JP2010008764A - Speech recognition method, speech recognition system and speech recognition device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech recognition method capable of enhancing the precision of speech recognition. <P>SOLUTION: The speech of an operator and the speech of a customer are recorded in an operator speech record section 21 and a customer speech record section 22, respectively (S21). First, recognition of speech of the operator is performed for speech of the operator, which is recorded by an operator speech record section 21 (S22). After recognition of speech of the operator is performed, a priority of a word appeared as a result of the recognition is set as a maximum value(S23). Then, according to a coincidence measure calculated by a coincidence measure calculation section 26, the priority of the word is set (S24). Then, speech recognition of the customer is performed for the speech of the customer, which is recorded by the customer speech record section 22 (S25). <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a voice recognition method, a voice recognition system, and a voice recognition device, and in particular, a voice recognition method, a voice recognition system, and a voice that recognize a first voice and a second voice different from the first voice. The present invention relates to a recognition device.

  In a general call center, an operator and a customer have a conversation via a telephone line. The operator then provides services such as technical support and product explanation to the customer. In general, an operator performs a service while operating a personal computer (personal computer) or the like during a call with a customer.

  In such a call center, it is necessary to record the conversation content between the operator and the customer as the business content of the operator. In this case, the operator may record the conversation contents while recalling the customer's inquiry contents and answer contents one by one after the call ends. However, with such a method, it is difficult to perform accurate recording, and there is a risk of omissions and errors. Also, during a call with a customer, there are cases where operations are performed by a personal computer or the like as described above, so it is very difficult to record the contents of the conversation during the call.

  Here, techniques relating to voice recognition for recognizing and outputting voices of operators and customers are disclosed in, for example, Japanese Patent Laid-Open No. 11-338494 (Patent Document 1) and Japanese Patent No. 3827704 (Patent Document 2). . According to Patent Document 1, the content of a call is converted into text, that is, automatically converted into text by a voice recognition technique. By doing so, the efficiency of the conversation content recording work is improved.

Such a general speech recognition technique cannot accurately recognize all words in a conversation but requires confirmation and correction of the output speech recognition result. According to Patent Document 2, the text recognition result is displayed on a personal computer display or the like, and a keyword or the like designated in advance is highlighted. By doing so, the operator can grasp the correction part in a short time and perform correction or the like. In this way, the efficiency of recording conversation contents is improved.
JP 11-338494 A Japanese Patent No. 3827704

  The general speech recognition techniques shown in Patent Literature 1 and Patent Literature 2 described above select a plurality of word candidates assumed from a recorded speech waveform from a dictionary prepared in advance. And the word with the highest evaluation value among them, that is, the word that will be most appropriate is output as the recognition result.

  Here, the accuracy of voice recognition is preferably higher from the viewpoint of reducing the number of correction points. Since the waveform of the voice is different for each individual, in order to increase the accuracy of voice recognition, the correspondence between the waveform of the voice to be recognized and the word is appropriately set, that is, the voice to be recognized in advance. Can be considered.

  Here, since an operator can identify an individual, the operator's voice can be learned as voice to be recognized in advance. That is, in the voice recognition of the operator's voice, it is possible to improve the recognition accuracy. However, the voices of an unspecified number of customers as call destinations cannot be learned in advance in order to improve recognition accuracy. Then, in the speech recognition of the customer's speech, the general speech recognition shown in Patent Document 1 and Patent Document 2 described above, that is, speech recognition that does not perform learning in advance must be adopted, and the recognition accuracy is poor. It will be enough. Such a situation deteriorates the operator's work efficiency.

  An object of the present invention is to provide a speech recognition method capable of increasing the accuracy of speech recognition.

  Another object of the present invention is to provide a speech recognition system capable of increasing the accuracy of speech recognition.

  Still another object of the present invention is to provide a speech recognition apparatus capable of increasing the accuracy of speech recognition.

  Still another object of the present invention is to provide a speech recognition program capable of increasing the accuracy of speech recognition.

  Still another object of the present invention is to provide a recording medium on which a voice recognition program capable of increasing the accuracy of voice recognition is recorded.

  The speech recognition method according to the present invention is a speech recognition method for recognizing a first speech and a second speech different from the first speech, and obtains data for speech recognition of the first speech in advance. A data acquisition step, a learning step for performing learning for improving the recognition accuracy of the first voice using the data obtained in the data acquisition step, and a first step of recognizing the first voice after the learning step. One voice recognition step, a priority setting step for setting a higher priority for words that appear frequently in the first voice recognized in the first voice recognition step, and a word priority set in the priority setting step And a second voice recognition step for recognizing the second voice.

