JP6450139B2 - Speech recognition apparatus, speech recognition method, and speech recognition program - Google Patents

Description

  The present invention relates to a speech recognition device that recognizes speech from an arbitrary direction, a speech recognition method, and a speech recognition program.

  In recent years, techniques for device operation, information acquisition, dialogue, and the like by voice recognition are becoming widespread. In particular, when a device such as a robot performs voice recognition and executes a process based on a voice recognition result, it is required that the device can accurately recognize a voice coming from an arbitrary direction. For this purpose, a device that estimates a sound source direction and sets the directivity direction of a microphone array in the sound source direction is known.

  For example, in Patent Document 1, in addition to a spatial filter having a blind spot directed to a sound source direction, a spatial filter having directivity directed to a sound source direction is also generated, a pattern of direction and gain is obtained for each, and based on both patterns It is described that a sound source direction is estimated. Patent Document 2 describes that the MUSIC method is used as a method of estimating the sound source direction. Japanese Patent Application Laid-Open No. 2004-228561 describes that the sound of a speaker hitting the palm of the hand is detected as a cue sound and the directivity direction of the microphone array is set.

JP 2012-150237 A JP 2010-121975 A International Publication No. 2011/055410

  However, in the methods described in Patent Documents 1 and 2, there is a possibility that the directivity direction is set for a sound source such as ambient noise different from the sound uttered by the device user. In addition, it is necessary to observe an audio signal for a certain period in order to estimate the sound source direction. Further, according to the method described in Patent Document 3, it is possible to prevent the directivity direction from being set with respect to ambient noise or the like. It is necessary to perform actions such as tapping. That is, since speech recognition is started after the sound source direction is estimated, it takes time until a speech recognition result is obtained, and the user's experience value is impaired.

  The present invention has been made in view of the above problems, and a voice recognition device, a voice recognition method, and the like that can accurately recognize a voice from an arbitrary direction without a delay associated with estimation of a sound source direction. And a speech recognition program.

  The speech recognition apparatus according to the present invention performs speech recognition on each of the sound acquisition means for acquiring each of the sound streams from a plurality of directions and each of the sound streams from the plurality of directions acquired by the sound acquisition means. When a speech recognition result satisfying a predetermined reliability criterion is obtained by the speech recognition processing means and the speech recognition processing means, the direction corresponding to the speech stream from which the speech recognition result is obtained is set as the sound source direction. Sound source direction determining means for determining.

  The speech recognition method according to the present invention is a speech recognition method executed by a speech recognition apparatus, and includes a speech acquisition step of acquiring each of speech streams from a plurality of directions, and a plurality of directions acquired in the speech acquisition step. A speech recognition processing step for performing speech recognition on each of the speech streams from and a speech recognition result that satisfies a predetermined reliability criterion in the speech recognition processing step. And a sound source direction determining step for determining a direction corresponding to the audio stream from which the sound stream is obtained as a sound source direction.

  The speech recognition program according to the present invention enables a computer to perform speech recognition on each of an audio acquisition unit that acquires audio streams from a plurality of directions and each of the audio streams from a plurality of directions acquired by the audio acquisition unit. When a speech recognition result that satisfies a predetermined reliability criterion is obtained by the speech recognition processing means that executes the above and the speech recognition processing means, the direction corresponding to the speech stream from which the speech recognition result is obtained is determined. This is executed as sound source direction determining means for determining the sound source direction.

  In the voice recognition apparatus according to the present invention, the voice recognition processing means obtains the voice recognition results for each of the voice streams from a plurality of directions acquired by the voice acquisition means. In addition, when a speech recognition result satisfying a predetermined reliability criterion is obtained by the speech recognition processing means, the direction corresponding to the sound stream from which the speech recognition result is obtained is indicated by the sound source direction determining means. Determined as direction. As described above, according to the voice recognition device, it is possible to determine the sound source direction while continuously performing the voice recognition, instead of starting the voice recognition after determining the sound source direction. That is, it is possible to execute speech recognition for speech from an arbitrary direction without a delay associated with estimation of the sound source direction. In addition, after the sound source direction is determined, for example, it is possible to perform higher-accuracy speech recognition on an audio stream from the determined sound source direction, so that the accuracy of speech recognition can be improved. It becomes. Therefore, according to the speech recognition apparatus, it is possible to recognize speech from any direction with high accuracy without the delay associated with estimation of the sound source direction.

  The voice recognition apparatus performs voice recognition with higher accuracy than the voice recognition by the voice recognition processing unit on the voice stream from the sound source direction determined by the sound source direction determination unit among the voice streams acquired by the voice acquisition unit. You may further provide the 2nd audio | voice recognition process means to perform.

  According to the speech recognition apparatus, when the sound source direction is determined by the sound source direction determination unit, the second speech recognition processing unit performs more accurate speech recognition on the audio stream from the sound source direction. Can do.

  In the voice recognition apparatus, the voice acquisition unit may acquire a voice stream corresponding to the beam direction of each directional beam by setting the directional beam in a plurality of predetermined directions.

  According to the voice recognition apparatus, it is possible to accurately acquire voice streams from each of a plurality of predetermined directions (fixed directions).

  In the voice recognition apparatus, the voice acquisition unit sets a directional beam in a plurality of directions that are candidates for the sound source direction estimated by a predetermined method, thereby generating a voice stream corresponding to the beam direction of each directional beam. You may get it.

  According to the speech recognition apparatus, for example, by setting directional beams in a plurality of directions that are candidates for the sound source direction estimated by the MUSIC method or the like, a sound stream from a direction that is highly likely to be a sound source direction is given priority. And the accuracy of speech recognition can be improved.

  In the speech recognition apparatus, the speech recognition processing means may determine that the speech recognition result satisfies a predetermined reliability criterion when a predetermined word is included in the speech recognition result. .

  According to the voice recognition device, it is possible to easily and accurately determine the reliability of the voice recognition result based on whether or not a predetermined word is voice-recognized.

  In the above speech recognition apparatus, the speech recognition processing means may execute an utterance section detection process for detecting an utterance section, and perform speech recognition on the utterance section detected by the utterance section detection process.

  According to the voice recognition device, it is possible to perform voice recognition only for the speech section detected by the speech section detection process. As a result, it is possible to prevent unnecessary speech recognition processing from being performed on sections other than the speech section in the voice stream, and to reduce power consumption.

  According to the present invention, it is possible to accurately recognize a voice from an arbitrary direction without a delay accompanying estimation of a sound source direction.

It is a block diagram which shows the function structure of the speech recognition apparatus which concerns on embodiment of this invention. It is a block diagram which shows the hardware constitutions of a speech recognition apparatus. It is a figure which shows an example of the beam direction set with a several microphone. It is a flowchart which shows operation | movement of a speech recognition apparatus. It is a block diagram which shows the module structure of a speech recognition program.

  Hereinafter, a voice recognition device, a voice recognition method, and a voice recognition program according to embodiments of the present invention will be described with reference to the drawings. Where possible, the same parts are denoted by the same reference numerals, and redundant description is omitted. FIG. 1 is a block diagram showing a functional configuration of the speech recognition apparatus 1 according to the present embodiment. As shown in FIG. 1, the speech recognition apparatus 1 includes a speech input unit 11, a directivity control unit 12, a first speech recognition processing unit 13, a sound source direction determination unit 14, and a second speech recognition processing unit 15. A speech recognition result output unit 16.

  The speech recognition device 1 is configured as a device that recognizes a user's uttered speech and executes processing according to the speech recognition result. For example, the voice recognition device 1 is configured as a device that is installed in a living room in a home, recognizes a user's uttered voice, and instructs a home appliance or the like to execute processing according to the voice recognition result by radio waves or the like. Alternatively, it may be incorporated in the device itself that executes processing according to the voice recognition result. The voice recognition device 1 may be configured as a voice interaction device (for example, a robot or the like) that presents a response result to a question from a user to the user by text, voice, or the like.

  FIG. 2 is a block diagram illustrating an example of a hardware configuration of the speech recognition apparatus 1. As shown in FIG. 2, the speech recognition apparatus 1 includes, for example, a CPU (Central Processing Unit) 10A, a RAM (Random Access Memory) 10B, a ROM (Read Only Memory) 10C, and an input device 10D as hardware configurations. A communication device 10E such as a radio wave module for performing communication with an external device, an auxiliary storage device 10F, and an output device 10G. The input device 10D includes a plurality of microphones corresponding to the audio input unit 11, and includes a keyboard and a mouse that are input devices, for example. The output device 10G is, for example, a display that outputs the response result as text, a speaker that outputs the response result as sound, or the like. Each function of the speech recognition apparatus 1 is realized, for example, by causing the RAM 10B or the like to read a speech recognition program P described later and causing the CPU 10A to execute the speech recognition program P.

  Note that the speech recognition apparatus 1 does not necessarily have to include all of the hardware configurations described above. For example, the speech recognition device 1 may not include the output device 10G when it does not have a function of outputting the response result as text and speech. The speech recognition apparatus 1 may be configured as a physically single device, or may be configured such that a plurality of physically separated devices operate in cooperation.

  The voice input unit 11 is a voice input unit that collects sounds around the voice recognition device 1 and acquires the sounds as signals of a plurality of channels (a plurality of frequency bands). The voice input unit 11 is constituted by a plurality of microphones, for example.

  The directivity control unit 12 is an audio acquisition unit that acquires audio streams from a plurality of directions. The directivity control unit 12 executes signal processing for emphasizing only speech that has arrived from a preset beam direction and suppressing speech that has arrived from another direction by using a known method such as a fixed beamformer. . More specifically, the directivity control unit 12 performs the above-described signal processing on a plurality of channel signals obtained from the audio input unit 11, so that each of a plurality of beam directions is determined from each beam direction. Only an incoming voice is emphasized, and an audio stream in which an incoming voice from another beam direction is suppressed is generated.

  The directivity control unit 12 may acquire an audio stream corresponding to the beam direction of each directional beam by setting directional beams in a plurality of predetermined directions. That is, the plurality of beam directions set by the directivity control unit 12 may be fixed beam directions set in advance. Thereby, the audio | voice stream from each of the predetermined several direction (fixed direction) can be acquired accurately. In particular, when it is not known in advance from which direction the voice comes, the beam direction may be set so that a plurality of beams cover all directions as shown in FIG. In the example of FIG. 3, the beam directions are set in eight directions centered on the speech recognition apparatus 1 and divided at 45 degree intervals in the horizontal direction. By setting the beam direction in this way, the directivity control unit 12 emphasizes, for example, the voice coming from the sound source X ahead of the beam direction a set from the voice recognition device 1 toward the upper right in FIG. Audio streams can be generated.

  In addition, when the candidates of the sound transmission source (sound source direction) input to the speech recognition apparatus 1 are limited to a certain range, the directivity control unit 12 has a plurality of beams only in the certain range. The beam direction may be set so as to cover the film. For example, consider a case where the speech recognition apparatus 1 is incorporated in a television receiver and the user instructs the television receiver to perform a predetermined operation (for example, channel change) by speaking. In this case, it is assumed that the position of the user is in front of the screen of the television receiver. That is, the sound source direction candidates are limited to a range of 180 degrees in front of the screen of the television receiver. Therefore, in this case, the directivity control unit 12 may set the beam direction so as to cover only the range of 180 degrees ahead of the screen of the television receiver. The number of beam directions to be set depends on the number of microphones and the sharpness of directivity set by signal processing, but more beam directions can be set than the number of normal microphones.

  Further, the directivity control unit 12 obtains an audio stream corresponding to the beam direction of each directional beam by setting directional beams in a plurality of directions that are candidates for the sound source direction estimated by a predetermined method. May be. That is, the directivity control unit 12 may set a plurality of sound source direction candidates estimated by, for example, the MUSIC method using an acoustic signal input to the voice input unit 11 as a plurality of beam directions. Alternatively, the directivity control unit 12 sets a plurality of sound source direction candidates estimated by sound source direction tracking by a technique such as a Kalman filter or a particle filter as a plurality of beam directions in addition to the sound source direction estimation by the MUSIC method or the like. Also good. When a plurality of beam directions are set in this way, the plurality of beam directions are set and changed depending on the acoustic signal, unlike the case of using the above-described fixed beamformer. Thus, by setting directional beams in a plurality of directions that are candidates for the sound source direction estimated by the MUSIC method or the like, it is possible to preferentially obtain an audio stream from a direction that is likely to be a sound source direction, The accuracy of voice recognition can be improved.

