JP2010044241A - Voice recognition device and control program of same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To lessen processing load and reduce the response time in voice recognition processing. <P>SOLUTION: This voice recognition device includes: a recognition processing request receiving means 10 to which a voice recognition request and a voice signal are input; a similarity measuring means 31 for measuring the similarity between a voice signal input together with a voice recognition request and a voice signal stored in a database in the past; and a voice recognition processing means 20 for performing the voice recognition processing for the input voice signal using an acoustic model 21 and a language model 22. The recognition processing request receiving means 10 measures the similarity to the voice signal of a relevant user stored in the database in the past before the recognition processing is performed, and reads the voice recognition result to the relevant voice signal from the database 33 to be output as the voice recognition result to the voice signal input together with the voice recognition request when there is a voice signal having high similarity. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a speech recognition technology that uses past speech recognition results.

Conventionally, when speech input is performed, a speech recognition device that measures the similarity with speech input in the past and outputs a recognition result corresponding to the speech when the similarity is greater than a certain threshold Is known (Patent Document 1). In this speech recognition apparatus, among standard patterns of recognition target words, a partial standard pattern to be subjected to similarity calculation with an input pattern obtained by the speech analysis unit analyzing a speech waveform, and a cumulative similarity of these patterns are stored. To do. Further, the cumulative similarity is compared with the current threshold value, and the number of partial standard patterns having a high similarity is counted. A pruning threshold value is calculated according to the correspondence relationship between the current threshold value and the number, and based on the pruning threshold value, a similarity with a feature amount is calculated for a partial standard pattern having a high similarity. Then, a word corresponding to the standard pattern having the highest similarity among the partial standard patterns is determined as a recognition result.
JP-A-10-153999

  However, the technique described in Patent Document 1 has a problem in that the number of pieces of voice data to be compared is increased and the processing time is increased because the degree of similarity with all voices input in the past is calculated. In addition, for the comparison target voice, the similarity is calculated for the voices of different users without taking into account the user information, so that the processing time becomes longer and an incorrect recognition result may be returned. Furthermore, since the similarity is calculated as the comparison target voice regardless of whether the recognition result is correct or incorrect, the processing time may be increased and an erroneous recognition result may be output. In addition, since it is also compared with the voice that the user uttered many times without noticing it, the recognition result that the user does not want may be output many times.

  The present invention has been made in view of such circumstances, and it is an object of the present invention to provide a speech recognition apparatus and a program that can reduce processing time required for similarity measurement and avoid errors in recognition results. To do.

  (1) In order to achieve the above object, the present invention takes the following measures. That is, the speech recognition apparatus of the present invention recognizes the speech recognition result having a high similarity when the speech signal input together with the speech recognition request and the speech signal accumulated in the past are high. A speech recognition apparatus for outputting a speech recognition result for a speech signal input together with a request, a recognition processing request receiving means for inputting a speech recognition request and a speech signal, a speech signal input together with the speech recognition request, and a database A similarity measurement means for measuring the similarity with the voice signal already stored in the voice, and a voice recognition processing means for performing voice recognition processing of the input voice signal using an acoustic model and a language model, The recognition processing request receiving means, when the speech signal having a high similarity exists in the speech signal already accumulated in the database as a result of the measurement, A speech recognition result for the speech signal having a high similarity is read from the database and output as a speech recognition result for the speech signal input together with the speech recognition request, while the similarity is added to the speech signal already stored in the database. When there is no voice signal having a high level, the voice signal input together with the voice recognition request is output to the voice recognition processing means to execute voice recognition processing.

  As described above, when a speech signal having a high similarity exists in the speech signal stored in the database in the past, a speech recognition result for the speech signal having a high similarity is read from the database and input together with the speech recognition request. Since it is output as a speech recognition result for the speech signal, it is not necessary to perform speech recognition processing with a large processing load, and the processing load can be reduced and the response time can be shortened.

  (2) In the speech recognition apparatus according to the present invention, the recognition process request receiving unit inputs a speech recognition request, a speech signal, and a user ID, and the similarity measuring unit receives the user ID input together with the speech recognition request. And a voice signal is acquired, and a similarity between the voice data input together with the voice recognition request and a voice signal specified by the user ID among voice signals already stored in a database is measured. It is said.