  By comprising in this way, a 2nd audio | voice can be recognized using the recognition result of a 1st audio | voice with high recognition accuracy. Therefore, the accuracy of the recognition result of the second voice can be improved, and the accuracy of the voice recognition can be improved as a whole.

  Preferably, the priority setting step includes a step of setting a word priority based on a co-occurrence scale between a word appearing in the first voice recognized by the first voice recognition step and a predetermined word.

  More preferably, the priority setting step sets the word priority to be higher as the co-occurrence scale is higher.

  More preferably, a co-occurrence scale calculating step for calculating in advance a co-occurrence scale between two predetermined words is provided before the first speech recognition step.

  In a further preferred embodiment, the first voice is an operator voice and the second voice is a customer voice.

  In another aspect of the present invention, a speech recognition system is a speech recognition system for recognizing a first speech and a second speech different from the first speech, respectively. A data obtaining means for obtaining data for learning, a learning means for performing learning for improving the recognition accuracy of the first speech using the data obtained by the data obtaining means, and after the learning means, First speech recognition means for recognizing speech, priority setting means for setting a higher priority for words that appear frequently in the first speech recognized by the first speech recognition means, and priority setting means. Second speech recognition means for recognizing the second speech based on the priority of the word.

  In still another aspect of the present invention, the speech recognition device is a speech recognition device that recognizes a first speech and a second speech different from the first speech, and recognizes the speech of the first speech in advance. A data acquisition unit for acquiring data for the first, a learning unit for performing learning for improving the recognition accuracy of the first speech using the data acquired by the data acquisition unit, Set by the first voice recognition unit that recognizes the voice, the priority setting unit that sets the priority of the words that frequently appear in the first voice recognized by the first voice recognition unit, and the priority setting unit And a second voice recognition unit for recognizing the second voice based on the priority of the word.

  In still another aspect of the present invention, the speech recognition program uses a data acquisition unit for acquiring data for speech recognition of the first speech in advance, and data acquired by the data acquisition unit. Learning means for performing learning for improving the recognition accuracy of the first voice, the first voice recognition means for recognizing the first voice, and the first voice recognized by the first voice recognition means after the learning means In order to function as a second voice recognition means for recognizing the second voice based on the priority setting means for setting the priority of a word with high frequency to be set high and the priority of the word set by the priority setting means Is a voice recognition program.

  In still another aspect of the present invention, the recording medium is a computer-readable recording medium on which the above-described voice recognition program is recorded.

  According to the present invention, the second voice can be recognized using the recognition result of the first voice with high recognition accuracy. Therefore, the accuracy of the recognition result of the second voice can be improved, and the accuracy of the voice recognition can be improved as a whole.

  In addition, the second speech is recognized using the recognition result of the first speech with high recognition accuracy even by such a speech recognition system, speech recognition device, speech recognition program, and recording medium on which the speech recognition program is recorded. can do. Therefore, the accuracy of the recognition result of the second voice can be improved, and the accuracy of the voice recognition can be improved as a whole.

  Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a system configuration diagram showing a hardware configuration of a speech recognition system according to an embodiment of the present invention. Referring to FIG. 1, a speech recognition system 11 includes a telephone 13 connected to a telephone line 12, a personal computer 14 connected to the telephone 13 and serving as an operator's operation terminal, a LAN (Local Area Network) cable, and the like. And a server 15 for storing data, programs and the like. The personal computer 14 includes a display 19 that displays text data and the like on its screen, a hard disk for storing data, and a keyboard and mouse (both not shown) that serve as an interface with the operator. The personal computer 14 is stored therein and is controlled by a control unit (not shown) that controls the operation of the entire personal computer 14.

  The telephone 13 and the personal computer 14 are connected by a line 16. The operator utters a voice using a microphone 18 to a customer connected to the telephone line 12 via the telephone 13. In addition, the voice from the customer via the telephone 13 is input through the speaker 17. The voice uttered by the operator as the first voice and the voice inputted from the customer as the second voice different from the first voice are inputted to the personal computer 14 as voice data. The input audio data is stored in a hard disk included in the personal computer 14. That is, the operator's voice and the customer's voice are recorded by the personal computer 14. In addition, audio data is also stored in the server 15 as necessary.

