JP2017134162A - Voice recognition device, voice recognition method, and voice recognition program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice recognition device capable of improving the accuracy in voice recognition without increasing processing load imposed on an external unit executing voice recognition.SOLUTION: A voice recognition device 10 includes: a storage section 13 that stores a piece of voice data received from a terminal which inputs a voice while associating a piece of identification information relevant to the terminal; an acquisition section 14 that acquires a piece of text data as voice recognition result of the voice data from an external unit; and a correction part 15 that corrects a character string included in the text data using a character string correction dictionary depending on the identification information.SELECTED DRAWING: Figure 3

Description

  The present invention relates to a voice recognition device, a voice recognition method, and a voice recognition program.

  2. Description of the Related Art Conventionally, there has been known a technique for converting a call voice or the like into text using a voice recognition engine installed in a voice recognition server for the purpose of quickly looking back on the call voice or the like at a terminal.

  In the speech recognition engine, the accuracy of speech recognition is reduced in the case of a speaker who makes utterances deviating from feature points. For example, a voice of “Occurrence of missed hearing” (pronounced by a person with a bad tongue) may be erroneously recognized as “occurrence of a functional foot”. For this reason, a technique for improving the accuracy of speech recognition by causing a speech recognition engine to learn the features of the speech of a speaker is known (see, for example, Patent Document 1).

JP 2010-175967 A

  However, in order for the voice recognition engine to learn the characteristics of the voice of a specific speaker, it may be necessary to input the voice of the speaker for several tens of hours. Further, when performing speech recognition based on the characteristics of a specific speaker's voice, there is a problem that the processing load on the voice recognition server increases.

  Therefore, an object of one aspect is to improve the accuracy of speech recognition without increasing the processing load of an external device that performs speech recognition.

  In one proposal, in the speech recognition apparatus, a result of speech recognition of the speech data from an external device and a recording unit that records speech data received from a terminal that inputs speech and identification information related to the terminal. An acquisition unit that acquires the text data, and a correction unit that corrects a character string included in the text data by using a character string correction dictionary according to the identification information.

  According to one aspect, the accuracy of speech recognition can be improved without increasing the processing load on an external device that performs speech recognition.

It is a figure which shows the structural example of the speech recognition system in embodiment. It is a figure which shows the hardware structural example of the speech recognition apparatus in embodiment. It is a functional block diagram of the speech recognition apparatus in the embodiment. It is a figure which shows an example of the data memorize | stored in a sound recording data storage part. It is a figure which shows an example of the data memorize | stored in a speech recognition result memory | storage part. It is a figure which shows an example of the data memorize | stored in a correction dictionary memory | storage part. It is a sequence diagram of the process which recognizes the recorded telephone call voice. It is a flowchart which shows an example of the correction process of a speech recognition result. It is a figure explaining the specific example of the correction process of a speech recognition result. It is a flowchart which shows an example of the registration process of a correction dictionary. It is a figure which shows the specific example of the registration process of a correction dictionary.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a diagram illustrating a configuration example of a voice recognition system according to an embodiment of the present invention. In FIG. 1, the speech recognition system 1 includes a speech recognition device 10, a telephone 20, and a speech recognition server 30 (an example of “external device”).

  The speech recognition apparatus 10 and the telephone 20 are connected so as to be communicable via a communication line such as a public telephone network, a mobile telephone network, or the Internet.

  The speech recognition apparatus 10 and the speech recognition server 30 are connected to be communicable via a communication line such as the Internet.

  The voice recognition device 10 is, for example, a smartphone, a tablet terminal, a mobile phone, a PC (Personal Computer), or the like.

  The voice recognition device 10 records voice such as a call received from the telephone 20 and causes the voice recognition server 30 to perform voice recognition of the recorded voice. The voice recognition device 10 corrects the text data of the voice recognition result based on a dictionary corresponding to the call partner. Note that the voice recognition device 10 may be a terminal similar to the telephone 20.

  The telephone 20 is, for example, a smartphone, a mobile phone, a fixed phone, an IP phone, a PC (Personal Computer), or the like. The telephone 20 transmits (inputs) voice to the voice recognition device 10 by voice, answering machine, voice message, or the like.