  The first voice recognition processing unit 13 is a voice recognition processing unit that performs voice recognition on each of voice streams from a plurality of directions acquired by the directivity control unit 12. Hereinafter, each of the audio streams from a plurality of directions acquired by the directivity control unit 12 is also referred to as a directional audio stream. A plurality of directional audio streams are continuously acquired by the processing of the audio input unit 11 and the directivity control unit 12 described above, and input to the first audio recognition processing unit 13. For this reason, the 1st audio | voice recognition process part 13 will perform an audio | voice recognition continuously about each of several directional audio | voice stream.

  In some cases, the directional audio stream contains only noise, not human voice. In addition, the directional audio stream may include only a section that is almost silent. Therefore, in order to prevent the first speech recognition processing unit 13 from misrecognizing the noise included in the directional speech stream as a human voice and obtaining an erroneous speech recognition result, Then, a known noise removal process may be executed before executing voice recognition. In addition, the first speech recognition processing unit 13 performs a well-known speech segment detection process on the directional speech stream in order to detect a speech segment (a segment including a human voice) in which speech recognition is to be performed. The speech recognition may be executed on the speech section detected by the speech section detection process. As a result, it is possible to prevent unnecessary speech recognition processing from being performed on sections other than the speech section in the directional sound stream, and power consumption can be reduced.

  The first speech recognition processing unit 13 is obtained by performing speech recognition on each directional speech stream (or each directional speech stream after performing the above-described noise removal processing, speech segment detection processing, and the like). It is determined whether or not the predetermined voice recognition result satisfies a predetermined reliability criterion. As the reliability used for such determination, for example, the reliability based on the likelihood of the output hypothesis, which is a well-known index in the field of statistical speech recognition, can be used.

  When one or more command words to be uttered (speech input) first to make the speech recognition apparatus 1 recognize speech of the user's speech and execute predetermined processing are determined in advance, the first speech recognition When the predetermined command word is included in the voice recognition result, the processing unit 13 may determine that the voice recognition result satisfies a predetermined reliability criterion. In this case, the first speech recognition processing unit 13 may use only the command word as a speech recognition target vocabulary. As a result, the first speech recognition processing unit 13 recognizes the directional speech stream and, when a speech recognition result (that is, a result indicating that any command word is recognized) is obtained, It can be determined that the recognition result satisfies a predetermined reliability criterion. That is, the determination of the reliability of the speech recognition result can be performed easily and accurately based on whether or not the command word is recognized as speech.

  When the first speech recognition processing unit 13 obtains a speech recognition result satisfying a predetermined reliability criterion from a certain directional speech stream, the first speech recognition processing unit 13 arrives at the direction of the directional speech stream (that is, the directional speech stream). It is estimated that human speech is coming from (the beam direction corresponding to), and notifies the sound source direction determination unit 14 of the beam direction. The first speech recognition processing unit 13 outputs a speech recognition result obtained by speech recognition for each directional stream to the speech recognition result output unit 16.

  The sound source direction determination unit 14 determines the direction corresponding to the audio stream from which the speech recognition result is obtained when the first speech recognition processing unit 13 obtains a speech recognition result that satisfies a predetermined reliability criterion. Sound source direction determining means for determining the sound source direction.

  For example, when the first sound recognition processing unit 13 is notified of the beam direction estimated to be the arrival direction of the human speech as described above, the sound source direction determination unit 14 determines the beam direction as the sound source direction. To do. Then, the sound source direction determination unit 14 acquires a directional audio stream corresponding to the beam direction from the directivity control unit 12 and outputs the directional audio stream to the second audio recognition processing unit 15. Alternatively, the sound source direction determination unit 14 is notified from the first speech recognition processing unit 13 by acquiring signals of a plurality of channels obtained from the voice input unit 11 and performing unique signal processing on the signals. A directional audio stream corresponding to the beam direction may be acquired.

  In the initial state in which no beam direction has been notified from the first speech recognition processing unit 13 in the past, the sound source direction determination unit 14 sets the initial beam direction in advance to the second speech recognition processing unit 15. A corresponding directional audio stream may be output, or no directional audio stream may be output.

  The sound source direction determination unit 14 is initialized when a predetermined time elapses after the beam direction is notified from the first speech recognition processing unit 13 in the past without receiving the beam direction notification. You may return to the state. The sound source direction determination unit 14 is notified of one beam direction from the first speech recognition processing unit 13 in the past, and further notified of another beam direction from the first speech recognition processing unit 13 at a later time. Alternatively, another beam direction notified later may be determined as the latest sound source direction, and a directional sound stream corresponding to the sound source direction may be output to the second sound recognition processing unit 15.

  The second voice recognition processing unit 15 performs second voice recognition processing on the voice stream from the sound source direction determined by the sound source direction determination unit 14 among the voice streams acquired by the directivity control unit 12. Means. The speech recognition processing by the second speech recognition processing unit 15 performs speech recognition with high accuracy using more words as speech recognition targets than the speech recognition by the first speech recognition processing unit 13. As a result, when the sound source direction is determined by the sound source direction determination unit 14, the second sound recognition processing unit 15 can execute more accurate sound recognition on the sound stream from the sound source direction. Become. For example, the voice recognition by the first voice recognition processing unit 13 is a local type voice recognition for executing the recognition process in the voice recognition device 1, and the voice recognition by the second voice recognition processing unit 15 is a server using an external server. Type speech recognition. Note that the speech recognition processing by the second speech recognition processing unit 15 may be the same as the speech recognition processing by the first speech recognition processing unit 13.

  When executing the above-described server-type speech recognition, the second speech recognition processing unit 15 sends the directional speech stream acquired from the sound source direction determining unit 14 to a server (not shown) having a function of performing speech recognition. The voice recognition may be executed by transmitting the data, causing the server to execute voice recognition, and receiving the voice recognition result (for example, text or the like; the same applies hereinafter) from the server. As described above, the second speech recognition processing unit 15 causes a server or the like equipped with a high-performance speech recognition engine to perform speech recognition, and thereby the sound stream from the sound source direction determined by the sound source direction determining unit 14 is processed. More accurate voice recognition can be performed. Note that data transmission / reception between the second speech recognition processing unit 15 and the server is performed via the Internet, a LAN (Local Area Network), or the like by using the communication function of the communication device 10E described above, for example. .

  The second speech recognition processing unit 15 performs speech recognition on the directional speech stream acquired from the sound source direction determining unit 14 as described above, and the speech recognition result obtained as a result is the speech recognition result output unit. 16 is output.

  The speech recognition result output unit 16 is a speech recognition result output unit that outputs a speech recognition result acquired from at least one of the first speech recognition processing unit 13 and the second speech recognition processing unit 15. The voice recognition result output unit 16 may output only the voice recognition result acquired from one of the first voice recognition processing unit 13 and the second voice recognition processing unit 15, or the first voice recognition processing unit 13 and the second voice recognition processing unit 15. You may output the text which combined the voice recognition result acquired from both of the voice recognition processing parts 15. As a specific method of outputting the speech recognition result, the speech recognition result output unit 16 may present the speech recognition result (text) to the user by outputting it to the output device 10G such as a display, The voice corresponding to the voice recognition result may be synthesized by a known method, and the obtained voice may be output to the output device 10G such as a speaker.

  Further, the voice recognition result output unit 16 may present information indicating some response result based on the voice recognition result to the user, in addition to outputting the voice recognition result by text or voice. For example, when the speech recognition device 1 is configured as a device that performs information retrieval based on the content of utterances from the user, the speech recognition result output unit 16 presents the search result to the user in text, speech, or the like. Also good. Similarly, when the voice recognition device 1 is configured as a device that executes a predetermined device operation (for example, an operation to turn off the light through a remote controller) based on the utterance content from the user, a voice recognition result output unit 16 may present the result of the device operation (for example, information indicating that the lamp has been turned off) to the user by text or voice. Further, when the voice recognition device 1 is configured as a device that realizes a dialog with the user by answering a user's question such as chat, the voice recognition result output unit 16 answers the user's question. The message may be presented to the user by text or voice.

  Next, an example of processing (voice recognition method) executed by the voice recognition device 1 will be described with reference to the flowchart shown in FIG. First, external audio is continuously input to the audio input unit 11 including a plurality of microphones. Then, the directivity control unit 12 performs signal processing on the signals of a plurality of channels obtained from the audio input unit 11, and corresponds to each of the audio streams from a plurality of directions (that is, to each of a plurality of beam directions). Directional audio stream) is acquired (step S1, audio acquisition step).

  Subsequently, voice recognition is performed on each directional voice stream by the first voice recognition processing unit 13 (step S2, voice recognition processing step). Then, the first speech recognition processing unit 13 determines whether or not the speech recognition result for each directional speech stream satisfies a predetermined reliability criterion (step S3).

  If it is determined in step S3 that a speech recognition result satisfying a predetermined reliability criterion has been obtained (step S3: YES), the first speech recognition processing unit 13 obtains the speech recognition result. The sound source direction determination unit 14 is notified of the beam direction corresponding to the directional audio stream. Then, the direction corresponding to the audio stream from which the speech recognition result is obtained is determined as the sound source direction by the sound source direction determination unit 14 (step S4, sound source direction determination step), and the directional audio stream corresponding to the sound source direction is determined. It is output to the second voice recognition processing unit 15.

  On the other hand, if it is not determined in step S3 that a speech recognition result satisfying a predetermined reliability criterion has been obtained (step S3: NO), the first speech recognition processing unit 13 to the sound source direction determining unit 14 Is not notified of the beam direction (the beam direction estimated as the sound source direction), and the sound source direction determination unit 14 determines the initial beam direction set in advance (or from the first speech recognition processing unit 13 within a certain past period). A directional audio stream corresponding to the notified beam direction is output to the second audio recognition processing unit 15.

  Subsequently, voice recognition is performed on the voice stream from the set sound source direction by the second voice recognition processing unit 15 (step S5, second voice recognition processing step). Here, the “set sound source direction” means that if it is determined that a speech recognition result satisfying the reliability criterion determined in advance in step S3 is obtained (step S3: YES), in step S4. If the sound source direction determined by the sound source direction determining unit 14 is not determined that a speech recognition result satisfying the reliability criterion determined in step S3 has been obtained (step S3: NO), the sound source direction is set in advance. The initial beam direction (or the beam direction notified from the first speech recognition processing unit 13 within a predetermined period in the past).

  Subsequently, the speech recognition result output unit 16 outputs at least one of the speech recognition result for each directional speech stream by the first speech recognition processing unit 13 and the speech recognition result by the second speech recognition processing unit 15 (step). S6). Here, the voice recognition result may be output in a form in which the voice recognition result is presented to the user as it is as text or voice, or information indicating some response result based on the voice recognition result (for example, search result, device, etc. As a result of the operation, an answer message to the user's inquiry may be output in a form presented to the user in text or voice.

  In the voice recognition device 1 described above, the first voice recognition processing unit 13 obtains the voice recognition result for each of the voice streams from the plurality of directions acquired by the directivity control unit 12. In addition, when the first speech recognition processing unit 13 obtains a speech recognition result that satisfies a predetermined reliability criterion, the sound source direction determination unit 14 corresponds to the speech stream from which the speech recognition result is obtained. Direction is determined as the sound source direction. As described above, according to the voice recognition device 1, it is possible to determine the sound source direction while continuously performing the voice recognition, instead of starting the voice recognition after determining the sound source direction. That is, it is possible to execute speech recognition for speech from an arbitrary direction without a delay associated with estimation of the sound source direction.

  More specifically, voice recognition by the first voice recognition processing unit 13 is continuously executed regardless of whether or not the sound source direction is determined. For this reason, when the user performs an utterance for causing the voice recognition apparatus 1 to execute some processing (for example, the above-described device operation or the like), at least voice recognition by the first voice recognition processing unit 13 is immediately executed. . Therefore, when the speech recognition apparatus 1 succeeds in the speech recognition, the speech recognition apparatus 1 can appropriately execute some processing based on the speech recognition result without the delay accompanying the estimation of the sound source direction.

  Furthermore, after the sound source direction is determined, the second speech recognition processing unit 15 performs higher-accuracy speech recognition on the sound stream from the determined sound source direction, thereby improving the accuracy of speech recognition. It becomes possible. Therefore, according to the speech recognition apparatus 1, it is possible to accurately recognize speech from an arbitrary direction without a delay associated with estimation of the sound source direction.

  Next, a voice recognition program for causing a computer to execute the above-described series of processing by the voice recognition device 1 will be described. The speech recognition program P1 is inserted into a computer and accessed, or stored in a program storage area formed on a recording medium included in the computer.

  As shown in FIG. 5, the speech recognition program P1 includes a speech input module P11, a directivity control module P12, a first speech recognition processing module P13, a sound source direction determination module P14, a second speech recognition processing module P15, and speech recognition results. An output module P16 is provided. Functions realized by executing the voice input module P11, the directivity control module P12, the first voice recognition processing module P13, the sound source direction determination module P14, the second voice recognition processing module P15, and the voice recognition result output module P16. The voice input unit 11, the directivity control unit 12, the first voice recognition processing unit 13, the sound source direction determination unit 14, the second voice recognition processing unit 15, and the voice recognition result output unit 16 of the voice recognition device 1 described above, respectively. It is the same.

  Note that a part or all of the voice recognition program P1 may be transmitted via a transmission medium such as a communication line and received and recorded (including installation) by another device. Further, each module of the speech recognition program P1 may be installed in any one of a plurality of computers instead of one computer. In that case, the above-described series of processing of the speech recognition apparatus 1 is performed by the computer system of the plurality of computers.

  As mentioned above, although preferred embodiment and the modification of this invention were demonstrated, this invention is not restricted to the said embodiment, A various deformation | transformation is possible in the range which does not deviate from the summary.

  DESCRIPTION OF SYMBOLS 1 ... Voice recognition apparatus, 11 ... Voice input part, 12 ... Directivity control part, 13 ... 1st voice recognition processing part, 14 ... Sound source direction determination part, 15 ... 2nd voice recognition processing part, 16 ... Voice recognition result output P1 ... voice recognition program, P11 ... voice input module, P12 ... directivity control module, P13 ... first voice recognition processing module, P14 ... sound source direction determination module, P15 ... second voice recognition processing module, P16 ... voice recognition Result output module.

Claims (7)

Audio acquisition means for acquiring each of audio streams from a plurality of directions;
Speech recognition processing means for performing speech recognition on each of the audio streams from a plurality of directions acquired by the sound acquisition means;
When the voice recognition processing unit obtains a voice recognition result that satisfies a predetermined reliability criterion, sound source direction determination is performed to determine the direction corresponding to the voice stream from which the voice recognition result is obtained as the sound source direction. Means,
A voice recognition with higher accuracy than the voice recognition by the voice recognition processing means is performed on the voice stream from the sound source direction determined by the sound source direction determination means among the voice streams acquired by the voice acquisition means. A speech recognition apparatus comprising: 2 speech recognition processing means . The voice acquisition unit, by setting the directional beam in a plurality of predetermined direction, acquires the audio stream corresponding to the beam direction of each directional beam, according to claim 1 Symbol placement of the speech recognition device. The audio acquisition unit acquires an audio stream corresponding to the beam direction of each directional beam by setting directional beams in a plurality of directions that are candidates for a sound source direction estimated by a predetermined method. 1 Symbol placement of the voice recognition device. The voice recognition processing means, when the word predetermined is included in the speech recognition result, the speech recognition result is judged to meet the criteria of the predetermined reliability, any claim 1-3 The speech recognition device according to claim 1. The voice recognition processing unit performs voice activity detection process for detecting a speech period, performing speech recognition on the detected speech segment by the speech section detection processing, any one of claims 1-4 The speech recognition apparatus according to one item. A speech recognition method executed by a speech recognition device,
An audio acquisition step for acquiring each of audio streams from a plurality of directions;
A speech recognition processing step of executing speech recognition for each of the sound streams from the plurality of directions acquired in the sound acquisition step;
In the speech recognition processing step, when a speech recognition result satisfying a predetermined reliability criterion is obtained, sound source direction determination is performed to determine a direction corresponding to the sound stream from which the speech recognition result is obtained as a sound source direction. Steps,
A voice recognition with higher accuracy than the voice recognition in the voice recognition processing step is performed on the voice stream from the sound source direction determined in the sound source direction determination step among the voice streams acquired in the voice acquisition step. A voice recognition method including two voice recognition processing steps . Computer
Audio acquisition means for acquiring each of audio streams from a plurality of directions;
Speech recognition processing means for performing speech recognition on each of the audio streams from a plurality of directions acquired by the sound acquisition means;
When the voice recognition processing unit obtains a voice recognition result that satisfies a predetermined reliability criterion, sound source direction determination is performed to determine the direction corresponding to the voice stream from which the voice recognition result is obtained as the sound source direction. and means,
A voice recognition with higher accuracy than the voice recognition by the voice recognition processing means is performed on the voice stream from the sound source direction determined by the sound source direction determination means among the voice streams acquired by the voice acquisition means. A voice recognition program to be executed as two voice recognition processing means .

Images