  Here, the configuration of the personal computer 14 and the like will be described. FIG. 2 is a block diagram of the speech recognition system 11 according to an embodiment of the present invention. Referring to FIGS. 1 and 2, a personal computer 14 provided in a speech recognition system 11 according to an embodiment of the present invention includes an operator speech recording unit 21 that records an operator's speech and a customer speech that records a customer's speech. Recording unit 22, voice recognition unit 23 for recognizing operator voice and customer voice respectively recorded by operator voice recording unit 21 and customer voice recording unit 22, and a result of voice recognized by voice recognition unit 23 are output. And a speech recognition result output unit 24.

  The personal computer 14 records the operator's voice as the first voice and the customer's voice as the second voice by the operator voice recording unit 21 and the customer voice recording unit 22, respectively. Then, the voice recognition unit 23 performs voice recognition separately.

  The voice recognition unit 23 performs voice recognition using the voice recognition dictionary unit 27. The speech recognition dictionary unit 27 stores a plurality of words corresponding to speech as a speech recognition dictionary. The voice recognition dictionary unit 27 is provided in the server 15 connected to the personal computer 14.

  The voice recognition unit 23 extracts words corresponding to the voices recorded by the operator voice recording unit 21 and the customer voice recording unit 22 from a plurality of words stored in the voice recognition dictionary in the voice recognition dictionary unit 27, and The result is output to the speech recognition result output unit 24. The speech recognition output is performed in a text format, that is, text data. Specifically, the text data as the output result is displayed on the display 19 of the personal computer 14 or the like.

  The personal computer 14 also includes a word priority setting unit 25 that sets the priority of words in the speech recognition dictionary when the speech recognition unit 23 recognizes speech. When the speech recognition unit 23 recognizes speech, the priority of words is used. If there is a similar word corresponding to the voice, it is reflected as a recognition result according to the priority, specifically, in descending order of priority.

  Here, the voice recognition unit 23 learns the operator's voice in advance. That is, among the word candidates assumed from the waveform of the operator's voice, the word having the highest evaluation value is recognized as an appropriate word in voice recognition, and corrected if it is incorrect. Such learning in the speech recognition unit 23 is possible because the operator is specified, and the accuracy of the recognition result of the operator's speech in the speech recognition unit 23 that has performed such learning is very high. .

  Specifically, voice data of an operator is input as a data acquisition step for acquiring data for voice recognition of the first voice in advance. Then, the above-described speech recognition learning is performed as a learning step for performing learning for improving the recognition accuracy of the first speech.

  The personal computer 14 also includes a co-occurrence scale calculation unit 26 that calculates a co-occurrence scale between two predetermined words from past business records. A server 15 is provided with a business record storage unit 28 that records past business records, that is, past conversations between an operator and a customer. The personal computer 14 sets the word priority setting unit 25 based on the co-occurrence scale calculated by the co-occurrence scale calculation unit 26.

  Here, a calculation method for calculating the co-occurrence scale by the co-occurrence scale calculation unit 26 will be described. FIG. 3 is a flowchart showing the operation of the control unit of the personal computer 14 when calculating the co-occurrence scale. Referring to FIGS. 1 to 3, first, morphological analysis is performed on past business records that have been converted to text (step S11 in FIG. 3, hereinafter, steps are omitted). Morphological analysis uses grammatical knowledge (a collection of grammar rules) and a dictionary (a word list with information such as parts of speech) as information sources, and divides sentences written in natural language into morpheme strings. Refers to the task of determining part of speech. Here, a morpheme is a minimum unit having meaning in a language. There are various tools for performing morphological analysis, including “ChaSen” which is free software. Here, any morphological analysis method may be used as long as it is a general method.

  Thereafter, a co-occurrence scale between two predetermined words is calculated for words appearing in all past business record texts (S12).

  Here, for the calculation of the co-occurrence scale, any of the following general scales may be used. For example, the number of occurrences of word X is | X |, the number of occurrences of word Y is | Y |, the number of business records in which at least one appears is | X ｜ Y |, and the number of business records in which both appear is | X If ∩Y |, the co-occurrence frequency is | X∩Y |, the Jaccard coefficient is | X∩Y | / | X∪Y |, and the Simpson coefficient is | X∩Y | / min (| X |, | Y |) and the cosine distance are represented by | X∩Y | / sqrt (| X || Y |).

  In this way, the co-occurrence scale is calculated. The co-occurrence scale calculated by the co-occurrence scale calculation unit 26 is used for setting of the word priority setting unit 25. That is, it is used when setting the priority of words for recognizing speech.