  The voice recognition server 30 recognizes voice data received from the voice recognition device 10 and transmits text data of the voice recognition result to the voice recognition device 10. Note that the voice recognition processing in the voice recognition server 30 may be performed using a known technique.

FIG. 2 is a diagram illustrating a hardware configuration example of the speech recognition apparatus 10 according to the embodiment. 2 includes a drive device 100, an auxiliary storage device 102, a memory device 103, a CPU 104, an interface device 105, a display device 106, an input device 107, and the like that are connected to each other via a bus B.
A voice recognition program for realizing processing in the voice recognition device 10 is provided by a recording medium 101 such as an SD memory card. When the recording medium 101 on which the voice recognition program is recorded is set in the drive device 100, the voice recognition program is installed from the recording medium 101 to the auxiliary storage device 102 via the drive device 100. However, the voice recognition program need not be installed from the recording medium 101 and may be downloaded from another computer via a network. The auxiliary storage device 102 stores the installed voice recognition program and also stores necessary files and data.
The memory device 103 reads the program from the auxiliary storage device 102 and stores it when there is an instruction to start the program. The CPU 104 realizes functions related to the speech recognition apparatus 10 according to a program stored in the memory device 103. The interface device 105 is used as an interface for connecting to a network. The display device 106 displays a GUI (Graphical User Interface) or the like by a program. The input device 107 includes a touch panel and buttons, or a keyboard and mouse, and is used to input various operation instructions.

  An example of the recording medium 101 is a portable recording medium such as an SD memory card, a CD-ROM, a DVD disk, or a USB memory. An example of the auxiliary storage device 102 is an HDD (Hard Disk Drive) or a flash memory. Both the recording medium 101 and the auxiliary storage device 102 correspond to computer-readable recording media.

  The hardware configuration of the speech recognition server 30 is a hardware configuration of a server computer, and may be the same as the hardware configuration example of the speech recognition apparatus 10 shown in FIG.

  Next, the functional configuration of the speech recognition apparatus 10 will be described with reference to FIG. FIG. 3 is a functional block diagram of the speech recognition apparatus 10. The voice recognition device 10 includes a communication unit 11, a recording unit 13, an acquisition unit 14, a correction unit 15, and a registration unit 16. Each of these units is realized by a process in which one or more programs installed in the speech recognition apparatus 10 are executed by the CPU 104 of the speech recognition apparatus 10.

  The voice recognition device 10 also includes a recording data storage unit 12A, a voice recognition result storage unit 12B, a correction dictionary storage unit 12C, and the like. Each of these storage units is realized using, for example, the auxiliary storage device 102 or the like.

  FIG. 4 is a diagram illustrating an example of data stored in the recording data storage unit 12A. The recorded data storage unit 12A stores the audio data received from the transmission source in association with the identification information of the transmission source of the audio data such as a call. The identification information of the voice data transmission source is information for identifying the telephone 20 or information for identifying the user of the telephone 20, for example, the telephone number of the telephone 20 that is the voice data transmission source, User account ID and the like.

  FIG. 5 is a diagram illustrating an example of data stored in the speech recognition result storage unit 12B. The speech recognition result storage unit 12B stores text data that is a speech recognition result for the speech data by the speech recognition server 30 in association with the identification information of the transmission source of the speech data. Further, when correction is necessary for the character string included in the text data, the text data corrected by the correction unit 15 is stored.

  FIG. 6 is a diagram illustrating an example of data stored in the correction dictionary storage unit 12C. The correction dictionary storage unit 12C stores a correction dictionary for each piece of identification information of the voice transmission source. The correction dictionary includes data items of “character string before correction”, “character string after correction”, and “noun within phrase”.

  The “character string before correction” is a character string that is a correction target in text data that is a voice recognition result by the voice recognition server 30. The “character string after correction” is a character string that corrects “character string before correction” by replacement. The “in-phrase noun” is a character string that is likely to be used together with the “character string before correction”.

  Returning to FIG. The communication unit 11 communicates with the telephone 20 and the voice recognition server 30.