  As described above, the user ID and the voice signal input together with the voice recognition request are acquired, and the voice data input together with the voice recognition request and the voice specified by the user ID among the voice signals already stored in the database. Since the similarity with the signal is measured, the similarity can be measured only for the audio signal specified by the user ID, and it becomes possible to shorten the processing time and avoid the recognition result error.

  (3) In the speech recognition apparatus of the present invention, the similarity measurer is specified by the user ID among the speech data input together with the speech recognition request and the speech signal already stored in the database. And a speech signal having a recognition score included in the recognition result of the speech recognition processing means higher than a predetermined threshold is measured.

  As described above, among the voice data input together with the voice recognition request and the voice signals already stored in the database, the voice signal specified by the user ID and recognized in the recognition result of the voice recognition processing means. Since the degree of similarity with an audio signal whose score is higher than a predetermined threshold is measured, it is possible to avoid outputting past erroneous recognition results and to reduce the processing time required for similarity measurement. .

  (4) Further, in the voice recognition device of the present invention, the similarity measurer includes a plurality of units corresponding to the user IDs within a predetermined period before the time when the voice signal input together with the voice recognition request is input. If these are similar to each other, they are excluded from the measurement target of the similarity, and the similarity among the audio signals corresponding to the user ID and the measurement target of the similarity is excluded. A speech recognition result for the highest speech signal having a degree equal to or higher than a threshold is read from the database and output as a speech recognition result for the speech signal input together with the speech recognition request, while the similarity is measured corresponding to the user ID If a voice signal having a similarity greater than or equal to a threshold value has not been input within the predetermined period, the voice recognition request is shared. Is characterized by executing the speech recognition processing the input audio signal is output to the speech recognition processing unit.

  As described above, when a plurality of audio signals corresponding to the user ID are input within a predetermined period before the time when the audio signal input together with the audio recognition request is input, and these are similar to each other, Since the similarity is excluded from the measurement target, it is possible to reduce the processing load and the processing time, and to avoid outputting an erroneous recognition result many times. In addition, a voice recognition result for a voice signal whose similarity is equal to or higher than a threshold among voice signals corresponding to the user ID is read from the database, and a voice for the voice signal input together with the voice recognition request is read. Since it outputs as a recognition result, it becomes possible to raise the probability of outputting a correct speech recognition result.

  (5) In the speech recognition apparatus of the present invention, the recognition processing request receiving means includes a plurality of correspondences corresponding to the user IDs at least a predetermined period before the time when the speech signal input together with the speech recognition request is input. If these are similar to the voice signal input together with the voice recognition request, the voice recognition result for the voice signal having the highest similarity with the similarity equal to or higher than a threshold is read from the database, and While outputting as a voice recognition result for a voice signal input together with a voice recognition request, if a voice signal whose similarity is greater than or equal to a threshold is not input more than a predetermined period before the time, along with the voice recognition request The input voice signal is output to the voice recognition processing means to execute voice recognition processing.

  As described above, a plurality of audio signals corresponding to the user ID are input a predetermined period or more before the time when the audio signal input together with the audio recognition request is input, and these are the audio signal input together with the audio recognition request If they are similar, the speech recognition result for the speech signal having the highest similarity that is equal to or higher than the threshold value is read from the database and output as the speech recognition result for the speech signal input together with the speech recognition request. Can be increased.

  (6) Moreover, in the speech recognition apparatus of the present invention, the recognition processing request receiving means includes a plurality of correspondences corresponding to the user ID at least a predetermined period before the time when the speech signal input together with the speech recognition request is input. If these are similar to the speech signal input together with the speech recognition request, the speech recognition result for the latest speech signal with the similarity equal to or greater than a threshold value is read from the database, and the speech recognition request When a voice signal whose similarity is equal to or higher than a threshold is not input before a predetermined period before the time, the voice recognition request is input together with the voice recognition request. A voice signal is output to the voice recognition processing means to execute voice recognition processing.