  Next, a method for recognizing speech using such a speech recognition system 11 will be described. FIG. 4 is a flowchart showing the operation of the control unit of the personal computer 14 when recognizing speech using the speech recognition system 11.

  1 to 4, first, the operator voice and the customer voice are separately recorded by the operator voice recording unit 21 and the customer voice recording unit 22 (S21 in FIG. 4).

  Next, the operator's voice recorded by the operator voice recording unit 21 is recognized as the first voice, as a first voice recognition step for recognizing the first voice (S22). In this case, since the operator's voice is learned in advance, the accuracy of the recognition result in the voice recognition unit 23 is high.

  After the recognition of the operator's voice is completed, the priority of words appearing from the recognition result is set to the maximum value (S23). Common words often appear in conversations between operators and customers. Therefore, by doing this, it is possible to increase the accuracy of customer's voice recognition performed later. This step is a priority setting step for setting a higher priority for words that frequently appear in the first speech recognized in the first speech recognition step.

  Next, priority is set for all words in the speech recognition dictionary according to the co-occurrence scale calculated by the co-occurrence scale calculation unit 26 (S24). This co-occurrence scale is calculated in the step shown in FIG. In this case, the higher the co-occurrence scale, the higher the priority. Specifically, a co-occurrence scale with all words appearing in the operator's speech recognition result is obtained for each word in the speech recognition dictionary, and an average value thereof is calculated. For a certain word X, assuming that the average value is Avr, the maximum value that can be set as the priority is Pmax, and the minimum value is Pmin, for example, (Pmax−Pmin) × Avr + Pmin is used as the priority of the word X.

  Thereafter, the customer's voice recorded by the customer voice recording unit 22 is recognized as the second voice (S25). This step is a second speech recognition step for recognizing the second speech based on the word priority set in the priority setting step. Next, the recognition results for the operator's voice and the customer's voice are output (S26).

  That is, the speech recognition method according to the present invention includes a data acquisition step for obtaining data for speech recognition of the first speech in advance, and a first speech recognition using the data obtained in the data acquisition step. A learning step for performing learning for improving accuracy, a first voice recognition step for recognizing the first voice after the learning step, and a high frequency of appearing in the first voice recognized by the first voice recognition step A priority setting step for setting a high word priority, and a second voice recognition step for recognizing the second voice based on the word priority set in the priority setting step. The speech recognition method according to the present invention further includes a co-occurrence scale calculating step for calculating in advance a co-occurrence scale between two predetermined words before the first speech recognition step.

  With this configuration, it is possible to recognize the customer's voice as the second voice by using the recognition result of the operator's voice as the first voice with high recognition accuracy. Therefore, the accuracy of the recognition result of the customer's voice can be improved, and the accuracy of the voice recognition of the operator's voice and the customer's voice can be improved as a whole.

  In this case, since speech recognition is performed by increasing the priority of words according to the co-occurrence scale, speech recognition can be performed more appropriately, that is, with higher accuracy.

  In the above embodiment, voice recognition is performed collectively for the operator's voice or the customer's voice. However, the present invention is not limited to this. For example, it is possible to divide by silence and perform speech recognition for each of the divided sounds.

  In the above embodiment, the operator's voice is recognized after the recording of both the operator's voice and the customer's voice, and then the customer's voice is recognized. However, the above-described voice recognition may be performed in real time or near real time while recording voice at any time during a call.

  In the above embodiment, the server stores the voice recognition dictionary unit and the business record storage unit. However, the present invention is not limited to this, and the voice recognition dictionary unit and the business record storage unit are stored in the hard disk of the personal computer. You may decide.

  In addition, the personal computer as the voice recognition device uses a data acquisition unit that obtains data for voice recognition of the first voice in advance, and the first voice recognition accuracy using the data obtained by the data acquisition unit. A learning unit that performs learning to improve performance, a first speech recognition unit that recognizes the first speech after the learning unit, and a word that frequently appears in the first speech recognized by the first speech recognition unit A priority setting unit that sets a higher priority of the second voice recognition unit, and a second voice recognition unit that recognizes the second voice based on the priority of the word set by the priority setting unit. .

  Further, a voice recognition system for other operators may be constructed on the network. In this case, the server may also be used.

  As the voice recognition program, the computer uses the data acquisition means for obtaining data for voice recognition of the first voice in advance, and the data obtained by the data acquisition means to improve the first voice recognition accuracy. Learning means for performing learning for improvement, first voice recognition means for recognizing the first voice after the learning means, and priority of words frequently appearing in the first voice recognized by the first voice recognition means Priority setting means for setting a higher value, and using a voice recognition program for functioning as second voice recognition means for recognizing the second voice based on the word priority set by the priority setting means Also good.

