transmitted voice signal.

  Preferably, the phrase information includes data representing a part of speech of a word included in the phrase, and the speech recognition unit indicates the transmitted voice signal with reference to data representing a part of speech of the received phrase information. Recognize voice for data.

  Preferably, the receiving unit receives an attribute of an unknown word in the speech recognition result received from the server, and the speech recognition unit refers to the attribute of the unknown word received by the receiving unit, and Voice recognition is performed on data indicating the transmitted voice signal.

  Preferably, the voice recognition means performs voice recognition when a voice recognition result received from the server meets a predetermined condition.

  Preferably, when the speech recognition result received from the server and the speech recognition result by the speech recognition means have a plurality of candidates for one word, a display for displaying at least two candidates among the plurality of candidates And an input means for receiving a command for selecting at least one of the candidates displayed by the display means, wherein the selection means is configured to recognize the recognition based on the selection command input by the input means. Select a result.

  Preferably, an extraction unit that extracts a feature parameter of speech input by the input unit is provided, and the transmission unit transmits the feature parameter extracted by the extraction unit to the server.

  A program according to a fourth aspect of the present invention receives a result of speech recognition from a server, a voice input means for inputting a voice signal, a transmission means for transmitting data indicating the voice signal to a server, and the computer. Receiving means, referring to a voice recognition result received by the receiving means, voice recognition means for recognizing the data indicating the transmitted voice signal, a voice recognition result received from the server, and the voice recognition means And a result of the voice recognition performed in step (1), and a selection means for selecting a recognition result from the result.

  According to the speech recognition system of the present invention, it is possible to recognize speech even for individual local words while performing continuous speech recognition by the server.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the drawings, the same or corresponding parts are denoted by the same reference numerals. In the present invention, speech recognition is a process of extracting the content being spoken as character data from a speech language.

(Embodiment 1)
FIG. 1 is a block diagram showing a configuration of a speech recognition system according to Embodiment 1 of the present invention. As shown in FIG. 1, the voice recognition system includes a client 1 and a server 3. The client 1 and server 3 communicate via a network (not shown). In FIG. 1, a single client 1 is shown as a representative. A plurality of clients 1 can communicate with the server 3.

  The client 1 inputs voice and transmits voice data to the server 3. The server 3 recognizes the voice from the received voice data and transmits the result to the client 1. The client 1 refers to the voice recognition result received from the server 3 and performs voice recognition on the same voice data. Then, an appropriate recognition result is selected for each voice portion from the plurality of voice recognition results.

  The client 1 includes a control unit 10, a voice input unit 11, a transmission unit 12, a reception unit 13, a voice recognition unit 14, a display unit 15, an input unit 16, a selection unit 17, a microphone 21, a display device 22, and an input device 23. Is done. The voice input unit 11, the transmission unit 12, the reception unit 13, the voice recognition unit 14, the display unit 15, the input unit 16, and the selection unit 17 are connected to the control unit 10 via the internal bus 20. The microphone 21 is connected to the audio input unit 11. The display device 22 is connected to the display unit 15. The input device 23 includes a keyboard or a numeric keypad, a mouse or a cross cursor key, an enter key, and the like, and is connected to the input unit 16. The control unit 10 controls the entire client 1.

  The server 3 includes a control unit 31, a reception unit 32, a voice recognition unit 33, and a transmission unit 34. The reception unit 32, the voice recognition unit 33, and the transmission unit 34 are connected to the control unit 31 via the internal bus 30. The receiving unit 32 receives data from the transmitting unit 12 of the client 1 via a network (not shown). The transmission unit 34 transmits data to the reception unit 13 of the client 1 via the network. The control unit 31 controls the entire server 3.

  The voice input unit 11 A / D converts the voice signal input from the microphone 21 to generate voice data. The voice input unit 11 may further convert the voice data into a compressed code. The voice data transmitted to the server 3 may be data obtained by performing A / D conversion on a voice signal, further encoded data, or feature amount data extracted in the first step of voice recognition processing from voice. In the following description, they are collectively referred to as audio data.

  The voice input unit 11 sends voice data to the transmission unit 12. The same voice data is sent to the voice recognition unit 14. The transmission unit 12 transmits the audio data to the server 3.

  The reception unit 32 of the server 3 receives audio data from the transmission unit 12 of the client 1. The receiving unit 32 sends the voice data received from the client 1 to the voice recognition unit 33. The voice recognition unit 33 recognizes voice from the received voice data and converts it into a character string.