  As described above, a plurality of audio signals corresponding to the user ID are input a predetermined period or more before the time when the audio signal input together with the audio recognition request is input, and these are the audio signal input together with the audio recognition request If they are similar, the speech recognition result for the latest speech signal with a similarity equal to or greater than the threshold value is read from the database and output as the speech recognition result for the speech signal input together with the speech recognition request, so the correct speech recognition result is output. The probability can be increased.

  (7) Further, the control program of the speech recognition apparatus according to the present invention provides a past speech having a high similarity when the similarity between the speech signal input together with the speech recognition request and the speech signal accumulated in the past is high. A control program for a speech recognition apparatus that outputs a recognition result as a speech recognition result for a speech signal input together with the speech recognition request, the processing for inputting the speech recognition request and the speech signal, and input together with the speech recognition request Processing for measuring the similarity between an audio signal and an audio signal already stored in the database, and when the result of the measurement is an audio signal having a high similarity in the audio signal already stored in the database Reads out a speech recognition result for a speech signal having a high degree of similarity from a database, and for a speech signal input together with the speech recognition request If there is no voice signal having a high similarity in the process of outputting as a voice recognition result and the voice signal already stored in the database, an acoustic model and A series of processes including a voice recognition process using a language model is converted into a command that can be read and executed by a computer.

  As described above, when a speech signal having a high similarity exists in the speech signal stored in the database in the past, a speech recognition result for the speech signal having a high similarity is read from the database and input together with the speech recognition request. Since it is output as a speech recognition result for the speech signal, it is not necessary to perform speech recognition processing with a large processing load, and the processing load can be reduced and the response time can be shortened.

  According to the present invention, since the user ID is used for the accumulated voice that measures the similarity to the input voice, it is possible to shorten the processing time required for the similarity measurement and to avoid outputting an erroneous recognition result. Can do. Also, since the accumulated speech for measuring the similarity is only targeted for those whose recognition score is equal to or greater than a certain threshold value, it is possible to avoid outputting past erroneous recognition results, and processing required for similarity measurement Time can be shortened. Furthermore, since the access information is used and voices with a similar similarity among a plurality of voices uttered in the near past are determined as vocabulary utterances or misrecognition utterances, and excluded from the target of similarity, processing load Can be reduced and the processing time can be shortened, and it is possible to avoid outputting erroneous recognition results many times.

  In the speech recognition apparatus according to the present embodiment, when speech similar to accumulated speech generated in the past is input by the user, past speech recognition results are returned without performing speech recognition processing. As a result, the speech recognition result can be output with a short response time, and the processing load can be reduced. The voice recognition processing has a large processing load and requires processing time, whereas the similarity determination processing has a processing load smaller than that of the voice recognition processing and processing time is short. Accordingly, when a voice similar to the accumulated voice generated in the past is input, the voice recognition process is not performed and the past voice recognition result is returned, thereby reducing the processing load and the response time. It becomes possible. Hereinafter, the present embodiment will be described with reference to the drawings.

  FIG. 1 is a diagram illustrating a schematic configuration of a speech recognition apparatus according to the present embodiment. The recognition process request receiving means 10 receives a voice recognition request and voice data from the user terminal, and instructs the voice recognition processing means 20 to perform a recognition process. In addition to returning the recognition result to the user terminal, the input voice, the recognition result, and the access information are stored in each DB. When receiving the voice recognition request, the process control unit 13 transmits the input voice to the similarity measurement unit 31 and instructs the similarity measurement unit 31 to measure the similarity with the past accumulated voice. If there is speech similar to the past, the speech recognition result for the speech is acquired from the past history DB 33 and output. If there is no speech similar to the past, the speech recognition processing means 20 is caused to execute speech recognition processing.

  The speech recognition processing means 20 performs recognition processing using the acoustic model 21 and the language model 22 in accordance with the recognition request received from the recognition processing request receiving means 10. Then, the recognition result is returned to the recognition process request receiving means 10.

  The similarity measurer 31 determines the similarity between the input voice and accumulated voice data uttered in the past. When the obtained distance between the audio data is equal to or less than the threshold, it is determined that the two are similar. The voice for measuring the similarity is limited to that of the corresponding user ID. The past history DB 33 includes an accumulated voice DB 33a, a recognition result DB 33b, and an access information DB 33c, as will be described later. Details will be described later.