  The recording unit 13 associates voice data such as a voice call, answering machine voice, and voice message received from the telephone 20 with identification information about the telephone 20 such as a telephone number and an account ID acquired at the time of outgoing / incoming call. And recorded in the recording data storage unit 12A.

  The acquisition unit 14 transmits the voice data stored in the recording data storage unit 12A to the voice recognition server 30, receives text data that is a result of voice recognition of the voice data from the voice recognition server 30, and performs voice recognition. Store in the result storage unit 12B. The voice recognition device 10 may be configured to recognize voice data stored in the recording data storage unit 12A.

  The correction unit 15 uses the correction dictionary corresponding to the identification information related to the telephone 20 that is the voice transmission source, and the character string included in the text data acquired by the acquisition unit 14 and stored in the voice recognition result storage unit 12B. Correct. The correction unit 15 corrects the character string by replacing the first character string included in the text data with the second character string registered in association with the first character string in the correction dictionary. To do.

  When the text data includes “character string before correction” and “noun in phrase” registered in the correction dictionary storage unit 12C, the correction unit 15 determines the character string of the “character string before correction”. Make corrections.

  The correction unit 15 stores the corrected text data by overwriting the uncorrected text data in the speech recognition result storage unit 12B.

  When the registration unit 16 receives an editing operation on any of the character strings included in the text data acquired by the acquisition unit 14 from the user, the registration unit 16 stores the correction dictionary before the editing in the correction dictionary corresponding to the identification information corresponding to the text data. Register the edited character string in association with the character string.

  More specifically, upon receiving the editing operation, the registration unit 16 accesses the correction dictionary associated with the identification information of the voice transmission source acquired from the outgoing / incoming call history and the telephone directory in the correction dictionary storage unit 12C. To do. And the registration part 16 extracts the noun etc. which are character strings different from the character string before edit from the clause of the said text data, matched with the character string before edit, and the said another character string, The edited character string is registered in the correction dictionary. Here, the registration unit 16 converts the character string before editing, another character string, and the edited character string into “character string before correction”, “noun within phrase”, and “character string after correction” in the correction dictionary. ”To each item.

  Next, processing for recognizing a recorded call voice will be described with reference to FIG. FIG. 7 is a sequence diagram of processing for recognizing a recorded call voice.

  In step S <b> 101, the voice recognition device 10 starts a call with the telephone 20.

  Subsequently, the telephone 20 transmits a call voice to the voice recognition device 10 (step S102).

  Subsequently, the voice recognition device 10 records the call voice from the telephone 20 (step S103).

  Subsequently, the voice recognition device 10 ends the call with the telephone 20 (step S104).

  Subsequently, the voice recognition device 10 transmits the voice data in which the call voice from the telephone 20 is recorded to the voice recognition server 30 (step S105).

  Subsequently, the voice recognition server 30 performs voice recognition on the received voice data, and generates text data as a voice recognition result (step S106).

  Subsequently, the voice recognition server 30 transmits the text data of the voice recognition result to the voice recognition device 10 (step S107).

  Subsequently, the voice recognition device 10 performs correction processing on the received text data of the voice recognition result based on the correction dictionary corresponding to the identification information of the voice transmission source stored in the correction dictionary storage unit 12C (step S108). ).

  Subsequently, the voice recognition device 10 displays the corrected text data on the display device 106 (step S109).

  Subsequently, the speech recognition apparatus 10 receives an edit (correction) operation on the text data from the user (step S110).

  Subsequently, the speech recognition device 10 registers “character string before correction”, “character string after correction”, and the like in association with the identification information of the transmission source of the call voice in the correction dictionary storage unit 12C (Step S1). S111).

  Next, a detailed example of the speech recognition result correction process will be described with reference to FIG. FIG. 8 is a flowchart illustrating an example of a speech recognition result correction process.

  In step S201, the acquisition unit 14 acquires text data of a voice recognition result.

  Subsequently, the correction unit 15 acquires the identification information of the voice transmission source from the outgoing / incoming call history and the telephone directory (step S202).