  In speech recognition, in general, acoustic features and linguistic features of speech are often handled separately. Using the acoustic features, phonemes are extracted based on the frequency characteristics of the phonemes to be recognized. As a representation of an acoustic model representing an acoustic feature, a hidden Markov model with a mixed normal distribution as an output probability is widely used.

  A phrase or word is discriminated by using a language model or rule grammar representing restrictions on how to arrange phonemes, which is a linguistic feature, and referring to dictionary data. At this time, the phrase time information is also acquired.

  The speech recognition unit 33 can use large vocabulary continuous speech recognition, small vocabulary word speech recognition, or the like according to the purpose of speech recognition. In general speech recognition, large vocabulary continuous speech recognition is desirable.

  The voice recognition unit 33 sends the result of voice recognition to the transmission unit 34. Here, the result of speech recognition includes information on the segment break position. In addition, the result of speech recognition may include information on phrase attributes, for example, character type such as kanji or katakana, word part of speech, or phrase time. The transmission unit 34 transmits the result of speech recognition to the client 1.

  The reception unit 13 of the client 1 receives the result of speech recognition from the server 3 and sends it to the speech recognition unit 14. The voice recognition unit 14 performs voice recognition on the same voice data with reference to the voice recognition result of the server 3. Similarly to the voice recognition unit 33 of the server 3, the voice recognition unit 14 can employ an appropriate voice recognition method according to the purpose of voice recognition. Further, a voice recognition method different from the voice recognition unit 33 of the server 3 can be used. Of the speech recognition results of the server 3, for example, isolated speech recognition can be performed for each phrase by using the segment break position.

  In the speech recognition of the client 1, speakers, fields, contexts, and the like can be limited. In addition, since local individual words can be registered in the speech recognition unit 14, recognition accuracy can be improved for proper nouns and speaker-specific expressions. As will be described later, by selecting a speech recognition result of the server 3 or a word input by the user and registering the word in the speech recognition unit 14, the speech recognition of the client 1 is learned so as to be adapted to the usage content of the user. Can be made. The speech recognition unit 14 sends the speech recognition result received from the server 3 and the result of speech recognition performed by the speech recognition unit 14 to the display unit 15.

  The display unit 15 displays the voice recognition results of the server 3 and the voice recognition unit 14 on the display device 22. Here, the same words in the two speech recognition results are merged and displayed, and the words having different recognition results are written side by side. Moreover, when there are a plurality of candidates in each speech recognition result, they may be displayed.

  The input unit 16 accepts an input of an instruction to select from the input device 23 for words with different displayed speech recognition results. Moreover, a user's character input is received about words with an unknown recognition result or words, such as a recognition error. Then, an instruction to select and a character input are sent to the selection unit 17. The selection unit 17 determines a word for a word having a plurality of speech recognition result candidates. The selection unit 17 may select a result having a higher likelihood for each word among the recognition results.

  The selection unit 17 displays the result of selection and confirmation from the display unit 15. The recognition result selected and confirmed by the selection unit 17 is displayed on the display unit 15 and can be used as an input for an e-mail or a memo. The dictionary data of the voice recognition unit 14 may be updated according to the selected result or character input. By selecting words from the speech recognition result of the server 3 or words inputted by the user and registering the words in the speech recognition unit 14, the voice recognition of the client 1 can be learned so as to be adapted to the usage contents of the user.

  FIG. 2 shows an example of the result of speech recognition. In FIG. 2, DSR (Distributed Speech Recognition) represents voice recognition of the server 3. LSR (Local Speech Recognition) represents voice recognition on the client 1 side. The feature amount may be extracted by the client 1 and the feature amount may be transmitted to the server 3, or may be independently performed by the client 1 and the server 3.

  It is shown that the server 3 recognizes “I / is / island / ni / live / is /.” With respect to the utterance “I am Hashimaniikimasu”. The number of the phrase time information below the recognition result represents the time from the start of the speech voice in seconds.

  The client 1 divides the voice with reference to the phrase time information of the voice recognition result received from the server 3, and recognizes the isolated voice for each divided part. As a result, it is shown that “cotton / ha / shima / ni / go / mari” was recognized.

  FIG. 3 shows an example of displaying a speech recognition result and selecting a recognition result. The example of FIG. 2 is used for voice and recognition results. The same words in the voice recognition result of the server 3 and the voice recognition result of the client 1 are displayed in a merged manner, and different words are written side by side (the third frame from the top in FIG. 3).

  The user sees the displayed recognition result and selects the correct one. The recognition result is confirmed by the user's selection. As a result, “I will go to Shima” is selected. The user can obtain desired text data by a simple operation of selecting and determining a recognition result candidate using a selection key (such as a cross cursor key) and a determination key.

  By registering local individual words for each user in the dictionary of the voice recognition unit 14 of the client 1, even a similar pronunciation word recognized as a general word by the server 3, Local individual words can be displayed in the choices as recognition result candidates. Further, by selecting a speech recognition result of the server 3 or a word inputted by the user and registering a word in the speech recognition unit 14, the speech recognition of the client 1 can be learned so as to be adapted to the use contents of the user. it can. As a result, speech recognition can be performed more accurately even for utterance contents including words unique to the client 1 side.

  FIG. 4 is a flowchart illustrating an example of the operation of the speech recognition system according to the first embodiment. First, a voice is input by the voice input unit 11 of the client 1 (step S11), and the voice data is transmitted from the transmission unit 12 to the server 3 (step S12).

  In the server 3, when the voice data is received by the receiving unit 32 (step T11), the voice recognition unit 33 performs voice recognition processing (step T12). Then, the speech recognition result is transmitted to the client 1 by the transmission unit 34 (step T13).

  When the receiving unit 13 of the client 1 receives the speech recognition result from the server 3 (step S13), the speech recognizing unit 14 performs a speech recognition process with reference to the result (step S14). The display unit 15 displays the speech recognition results of the server 3 and the client 1 (step S15).

  The input unit 16 receives a selection input from the user (step S16), and the selection unit 17 selects and confirms the recognition result according to the input instruction (step S17).

  As described above, according to the speech recognition system of the first embodiment, it is possible to perform speech recognition on local individual words while performing continuous speech recognition by the server 3. As a result, the voice recognition rate can be improved. Further, by referring to the voice recognition result of the server 3, the voice recognition efficiency on the client 1 side can be improved.

  In addition, the speech recognition system of Embodiment 1 has the following effects. Commonly used vocabularies, language models, or rule grammars are reflected on the server 3, and individual words and rules are reflected on the client 1 side, so that the resources of the server 3 and the client 1 can be used effectively. Further, by selecting a speech recognition result of the server 3 or a word inputted by the user and registering a word in the speech recognition unit 14, the speech recognition of the client 1 can be learned so as to be adapted to the use contents of the user. it can. And the update of the voice recognition dictionary on the client 1 side does not affect the server 3 side, and the flexibility of the system is high.

  Furthermore, the speech recognition model used by the speech recognition unit 14 of the client 1, that is, the acoustic model, the language model, the dictionary, and the like can be arbitrarily changed by the user. By setting the voice recognition model of the client 1 according to the target of voice recognition, the accuracy and efficiency of voice recognition can be improved.

  In the speech recognition system of the present invention, the speech recognition algorithm on the client 1 side can be selected in accordance with the processing capability on the client 1 side. For example, when the processing capability on the client 1 side is low, isolated voice recognition can be used. If the processing capability on the client 1 side is high, continuous speech recognition may be used.

(Modification 1 of Embodiment 1)
FIG. 5 shows an example of recognition result selection when part-of-speech information is used. The speech of the speech recognition target in FIG. 5 is “Anohito Gakenzides.” From the server 3, it is shown that “prosecutor” of the common noun “kenji” is transmitted as a voice recognition result.

  In the client 1, for example, it is assumed that the proper noun “Kenji” is a candidate for “Kenji” as a result of speech recognition for each phrase. When the recognition result of the server 3 is a common noun and the speech recognition result of the client 1 is a proper noun, the selection unit 17 selects the recognition result of the proper noun client 1.

  By referring to the part-of-speech information, the accuracy of voice recognition on the client 1 side can be improved.

(Modification 2 of Embodiment 1)
FIG. 6 shows an example of recognition result selection when unknown word information is used. An unknown word means that the word registered in the voice recognition unit 33 has no word corresponding to the voice. If the speech recognition unit 33 determines that the word is an unknown word as a result of the speech recognition, the speech recognition unit 33 may recognize the part of speech information that the word will have as an attribute. Further, in many cases, unknown words can recognize phoneme strings, and information on whether a character string corresponding to the phoneme includes kanji or katakana may be recognized as an attribute. The voice recognition unit 33 transmits the attributes of these unknown words to the client 1.