  FIG. 2 is a diagram showing a schematic configuration of the past history DB 33 shown in FIG. The stored voice DB 33a is a DB that stores input voice. The voice data to be stored may be spectrum data, cepstrum domain data, VQ data, etc. in addition to voice data in the PCM format or the like. The recognition result DB 33b is a DB that accumulates recognition results. The recognition results that are accumulated are recognized characters and recognition scores. The recognition score may be further held separately for acoustic likelihood and language probability. The access information DB 33c is a DB that accumulates access information. The access information to be accumulated includes access time, access user ID, voice data name stored in the corresponding stored voice DB, and recognition result file name stored in the corresponding recognition result DB.

  FIG. 3 is a block diagram showing a schematic configuration of the similarity measuring means 31 shown in FIG. In the similarity measurer 31, the similarity determination controller 31a receives the input voice data and the user ID from the recognition process request receiver 10, and then acquires the voice data of the same user ID in the accumulated voice DB 33a. The data processing unit 31c processes the input voice and the voice data acquired from the stored voice DB 33a into the same type of data format. For example, if both are audio data such as PCM or spectral domain data, they are processed into spectral domain data, cepstrum domain data, VQ data, and the like. If both are cepstrum area data, they are processed into cepstrum area data, VQ data, and the like. When both are VQ data, they are left as they are.

  The distance calculation means 31d calculates the distance between the voice data or the distance of the recognition result. The obtained distance is output to the inter-speech distance information table, the recognition result inter-distance information table, and the similarity determination means 31e. The similarity determination unit 31e determines that they are similar when the distance obtained by the distance calculation unit 31d is equal to or less than a threshold value. Information on the determined result and the corresponding data is output to the recognition process request receiving means 10.

The similarity measurer 31 shown in FIG. 1 can take other configurations. FIG. 4 is a diagram illustrating another configuration of the similarity measuring unit 31. In FIG. 4, the access information analyzing unit 31b receives the input voice data and the user ID from the processing control unit 13 in the recognition processing request receiving unit 10, and then acquires the access information of the corresponding user ID from the access information DB 33c. Audio data used for determining the degree is selected. The access information analyzing unit 31b selects one satisfying one of the following two conditions as voice data for measuring the similarity. In FIG. 4, the other components are the same as those in FIG.
(Condition 1) If a plurality of voices uttered within T from the current time are similar to each other, they are excluded from the similarity measurement target. Then, only the audio signal having the highest similarity (the distance is small) is selected from the audio signals corresponding to the user ID and whose similarity is to be measured.
(Condition 2) Audio data having an interval of T ′ or more from the current time is selected.

  Further, the similarity measuring unit 31 shown in FIG. 1 can adopt other configurations. FIG. 5 is a diagram showing another configuration of the similarity measuring unit 31. In FIG. 5, the access information / recognition result analyzing unit 31f receives the input voice data and the user ID from the processing control unit 13 in the recognition processing request receiving unit 10, and then acquires the access information of the corresponding user ID from the access information DB 33c. In addition, only the recognition result having a high recognition score is acquired from the recognition result DB, and the audio data used for determining the similarity is selected. In FIG. 5, the other components are the same as those in FIGS.

FIG. 6 is a diagram showing the concept of the distance calculation method performed by the distance calculation means 31d shown in FIGS. In this distance calculation, the distance between two voices having different numbers of frames is obtained using DTW (Dynamic Time Stretching Method). Examples of distances between frames include the following distance measures. (1) Logarithmic spectrum, LPC spectrum, cepstrum, Euclidean distance of VQ data.
(2) Maximum likelihood spectral distance using LPC spectrum.
(3) Cosh scale.

  FIG. 7 is a diagram showing a concept of operation of the similarity determination unit 31e shown in FIGS. When the access time is a plurality of utterances within T from the current time, the similarity determination unit 31e excludes these from the similarity measurement target if they are similar to each other. Then, among the audio signals corresponding to the user ID and whose similarity is to be measured, only the signal with the highest similarity (small distance) is set as the determination target. In addition, a voice whose access time is at an interval equal to or greater than T ′ is determined.