  Subsequently, the correction unit 15 refers to the correction dictionary storage unit 12C and acquires a correction dictionary corresponding to the identification information of the call voice transmission source (step S203).

  Subsequently, the correction unit 15 performs morphological analysis on the text data of the speech recognition result, and decomposes each clause included in the text data into a character string of a predetermined unit (step S204).

  Subsequently, the correcting unit 15 extracts a character string that is a noun from the decomposed character strings (step S205). The set of extracted character strings is hereinafter referred to as “character string set L”.

  Subsequently, the correcting unit 15 determines whether each character string of the extracted noun (each character string included in the character string set L) is registered in the “character string before correction” in the correction dictionary (step S206). .

  When each character string of the extracted noun is registered (YES in step S206), the correction unit 15 excludes each character string from the character string set L (step S207).

  Subsequently, the correction unit 15 determines whether each character string of the noun remaining without being excluded is a character string such as “temporal noun” stored in advance (step S208).

  When each character string to be determined is a character string stored in advance (YES in step S208), the correction unit 15 excludes each character string from the character string set L (step S209).

  Subsequently, the correcting unit 15 extracts the character string of the noun that remains without being excluded as “noun in phrase” (step S210).

  Subsequently, the correcting unit 15 determines that the combination of each character string of the noun registered in the correction dictionary excluded in step S207 and the character string of “noun in phrase” extracted in step S210 is “correction” in the correction dictionary. It is determined whether it is registered in “previous character string” and “noun in phrase” (step S211).

  If the set of the character strings is registered (YES in step S211), the correction unit 15 “character string before correction” excluded in step S207 from the character strings included in the text data of the speech recognition result. Is corrected (replaced) to “character string after correction” in the correction dictionary (step S212).

  Subsequently, the correction unit 15 stores the corrected text data in the voice recognition result storage unit 12B (step S213).

  Next, a specific example of the speech recognition result correction process will be described with reference to FIG. FIG. 9 is a diagram illustrating a specific example of the speech recognition result correction process.

  FIG. 9 illustrates an example in which the correction unit 15 acquires text data including a phrase “A function foot has occurred in the previous meeting” in step S201 of FIG.

  In this case, the correction unit 15 morphologically analyzes the phrase in step S204 of FIG. 8 and decomposes the phrase into character strings of a predetermined unit. In step S205 of FIG. 8, the correction unit 15 extracts nouns from the decomposed character string. In FIG. 9, for example, “the other day”, “function”, “foot”, and “occurrence” are extracted as nouns. In addition, the process which extracts a noun by performing a morphological analysis on a phrase may be performed using a known technique.

  In step S207 of FIG. 8, the correcting unit 15 includes “function” and “foot” because “function” and “foot” are continuous among the extracted nouns, and “functional foot” is the word before correction. "Is excluded. As a result, “the other day” and “occurrence” remain. In step S209 in FIG. 8, the correction unit 15 is a character string such as “temporal noun” in which “the other day” is stored in advance, and has a low probability of being used simultaneously with the character string to be corrected. "Is excluded. As a result, “occurrence” remains. As a result, “occurrence” is extracted as “noun in phrase” in step S210 of FIG.

  In step S212 of FIG. 8, the correcting unit 15 corrects the character string “functional foot” to the character string “missing”. In step S213 of FIG. 8, the correction unit 15 stores text data “Learning missed in the previous meeting” in the speech recognition result storage unit 12B.

  Next, a detailed example of registration processing in the correction dictionary storage unit 12C will be described with reference to FIG. FIG. 10 is a flowchart illustrating an example of a correction dictionary registration process.

  The registration unit 16 displays the text data stored in the voice recognition result storage unit 12B on the screen (step S301).

  Subsequently, the registration unit 16 receives an editing operation for the text data from the user (step S302).

  Subsequently, the registration unit 16 extracts “in-phrase nouns” from the phrases including the character string that has been edited in the text data (step S303). Note that the “noun in a phrase” is extracted by the same processing as Step S202 to Step S210 in FIG.

  Subsequently, the registration unit 16 associates with the identification information of the transmission source of the voice acquired from the outgoing / incoming call history or the phone book, the character string before editing, the character string after editing, and the “noun in phrase” extracted in step S303. Are registered in the items of “character string before correction”, “character string after correction”, and “noun within phrase” in the correction dictionary storage unit 12C (step S304).

  Next, a specific example of the correction dictionary registration process will be described with reference to FIG. FIG. 11 shows a specific example of correction dictionary registration processing.

  FIG. 11A is a display screen example when the text data stored in the speech recognition result storage unit 12B is displayed on the screen in step S301 of FIG. In the example of FIG. 11 (A), text data 501 “Hit was missed” is displayed. FIG. 11B to FIG. 11E are diagrams showing an operation and a display screen example when accepting an editing operation on text data in step S302 of FIG.

  For example, as illustrated in FIG. 11B, the registration unit 16 receives an operation of selecting a character string 502 that has been erroneously recognized. Subsequently, as illustrated in FIG. 11C, the registration unit 16 accepts a long press operation of the selected character string 502. Subsequently, as illustrated in FIG. 11D, the registration unit 16 displays a pop-up menu 503 for “registering in the correction dictionary” and accepts a pressing operation of the pop-up menu 503. Subsequently, as shown in FIG. 11D, the registration unit 16 displays a character string 502 before correction and a character string input field 505 after correction, and inputs the corrected character string into the input field 505. Accept the operation.

  After that, the registration unit 16 extracts the “character string before correction” and “character string after correction” input by the user in association with the identification information of the voice transmission source acquired from the outgoing / incoming call history and the phone book. The “noun in phrase” is registered in the correction dictionary storage unit 12C.

<Summary>
In the present embodiment, the voice recognition device 10 corrects a character string included in text data of a voice recognition result acquired from the voice recognition server 30 based on a correction dictionary corresponding to a voice transmission source.

  Thereby, the accuracy of voice recognition can be improved without increasing the processing load on the voice recognition server 30 side. Further, since it is not necessary to send the identification information of the voice transmission source such as a telephone number to an external device such as the voice recognition server 30, information security can be ensured.

  In addition, the correction dictionary includes “noun in phrase” which is a character string that is likely to be used together with “character string before correction”, and “text string before correction” and “ When “noun in phrase” is included, “character string before correction” is corrected to “character string after correction”. Thereby, erroneous correction processing can be suppressed.

  As mentioned above, although the Example of this invention was explained in full detail, this invention is not limited to such specific embodiment, In the range of the summary of this invention described in the claim, various deformation | transformation・ Change is possible.

Regarding the above description, the following items are further disclosed.
(Appendix 1)
A recording unit that records voice data received from a terminal that inputs voice and identification information about the terminal in association with each other;
An acquisition unit that acquires text data that is a result of voice recognition of the voice data from an external device;
A correction unit that corrects a character string included in the text data using a character string correction dictionary according to the identification information;
A speech recognition apparatus comprising:
(Appendix 2)
The correction unit corrects the first character string included in the text data to a second character string registered in association with the first character string in the correction dictionary. The speech recognition apparatus according to appendix 1.
(Appendix 3)
When an editing operation on any character string included in the text data acquired by the acquisition unit is accepted, editing is performed in association with the character string before editing in the correction dictionary corresponding to the identification information corresponding to the text data. The speech recognition apparatus according to appendix 1 or 2, further comprising a registration unit that registers a subsequent character string.
(Appendix 4)
When the registration unit accepts an editing operation on any of the character strings included in the text data acquired by the acquisition unit, the registration unit extracts a character string different from the character string before the editing from the text data, In association with the character string before editing and the other character string, the edited character string is registered in the correction dictionary corresponding to the identification information corresponding to the text data,
The correction unit corrects the character string before editing when the character string before editing and the other character string are included in the text data acquired from the external device. 4. The speech recognition device according to 3.
(Appendix 5)
Voice recognition device
A process of associating and recording voice data received from a terminal that inputs voice and identification information about the terminal;
Processing to obtain text data as a result of voice recognition of the voice data from an external device;
A process of correcting a character string included in the text data using a character string correction dictionary according to the identification information;
The voice recognition method characterized by performing.
(Appendix 6)
In the correcting process, the first character string included in the text data is corrected to a second character string registered in association with the first character string in the correction dictionary. The speech recognition method according to appendix 5.
(Appendix 7)
When an editing operation on any character string included in the text data acquired by the acquisition process is received, the correction dictionary corresponding to the identification information corresponding to the text data is associated with the character string before editing. The speech recognition method according to appendix 5 or 6, wherein a process of registering the edited character string is executed.
(Appendix 8)
When the registering process accepts an editing operation on any of the character strings included in the text data acquired by the acquiring process, the character string different from the character string before the editing is extracted from the text data. Then, in association with the character string before editing and the other character string, the edited character string is registered in a correction dictionary corresponding to the identification information corresponding to the text data,
In the correction process, when the text data acquired from the external device includes the character string before editing and the other character string, the correction is performed on the character string before editing. The speech recognition method according to appendix 7.
(Appendix 9)
In voice recognition device,
A process of associating and recording voice data received from a terminal that inputs voice and identification information about the terminal;
Processing to obtain text data as a result of voice recognition of the voice data from an external device;
A process of correcting a character string included in the text data using a character string correction dictionary according to the identification information;
A speech recognition program characterized in that
(Appendix 10)
In the correcting process, the first character string included in the text data is corrected to a second character string registered in association with the first character string in the correction dictionary. The voice recognition program according to appendix 9.
(Appendix 11)
When an editing operation on any character string included in the text data acquired by the acquisition process is received, the correction dictionary corresponding to the identification information corresponding to the text data is associated with the character string before editing. The speech recognition program according to appendix 9 or 10, wherein a process for registering the edited character string is executed.
(Appendix 12)
When the registering process accepts an editing operation on any of the character strings included in the text data acquired by the acquiring process, the character string different from the character string before the editing is extracted from the text data. Then, in association with the character string before editing and the other character string, the edited character string is registered in a correction dictionary corresponding to the identification information corresponding to the text data,
In the correction process, when the text data acquired from the external device includes the character string before editing and the other character string, the correction is performed on the character string before editing. The speech recognition program according to attachment 11.

DESCRIPTION OF SYMBOLS 10 Speech recognition apparatus 11 Communication part 12 Storage part 12A Recording data 12B Speech recognition result data 12C Correction dictionary data 13 Recording part 14 Acquisition part 15 Correction part 16 Registration part

Claims (6)

A recording unit that records voice data received from a terminal that inputs voice and identification information about the terminal in association with each other;
An acquisition unit that acquires text data that is a result of voice recognition of the voice data from an external device;
A correction unit that corrects a character string included in the text data using a character string correction dictionary according to the identification information;
A speech recognition apparatus comprising:   The correction unit corrects the first character string included in the text data to a second character string registered in association with the first character string in the correction dictionary. The speech recognition apparatus according to claim 1.   When an editing operation on any character string included in the text data acquired by the acquisition unit is accepted, editing is performed in association with the character string before editing in the correction dictionary corresponding to the identification information corresponding to the text data. The speech recognition apparatus according to claim 1, further comprising a registration unit that registers subsequent character strings. When the registration unit accepts an editing operation on any of the character strings included in the text data acquired by the acquisition unit, the registration unit extracts a character string different from the character string before the editing from the text data, In association with the character string before editing and the other character string, the edited character string is registered in the correction dictionary corresponding to the identification information corresponding to the text data,
The correction unit corrects the character string before editing when the text data acquired from the external device includes the character string before editing and the other character string. Item 4. The speech recognition device according to Item 3. Voice recognition device
A process of associating and recording voice data received from a terminal that inputs voice and identification information about the terminal;
Processing to obtain text data as a result of voice recognition of the voice data from an external device;
A process of correcting a character string included in the text data using a character string correction dictionary according to the identification information;
The voice recognition method characterized by performing. In voice recognition device,
A process of associating and recording voice data received from a terminal that inputs voice and identification information about the terminal;
Processing to obtain text data as a result of voice recognition of the voice data from an external device;
A process of correcting a character string included in the text data using a character string correction dictionary according to the identification information;
A speech recognition program characterized in that

Images