  The speech of the speech recognition target in FIG. 6 is “Anohito Katachimorides.” From the server 3, it is transmitted as a speech recognition result that “Tachimori” is an unknown word that is not registered in the server 3.

  As a result of the voice recognition in the client 1, it is assumed that the proper noun “Sun Moon” is a candidate for “Tachimori”. Since the recognition result of the server 3 is an unknown word, the selection unit 17 selects the recognition result of the client 1.

  The client 1 may perform speech recognition only for unknown words or words with low likelihood of the speech recognition result received from the server 3. If the recognition result of the server 3 is not an unknown word, it is considered that the recognition result is probable. In this case, the client 1 does not perform voice recognition, thereby speeding up the voice recognition processing operation of the client 1. it can. Further, when the recognition result of the server 3 is an unknown word, the recognition result is considered to be inaccurate. In this case, the speech recognition rate can be improved by performing speech recognition with the client 1. By referring to information such as the part of speech of the unknown word received from the server 3, the attribute can be limited, so that the voice recognition processing operation of the client 1 can be speeded up. The example of FIG. 6 shows that the speech recognition rate can be improved at the client 1 for local individual words.

  As described above, according to the speech recognition system of the first embodiment, the speech recognition speed can be improved as a whole.

  Note that the voice recognition unit 14 of the client 1 may actually be on the server 3 side. For example, the server 3 may be incorporated as an agent or module for each client 1. The voice recognition unit 14 may be configured such that a voice recognition dictionary and rule grammar for each client 1 are stored as data of the server 3, and a common module is executed by the server 3 as a process for each client 1. In that case, the transmission unit 34 of the server 3 transmits the speech recognition result of the speech recognition unit 33 and / or the speech recognition result of the speech recognition unit 14. The reception unit 13 of the client 1 receives the voice recognition result of the voice recognition unit 33 and / or the voice recognition result of the voice recognition unit 14, and selects either the client 1 or the voice of the voice recognition unit 33. The recognition result and / or the voice recognition result selected from the voice recognition unit 14 is received.

  Similarly, the voice recognition unit 33 of the server 3 can be incorporated into the client 1. In this case, data exchange among the transmission unit 12, the reception unit 32, the transmission unit 34, and the reception unit 13 is data transmission within the apparatus or communication between processes. When the speech recognition unit 33 is in the client 1, it can be considered as a back-end processor or a core part of speech recognition processing. The voice recognition unit 14 can be regarded as a part that realizes a dictionary or rule grammar for each user.

(Embodiment 2)
FIG. 7 is a block diagram showing the configuration of the speech recognition system according to the second embodiment. In the second embodiment, the client 1 includes a result determination unit 18.

  The result determination unit 18 examines the voice recognition result of the server 3 and determines whether or not the voice recognition unit 14 recognizes the voice. For example, when the likelihood representing the likelihood is transmitted for each word as a result of the speech recognition from the server 3, when the likelihood is equal to or greater than a predetermined threshold, the speech recognition unit 14 performs speech for the word. Do not recognize.

  Whether or not speech recognition is performed by the speech recognition unit 14 can be determined using likelihood, unknown word or part of speech, or a condition that combines them. For example, in the case shown in FIG. 6, the voice recognition process may be performed by the voice recognition unit 14 of the client 1 for only unknown words. In addition, the likelihood as a threshold value may be changed depending on whether the part of speech information is a common noun or a proper noun.

  FIG. 8 is a flowchart showing an example of the operation of the speech recognition system according to the second embodiment. The process of receiving the recognition result (step S23) from the voice input (step S21) of the client 1 is the same as the operation of FIG. In the example of FIG. 8, the server 3 transmits a recognition result including likelihood information to the client 1.

  The result determination unit 18 of the client 1 examines the speech recognition result of the server 3 (step S24), and if the likelihood is smaller than a threshold for each word (step S25; YES), the speech recognition unit 14 recognizes the speech for the phrase. (Step S26). If the likelihood is equal to or greater than the threshold (step S25; NO), the speech recognition process is not performed.

  If there is remaining voice data to be voice-recognized (step S27; YES), the result determination unit 18 checks the voice recognition result for the next word. If there is no remaining voice data (step S27; NO), the recognition result is displayed (step S28). The subsequent processing is the same as the operation of FIG.

  As described above, in the second embodiment, voice recognition is performed when the voice recognition result received from the server 3 meets a predetermined condition. As a result, the load on the client 1 for voice recognition can be reduced.

(Embodiment 3)
FIG. 9 is a block diagram showing the configuration of the speech recognition system according to the third embodiment. In the third embodiment, the client 1 further includes a feature amount extraction unit 19. It can be considered that the feature amount extraction included in the speech recognition unit 14 of FIG. 1 or 7 is shown separately.

  The feature amount extraction unit 19 extracts a feature amount for speech recognition from the input speech. The feature amount extraction unit 19 sends the feature amount data to the transmission unit 12 and the voice recognition unit 14. The transmission unit 12 transmits feature amount data to the server 3 as audio data. The voice recognition unit 33 of the server 3 performs voice recognition using the feature amount data received from the client 1.

  FIG. 10 is a flowchart illustrating an example of the operation of the speech recognition system according to the third embodiment. When voice is input through the voice input unit 11 of the client 1 (step S31), the feature amount extraction unit 19 extracts feature amount data (step S32). Then, the feature amount data is transmitted as audio data from the transmission unit 12 to the server 3 (step S33).

  In the server 3, when the feature amount data is received by the receiving unit 32 (step T31), the speech recognition unit 33 performs speech recognition processing using the feature amount data (step T32). Then, the speech recognition result is transmitted to the client 1 by the transmission unit 34 (step T33). The subsequent processing is the same as the operation of the second embodiment in FIG.

  In the third embodiment, since the characteristic parameter is transmitted as voice data, there is no deterioration in sound quality in the process of transmitting to the server 3, and high-accuracy voice recognition is realized. Since the characteristic parameter is smaller than the original audio signal data, the transmission time of data transmission / reception is short. Therefore, the voice recognition response is accelerated.

  FIG. 11 is a block diagram illustrating an example of a hardware configuration of the client 1 illustrated in FIG. As shown in FIG. 11, the client 1 includes a control unit 41, a main storage unit 42, an external storage unit 43, an operation unit 44, a display unit 45, an input / output unit 46, and a transmission / reception unit 47. The main storage unit 42, the external storage unit 43, the operation unit 44, the display unit 45, the input / output unit 46, and the transmission / reception unit 47 are all connected to the control unit 41 via the internal bus 40.

  The control unit 41 includes a CPU (Central Processing Unit) and the like, and in accordance with a control program 50 stored in the external storage unit 43, the voice input unit 11, the transmission unit 12, the reception unit 13, the voice recognition unit 14, Each process of the display part 15, the input part 16, the selection part 17, the result determination part 18, and the feature-value extraction part 19 is performed.

  The main storage unit 42 is composed of a RAM (Random-Access Memory) or the like, loads a control program 50 stored in the external storage unit 43, and is used as a work area of the control unit 41.

  The external storage unit 43 includes a nonvolatile memory such as a flash memory, a hard disk, a DVD-RAM (Digital Versatile Disc Random-Access Memory), a DVD-RW (Digital Versatile Disc ReWritable), and the control unit 41 controls the processing of the client 1. In accordance with an instruction from the control unit 41, the data stored in the program is supplied to the control unit 41, and the data supplied from the control unit 41 is stored.

  The operation unit 44 includes a pointing device such as a keyboard and a mouse, and an interface device that connects the keyboard and the pointing device to the internal bus 40. A voice recognition result selection instruction or the like is input via the operation unit 44 and supplied to the control unit 41.

  The display unit 45 is composed of a CRT (Cathode Ray Tube), an LCD (Liquid Crystal Display), or the like, and displays voice recognition results, input characters, words registered in the voice recognition dictionary, and the like.

  The input / output unit 46 includes a serial interface or a parallel interface. The input / output unit 46 connects the microphone 21 and inputs an audio signal. In addition, a speaker (not shown) is connected to reproduce sound.

  The transmission / reception unit 47 includes a network termination device or a wireless communication device connected to the network, and a serial interface or a LAN (Local Area Network) interface connected thereto. The transmission / reception unit 47 transmits voice data to the server 3 via the network and receives a voice recognition result.

  The voice input unit 11, the transmission unit 12, the reception unit 13, the voice recognition unit 14, the display unit 15, the input unit 16, the selection unit 17, the result determination unit 18, and the feature amount extraction unit of the client 1 illustrated in FIG. 19 is processed by the control program 50 using the control unit 41, main storage unit 42, external storage unit 43, operation unit 44, display unit 45, input / output unit 46, transmission / reception unit 47, and the like as resources. Execute.

  Note that the configuration of the voice recognition system described in each embodiment is an example, and can be arbitrarily changed and modified. The configuration of the client 1 or the server 3 is not limited to that shown in the embodiment, and is not limited to these. For example, as described in the first embodiment, the server 3 may include the voice recognition unit 14 for each client 1. The voice recognition unit 33 may be in the client 1.

  In addition, the above-described hardware configuration and flowchart are examples, and can be arbitrarily changed and modified.

  Speech recognition of the client 1 including the speech input unit 11, the transmission unit 12, the reception unit 13, the speech recognition unit 14, the display unit 15, the input unit 16, the selection unit 17, the result determination unit 18, the feature amount extraction unit 19, and the like The central part for processing can be realized by using a normal computer system, not by a dedicated system. For example, a computer program for executing the above operation is stored and distributed in a computer-readable recording medium (flexible disk, CD-ROM, DVD-ROM, etc.), and the computer program is installed in the computer. Thus, the client 1 that executes the above-described processing may be configured. The client 1 may be configured by storing the computer program in a storage device included in a server device on a communication network such as the Internet and downloading the computer program by a normal computer system.

  When the client 1 is realized by sharing of an OS (operating system) and an application program, or by cooperation between the OS and the application program, only the application program portion may be stored in a recording medium or a storage device. .

  It is also possible to superimpose a computer program on a carrier wave and distribute it via a communication network. For example, the computer program may be posted on a bulletin board (BBS, Bulletin Board System) on a communication network, and the computer program distributed via the network. The computer program may be started and executed in the same manner as other application programs under the control of the OS, so that the above-described processing may be executed.

It is a block diagram which shows the structure of the speech recognition system which concerns on Embodiment 1 of this invention. It is a figure which shows the example of the result of speech recognition. It is a figure which shows the example of a display of a speech recognition result, and recognition result selection. 4 is a flowchart illustrating an example of an operation of the voice recognition system according to the first embodiment. It is a figure which shows the example of recognition result selection in the case of using part of speech information. It is a figure which shows the example of recognition result selection in the case of using the information of an unknown word. It is a block diagram which shows the structure of the speech recognition system which concerns on Embodiment 2. FIG. 10 is a flowchart illustrating an example of an operation of the voice recognition system according to the second embodiment. FIG. 10 is a block diagram illustrating a configuration of a voice recognition system according to Embodiment 3. 10 is a flowchart illustrating an example of the operation of the speech recognition system according to the third embodiment. It is a block diagram which shows an example of the hardware constitutions of the client which concerns on embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Client 3 Server 10 Control part 11 Voice input part 12 Transmission part 13 Reception part 14 Voice recognition part 15 Display part 16 Input part 17 Selection part 18 Result determination part 19 Feature-value extraction part 20 Internal bus 21 Microphone 22 Display apparatus 23 Input apparatus DESCRIPTION OF SYMBOLS 30 Internal bus 31 Control part 32 Reception part 33 Voice recognition part 34 Transmission part 40 Internal bus 41 Control part 42 Main memory part 43 External memory part 44 Operation part 45 Display part 46 Input / output part 47 Transmission / reception part 50 Control program

Claims (16)

A speech recognition system composed of a server and a client,
The client
A voice input means for inputting a voice signal;
Transmitting means for transmitting data indicating the audio signal to a server;
With
The server
Server receiving means for receiving data indicating an audio signal from the client;
Server voice recognition means for voice recognition from the data received by the server reception means;
Server transmission means for transmitting the result of voice recognition by the server voice recognition means to the client;
With
The client further includes:
Receiving means for receiving the voice recognition result from the server;
Referring to the voice recognition result received from the server by the receiving means, voice recognition means for voice recognition for data indicating the transmitted voice signal corresponding to the received voice recognition result;
A selection means for selecting a recognition result from the voice recognition result received from the server and the result of voice recognition by the voice recognition means;
Comprising
A speech recognition system characterized by that. The server speech recognition means extracts phrase information including data representing a phrase break position and a phrase attribute from data indicating the voice signal,
The server transmission means transmits the phrase information to the client,
The voice recognition system according to claim 1, wherein the voice recognition unit of the client refers to the phrase information received from the server and recognizes the voice indicating data indicating the transmitted voice signal. The phrase information includes data representing a part of speech of a word included in the phrase,
The client voice recognition means refers to data representing the part of speech of the phrase information received from the server, and recognizes voice for the data indicating the transmitted voice signal.
The voice recognition system according to claim 2. The server speech recognition means estimates an attribute of an unknown word among results of speech recognition from data indicating the speech signal,
The server transmission means transmits the unknown word attribute estimated by the server speech recognition means to the client,
The voice recognition means of the client refers to the attribute of the unknown word received from the server, and recognizes voice about data indicating the transmitted voice signal.
The speech recognition system according to any one of claims 1 to 3, wherein   5. The speech recognition according to claim 1, wherein the speech recognition unit of the client performs speech recognition when a speech recognition result received from the server meets a predetermined condition. system. The client
When the speech recognition result received from the server and the result of speech recognition by the speech recognition means have a plurality of candidates for one word, display means for displaying at least two candidates among the plurality of candidates;
An input means for receiving an instruction to select at least one of the candidates displayed by the display means;
With
The selection means selects the recognition result based on the instruction to be selected input by the input means.
The speech recognition system according to claim 1, wherein: The client includes an extraction unit that extracts a feature parameter of speech input by the input unit,
The transmission means transmits the characteristic parameter extracted by the extraction means to the server,
The server speech recognition means recognizes speech using the feature parameter received from the client;
The speech recognition system according to claim 1, wherein A speech recognition method in a speech recognition system composed of a server and a client,
A voice input step of inputting a voice signal at the client;
A transmission step of transmitting data indicating the audio signal to a server;
A server receiving step of receiving data indicating an audio signal from the client at the server;
A server voice recognition step for voice recognition from the data received in the server reception step;
A server transmission step of transmitting the result of speech recognition in the server speech recognition step to the client;
A receiving step of receiving the voice recognition result from the server at the client;
With reference to the voice recognition result received in the reception step, the voice recognition step for voice recognition for the data indicating the transmitted voice signal;
A selection step for selecting a recognition result from the voice recognition result received from the server, and the voice recognition result in the voice recognition step;
A speech recognition method comprising: A voice input means for inputting a voice signal;
Transmitting means for transmitting data indicating the audio signal to a server;
Receiving means for receiving a result of voice recognition from the server;
With reference to the voice recognition result received by the receiving means, voice recognition means for voice recognition for the data indicating the transmitted voice signal;
A selection means for selecting a recognition result from the voice recognition result received from the server and the result of voice recognition by the voice recognition means;
A voice recognition client comprising: The receiving means receives clause information including data representing clause segmentation positions and clause attributes from the server,
The speech recognition means refers to the phrase information received by the reception means, and recognizes speech for data indicating the transmitted speech signal;
The voice recognition client according to claim 9. The phrase information includes data representing a part of speech of a word included in the phrase,
The speech recognition means refers to data representing a part of speech of the received phrase information, and recognizes speech for data indicating the transmitted speech signal;
The voice recognition client according to claim 10. The receiving means receives an attribute of an unknown word in the speech recognition result received from the server;
12. The voice recognition unit according to claim 9, wherein the voice recognition unit recognizes voice of data indicating the transmitted voice signal with reference to the attribute of the unknown word received by the reception unit. Voice recognition client as described in   The voice recognition client according to any one of claims 9 to 12, wherein the voice recognition means performs voice recognition when a voice recognition result received from the server meets a predetermined condition. When the speech recognition result received from the server and the result of speech recognition by the speech recognition means have a plurality of candidates for one word, display means for displaying at least two candidates among the plurality of candidates;
An input means for receiving an instruction to select at least one of the candidates displayed by the display means;
With
The selection means selects the recognition result based on the instruction to be selected input by the input means.
The voice recognition client according to any one of claims 9 to 13, wherein the voice recognition client is provided. Comprising extraction means for extracting feature parameters of speech input by the input means;
The transmission means transmits the characteristic parameters extracted by the extraction means to the server;
The voice recognition client according to claim 9, wherein the voice recognition client is a voice recognition client. Computer
A voice input means for inputting a voice signal;
Transmitting means for transmitting data indicating the audio signal to a server;
Receiving means for receiving a result of voice recognition from the server;
With reference to the voice recognition result received by the receiving means, voice recognition means for voice recognition for the data indicating the transmitted voice signal;
A program that functions as a selection unit that selects a recognition result from a voice recognition result received from the server and a result of voice recognition by the voice recognition unit.

Images