  FIG. 8 is a flowchart showing the operation of the speech recognition apparatus according to this embodiment. First, a voice recognition request is received (step S1), and the input voice and user ID are transmitted to the similarity measurer 31 (step S2). Next, the voice of the user ID whose similarity is to be measured is acquired from the accumulated voice DB 33a (step S3), and the data formats of the input voice and the past voice are unified (step S4). Then, the distance between the input voice and the past voice is calculated (step S5).

  Next, similarity determination is performed based on the distance between voices (step S6). If they are not similar, that is, if the distance between voices is greater than or equal to a threshold, voice recognition processing is performed using the input voice ( In step S7), the recognition result is returned (step S8). On the other hand, if they are similar in step S6, that is, if the distance between the voices is less than or equal to the threshold, the voice recognition result of the past voice similar to the input voice is acquired from the recognition result DB 33b (step S9). The acquired past speech recognition result is returned (step S10).

  The characteristic operations of the present embodiment as described above can be performed by causing a computer to execute a program. In other words, the control program for the speech recognition apparatus according to the present embodiment is configured so that when the similarity between the speech signal input together with the speech recognition request and the speech signal accumulated in the past is high, the past speech recognition with the high similarity is performed. A control program for a speech recognition apparatus for outputting a result as a speech recognition result for a speech signal input together with the speech recognition request, a process for inputting the speech recognition request and the speech signal, and speech input together with the speech recognition request A process of measuring the similarity between the signal and the voice signal already stored in the database, and if the result of the measurement is that the voice signal already stored in the database contains a voice signal having a high similarity The speech recognition result for the speech signal having a high similarity is read from the database, and the speech signal input together with the speech recognition request is read. If the speech signal having the high similarity does not exist in the speech signal already stored in the database and the processing to output as the speech recognition result, an acoustic model is generated for the speech signal input together with the speech recognition request. In addition, a series of processes including a voice recognition process using a language model is converted into a command that can be read and executed by a computer.

  As described above, when a speech signal having a high similarity exists in the speech signal stored in the database in the past, a speech recognition result for the speech signal having a high similarity is read from the database and input together with the speech recognition request. Since it is output as a speech recognition result for the speech signal, it is not necessary to perform speech recognition processing with a large processing load, and the processing load can be reduced and the response time can be shortened.

It is a figure which shows schematic structure of the speech recognition apparatus which concerns on this embodiment. It is a figure which shows schematic structure of the past log | history DB33 shown in FIG. It is a block diagram which shows schematic structure of the similarity measuring means 31 shown in FIG. FIG. 6 is a diagram showing another configuration of the similarity measurer 31. FIG. 6 is a diagram showing another configuration of the similarity measurer 31. It is a figure which shows the concept of the distance calculation method which the distance calculation means 31d shown in FIG. 3 and FIG. 4 performs. It is a figure which shows the concept of operation | movement of the similarity determination means 31e shown in FIG. 3 and FIG. It is a flowchart which shows operation | movement of the speech recognition apparatus which concerns on this embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Recognition process request | requirement reception means 13 Process control means 20 Speech recognition processing means 21 Acoustic model 22 Language model 31 Similarity measurement means 31a Similarity determination control means 31b Access information analysis means 31c Data processing means 31d Distance calculation means 31e Similarity determination means 31f Access information / recognition result analysis means 33 Past history DB
33a Accumulated voice DB
33b Recognition result DB
33c Access information DB

Claims (7)