  Further, a computer-readable recording medium on which the voice recognition program described above is recorded may be used.

  Further, in this embodiment, the voice recognition of the operator voice and the customer voice in the call center has been described, but the present invention is not limited to this, such as telephone examination / health guidance, telephone consulting, telephone shopping, telephone questionnaire survey and public opinion survey, etc. It can be applied to various applications.

  As mentioned above, although embodiment of this invention was described with reference to drawings, this invention is not limited to the thing of embodiment shown in figure. Various modifications and variations can be made to the illustrated embodiment within the same range or equivalent range as the present invention.

  The voice recognition method, voice recognition system, voice recognition apparatus, voice recognition program, and recording medium according to the present invention are effectively used when it is necessary to recognize the voice of an operator and the voice of a customer such as a call center. .

1 is a system configuration diagram showing a hardware configuration of a speech recognition system according to an embodiment of the present invention. 1 is a block diagram of a voice recognition system according to an embodiment of the present invention. It is a flowchart which shows the operation | movement in the case of calculating a co-occurrence scale. It is a flowchart which shows the operation | movement at the time of recognizing an operator's voice and a customer's voice.

Explanation of symbols

  11 voice recognition system, 12 telephone line, 13 telephone, 14 personal computer, 15 server, 16 line, 17 speaker, 18 microphone, 19 display, 21 operator voice recording part, 22 customer voice recording part, 23 voice recognition part, 24 voice recognition Result output unit, 25 word priority setting unit, 26 co-occurrence scale calculation unit, 27 speech recognition dictionary unit, 28 business record storage unit.

Claims (9)

A voice recognition method for recognizing a first voice and a second voice different from the first voice,
A data acquisition step for acquiring data for speech recognition of the first voice in advance;
A learning step for performing learning for improving the recognition accuracy of the first voice, using the data obtained in the data obtaining step;
A first speech recognition step for recognizing the first speech after the learning step;
A priority setting step for setting a high priority for words that frequently appear in the first voice recognized in the first voice recognition step;
A speech recognition method comprising: a second speech recognition step for recognizing the second speech based on the priority of the word set in the priority setting step. The priority setting step includes a step of setting a word priority based on a co-occurrence scale between a word appearing in the first voice recognized by the first voice recognition step and a predetermined word. 2. The speech recognition method according to 1. The speech recognition method according to claim 2, wherein the priority setting step sets the word priority to be higher as the co-occurrence scale is higher. The speech recognition method according to claim 2 or 3, further comprising a co-occurrence scale calculation step for calculating in advance a co-occurrence scale between two predetermined words before the first speech recognition step. The first voice is an operator voice,
The voice recognition method according to claim 1, wherein the second voice is a customer voice. A voice recognition system for recognizing a first voice and a second voice different from the first voice,
Data acquisition means for acquiring data for speech recognition of the first voice in advance;
Learning means for performing learning for improving the recognition accuracy of the first voice, using the data obtained by the data obtaining means;
A first voice recognition means for recognizing the first voice after the learning means;
Priority setting means for setting a high priority for words that frequently appear in the first voice recognized by the first voice recognition means;
A voice recognition system comprising: second voice recognition means for recognizing the second voice based on the priority of the word set by the priority setting means. A voice recognition device for recognizing a first voice and a second voice different from the first voice,
A data acquisition unit for acquiring data for speech recognition of the first voice in advance;
A learning unit that performs learning for improving the recognition accuracy of the first voice, using the data obtained by the data obtaining unit;
A first voice recognition unit for recognizing the first voice after the learning unit;
A priority setting unit that sets a high priority for words that frequently appear in the first speech recognized by the first speech recognition unit;
A speech recognition apparatus comprising: a second speech recognition unit that recognizes the second speech based on the word priority set by the priority setting unit. Computer
Data acquisition means for acquiring data for voice recognition of the first voice in advance;
Learning means for performing learning for improving the recognition accuracy of the first voice, using the data obtained by the data obtaining means;
A first voice recognition means for recognizing the first voice after the learning means;
Priority setting means for setting a high priority for words that frequently appear in the first voice recognized by the first voice recognition means;
And a speech recognition program for causing the second speech recognition means to recognize the second speech based on the priority of the word set by the priority setting means. A computer-readable recording medium on which the voice recognition program according to claim 8 is recorded.

Images