When the similarity between the speech signal input together with the speech recognition request and the speech signal accumulated in the past is high, the speech recognition for the speech signal input together with the speech recognition request is performed based on the past speech recognition result having the high similarity. A speech recognition device that outputs as a result,
A recognition processing request receiving means for inputting a voice recognition request and a voice signal;
Similarity measuring means for measuring the similarity between the voice signal input together with the voice recognition request and the voice signal already stored in the database;
Voice recognition processing means for performing voice recognition processing of an input voice signal using an acoustic model and a language model,
The recognition processing request receiving means, when there is a voice signal having a high similarity in the voice signal already stored in the database as a result of the measurement, a voice recognition result for the voice signal having a high similarity is stored in the database. Is output as a speech recognition result for the speech signal input together with the speech recognition request, while the speech recognition having a high similarity does not exist in the speech signal already stored in the database. A speech recognition apparatus characterized in that a speech signal input together with a request is output to the speech recognition processing means to execute speech recognition processing. The recognition processing request receiving means inputs a voice recognition request, a voice signal and a user ID,
The similarity measuring unit acquires a user ID and a voice signal input together with the voice recognition request, and the user ID among the voice data input together with the voice recognition request and the voice signal already stored in the database. The speech recognition apparatus according to claim 1, wherein the degree of similarity with the speech signal specified in (1) is measured.   The similarity measuring means is a voice signal specified by the user ID among voice data input together with the voice recognition request and voice signals already stored in a database, and the voice recognition processing means The speech recognition apparatus according to claim 2, wherein the degree of similarity with a speech signal having a recognition score included in the recognition result is higher than a predetermined threshold is measured.   The similarity measurer receives a plurality of voice signals corresponding to the user ID within a predetermined period before the time when the voice signal input together with the voice recognition request is input, and these are similar to each other. In this case, these are excluded from the similarity measurement target, and among the speech signals corresponding to the user ID and the similarity measurement target, the speech recognition result for the highest speech signal having the similarity equal to or higher than a threshold value Is read out from the database and output as a speech recognition result for the speech signal input together with the speech recognition request, while the similarity is equal to or greater than a threshold among speech signals corresponding to the user ID and the similarity is to be measured. If the voice signal to be input is not input within the predetermined period, the voice signal input together with the voice recognition request is converted into the voice recognition processing means. Output speech recognition apparatus according to claim 2 or claim 3, wherein the executing the speech recognition process.   The recognition processing request receiving means receives a plurality of audio signals corresponding to the user ID at least a predetermined period before the time when the audio signal input together with the audio recognition request is input, and these are input to the audio recognition request If the voice signal is similar to the voice signal input together with the voice recognition request, the voice recognition result for the voice signal having the highest similarity with the similarity equal to or higher than a threshold value is read from the database, and the voice recognition result for the voice signal input with the voice recognition request is read. On the other hand, if the audio signal having the similarity equal to or higher than the threshold value has not been input before the predetermined time, the audio signal input together with the audio recognition request is output to the audio recognition processing means. The voice recognition apparatus according to claim 2, wherein voice recognition processing is executed.   The recognition processing request receiving means receives a plurality of audio signals corresponding to the user ID at least a predetermined period before the time when the audio signal input together with the audio recognition request is input, and these are input to the audio recognition request When the voice signal is similar to the voice signal input together with the voice recognition request, the voice recognition result for the latest voice signal with the similarity equal to or higher than a threshold value is read from the database and output as the voice recognition result for the voice signal input together with the voice recognition request. On the other hand, if the audio signal whose similarity is equal to or greater than the threshold value has not been input before a predetermined period of time before the time, the audio signal input together with the audio recognition request is output to the audio recognition processing means. The speech recognition apparatus according to claim 2 or 3, wherein recognition processing is executed. When the similarity between the speech signal input together with the speech recognition request and the speech signal accumulated in the past is high, the speech recognition for the speech signal input together with the speech recognition request is performed based on the past speech recognition result having the high similarity. A speech recognition device control program that outputs as a result,
A process of inputting a voice recognition request and a voice signal;
A process of measuring the similarity between the voice signal input together with the voice recognition request and the voice signal already stored in the database;
As a result of the measurement, if a speech signal having a high similarity exists in the speech signal already stored in the database, a speech recognition result for the speech signal having a high similarity is read from the database, and the speech recognition request Processing to output as a voice recognition result for the voice signal input together with,
If the speech signal already stored in the database does not contain a speech signal with a high degree of similarity, speech recognition processing is performed using an acoustic model and a language model for the speech signal input together with the speech recognition request. A control program for a speech recognition apparatus, characterized in that a series of processes including: