JP6517670B2 - Speech recognition apparatus, speech recognition method and speech recognition program - Google Patents

Description

  The present invention relates to a speech recognition apparatus, a speech recognition method, and a speech recognition program.

  Conventionally, in an internet browser, smart TV, tablet, smart phone terminal, the voice uttered by the user is compressed by the terminal (hereinafter, the terminal is also referred to as a client) and sent to the cloud side as it is. Speech recognition is performed using a cloud-based speech recognition system in which recognition processing is performed and the terminal side receives and uses the result via a network. In addition, it does not send to the cloud side as it is compressed, only the acoustic analysis is performed by the terminal, only the feature amount is sent, and a cloud type speech recognition system using a method called DSR (Distributed Speech Recognition) that performs speech recognition on the cloud side is used. Speech recognition is also performed.

  In addition, there is also a technology that uses the HTML5 browser to acquire speech with a Web application, sends compressed speech to the cloud side with Web socket communication, performs speech recognition on the cloud side, and uses the result on the terminal (For example, refer to nonpatent literature 1).

  In these mechanisms, with the abundant CPU resources and statistical data on the cloud side, removal of background noise, separation of speech and non-voice sections by sound pressure and acoustic model, phoneme analysis, statistical analysis by language model, etc. Speech recognition can be performed with accuracy. Also, in recent years, a speech recognition method using deep learning technology to combine CPU and GPU resources to make each process more accurate, or using an deep learning technology to convert an acoustic feature directly to a sentence into an integrated model Is also proposed.

  Furthermore, by increasing the accuracy of the language model daily, such as increasing weighting, for words that are frequently used by current users, such as new words, buzzwords, and current events words, for example, big data obtained on the cloud side, for example Accurate and practical speech recognition can be performed.

  For example, the following (1) to (3) can be mentioned as conventional techniques for improving the speech recognition accuracy, but there are problems as described below.

(1) Echo canceler When a loop is formed in which a signal reproduced by a speaker is collected by a microphone and the collected audio signal is included in a signal reproduced by the speaker, echo or howling occurs. . An echo canceller is known as a conventional technique for removing and reducing the echo and howling. An echo canceller is used to reduce echo and howling, and by applying a filtering process to remove an acoustic signal generated by a speaker from an acoustic signal obtained by a microphone, echo or howling is affected in the apparatus. Except for However, in the case where a plurality of independent devices are combined and speakers are different as in a living room of a home, voices flowing out from other speaker devices can not be removed or reduced by the echo canceller. Also, even when voice is sent directly from the smartphone application or Web API to the cloud, the voice of the video is directly output from the video playback application that is another application, and the acoustic signal to be canceled is from the other application or API Is difficult to grasp and can not be removed or reduced by the echo canceller.

(2) Difference in sound pressure As a conventional technique for distinguishing between background sound and speech uttered by a speaker toward a microphone, there is a method of using the difference in sound pressure, that is, the power of an acoustic signal. However, when the playback volume of a television or radio is large, when the speaker of the television or radio is close to the microphone, or when the microphone and the speaker are away, the sound pressure between the background sound and the speaker's voice is large In some cases, it is difficult to distinguish between the background sound and the speaker's voice because there is no

(3) Separation by Acoustic Model As conventional technology for separating human voice and environmental noise such as air conditioning noise and engine sound, there is a sound source separation technology focusing on the difference in generation model of the sound of the generation source. Further, in recent years, a sound source separation technology based on advanced feature amount determination using deep learning is also being developed, and the accuracy of the sound source separation has been rapidly improved (see, for example, Non-Patent Document 2). However, live radio television radios and information broadcasts themselves are "human voice", so if the playback quality of the audio device that plays the audio of the broadcast in the background is high or the volume of playback is high. There are cases where the background speech and the speech of the speaker can not be sufficiently separated.

[Online], [October 13, 2015 search], Internet <http://www.ntt.co.jp/news2014/1409/140911a.html> [Online], [October 15, 2015 search], Internet <http://www.ntt.co.jp/journal/1509/files/jn201509017.pdf>

  However, when there is a voice with a relatively large volume in the background of the voice that the speaker wants to recognize, there is a background other than the sound signal of the voice emitted by the speaker in the sound signal targeted for voice recognition. The voice recognition result of the background is not included in the voice recognition result, and the voice recognition result different from the voice recognition result desired by the speaker is included. There is a problem of being obtained.

  For example, when a television or radio broadcast is constantly flowing at a relatively high volume and in the background at a location that emits voice recognition target voices, voices emitted from the speakers of these televisions or radio during an identification period of voice recognition And serif) are mixed unintentionally and there is a problem that they may not be recognized correctly (task 1).

  In addition, voice recognition is performed when sound recognition is performed on guidance broadcasts such as outdoor stadiums, lecture halls, public viewing venues, in a train, etc., and environments where in-house broadcasts frequently flow at high volumes. There is also a problem that voices (announcements or words) in a guide broadcast or in-house broadcast enter in the middle of the identification period of, and those voice recognition results are unintentionally mixed with the voice search results (Problem 2).

  In addition, when playing a recorded movie or playing VOD while downloading, voice recognition identification is performed in the case where the application voice during video playback and the voice recognition application individually operate in multitasking. During a period, there is also a problem that the voice of the moving image (announcement or speech) is unintentionally mixed with the voice recognition result (task 3).

  The present invention has been made in view of such circumstances, and performs voice recognition of an acoustic signal including the voice of the input speaker to obtain a recognition result, and performs voice recognition of another voice signal input. To reduce the possibility of including unnecessary speech recognition results by obtaining recognition results and removing common ones of these recognition results from speech recognition results of acoustic signals including the voice of the input speaker. It is an object of the present invention to provide a speech recognition device, a speech recognition method and a speech recognition program capable of improving the speech recognition rate for the speech of the speaker.

  In order to solve the above problems, the present invention performs speech recognition of an acoustic signal including the voice of the input speaker to obtain a recognition result, and performs speech recognition of another acoustic signal input to obtain a recognition result. These common recognition results are removed from the speech recognition result of the sound signal including the speech of the input speaker.

  One aspect of the present invention is a first sound signal which is a sound signal collected by the first sound collecting means including the voice of the first speaker and one or more different from the first sound collecting means The second to N-th (N is an integer of 2 or more) sound-pickup means, which are sound-pickup means, respectively recognize the sound signals of the second sound signal to the N-th sound signal which are sound signals Voice recognition means for obtaining a first voice recognition result to an N-th voice recognition result as voice recognition results for respective sound signals, and at least one voice recognition result of the second voice recognition result to the N-th voice recognition result The partial speech recognition result included in the partial speech recognition result and the partial speech recognition result included in the first speech recognition result have the same content of the partial speech recognition result, and the partial speech recognition result corresponding to the acoustic signal at substantially the same time When the partial speech recognition result is the first speech A speech recognition result processing means for obtaining those removed from identification result as the speech recognition result of the first speaker, a voice recognition device equipped with.

  According to one aspect of the present invention, there is provided a first acoustic signal which is an acoustic signal collected by the first sound collecting means including the voice of a first speaker and a first broadcast which is an acoustic signal of one or more broadcasts. The voice signal of each of the sound signal to the Mth broadcast sound signal (M is an integer of 1 or more) is voice-recognized, and the first voice recognition result and the first broadcast voice recognition as the voice recognition result for each sound signal Speech recognition means for obtaining a result of the M-th broadcast speech recognition, a partial speech recognition result included in at least one or more speech recognition results of the first broadcast speech recognition result to the M-th broadcast speech recognition result, and the first speech If the partial speech recognition result contained in the recognition result is the same as the partial speech recognition result corresponding to the acoustic signals at approximately the same time, the content of the partial speech recognition result is the same, the partial speech recognition result is Deleted from the first speech recognition result A speech recognition apparatus and a speech recognition result processing means for obtaining a speech recognition result of the first speaker.

  One aspect of the present invention is a first sound signal which is a sound signal collected by the first sound collecting means including the voice of the first speaker and one or more different from the first sound collecting means Sound signals of the second to N-th sound signals which are sound signals collected respectively by the second to N-th (N is an integer of 2 or more) sound-pickup means. Voice recognition means for obtaining a first voice recognition result to a N-th voice recognition result as voice recognition results for respective sound signals by performing voice recognition, and the second voice recognition result to the N-th voice recognition result The partial speech recognition result included in at least one or more speech recognition results corresponding to the sound signal collected in the vicinity of the sound collection means and the partial speech recognition result included in the first speech recognition result are partial speech Partial speech recognition results at the same time, with the same content of the recognition result If it is a speech recognition apparatus, comprising: a speech recognition result processing means for obtaining the partial speech recognition result obtained by deleting from the first speech recognition result as a voice recognition result of the first speaker, a.

  According to one aspect of the present invention, there is provided a first acoustic signal which is an acoustic signal collected by the first sound collecting means including the voice of a first speaker and a first broadcast which is an acoustic signal of one or more broadcasts. The voice signal of each of the sound signal to the Mth broadcast sound signal (M is an integer of 1 or more) is voice-recognized, and the first voice recognition result and the first broadcast voice recognition as the voice recognition result for each sound signal Voice recognition means for obtaining a result of the Mth broadcast speech recognition, and at least one or more of the first broadcast speech recognition result to the Mth broadcast speech recognition result in which the first sound collecting means is in a broadcast reception target area The partial speech recognition result included in the speech recognition result and the partial speech recognition result included in the first speech recognition result have the same content of the partial speech recognition result, and correspond to acoustic signals at substantially the same time In the case of a partial speech recognition result, the partial sound Recognition result is a voice recognition device and a speech recognition result processing means for obtaining a speech recognition result of the first speaker and those removed from the first speech recognition result.

  One aspect of the present invention is a first sound signal which is a sound signal collected by the first sound collecting means including the voice of the first speaker and one or more different from the first sound collecting means Sound signals of the second to N-th sound signals which are sound signals collected respectively by the second to N-th (N is an integer of 2 or more) sound-pickup means. A voice recognition step of obtaining a first voice recognition result to an N-th voice recognition result which is voice recognition result for each sound signal by voice recognition, and at least one of the second voice recognition result to the N-th voice recognition result The partial speech recognition result included in the speech recognition result and the partial speech recognition result included in the first speech recognition result have the same content of the partial speech recognition result, and correspond to acoustic signals at substantially the same time If it is a partial speech recognition result, the partial speech recognition result A speech recognition method and a speech recognition result processing step of obtaining those removed from the first speech recognition result as a voice recognition result of the first speaker.

  According to one aspect of the present invention, there is provided a first acoustic signal which is an acoustic signal collected by the first sound collecting means including the voice of a first speaker and a first broadcast which is an acoustic signal of one or more broadcasts. The voice signal of each of the sound signal to the Mth broadcast sound signal (M is an integer of 1 or more) is voice-recognized, and the first voice recognition result and the first broadcast voice recognition as the voice recognition result for each sound signal A voice recognition step of obtaining a result of the M-th broadcast speech recognition result, a partial speech recognition result included in at least one or more voice recognition results of the first broadcast speech recognition result to the M-th broadcast speech recognition result, and the first voice If the partial speech recognition result contained in the recognition result is the same as the partial speech recognition result corresponding to the acoustic signals at approximately the same time, the content of the partial speech recognition result is the same, the partial speech recognition result is Deleted from the first speech recognition result A speech recognition result processing step of obtaining a speech recognition result of the first speaker from a speech recognition method comprising.

  One aspect of the present invention is a first sound signal which is a sound signal collected by the first sound collecting means including the voice of the first speaker and one or more different from the first sound collecting means Sound signals of the second to N-th sound signals which are sound signals collected respectively by the second to N-th (N is an integer of 2 or more) sound-pickup means. A voice recognition step of performing voice recognition to obtain first to Nth voice recognition results as voice recognition results for respective acoustic signals; and the second voice recognition result to the Nth voice recognition result The partial speech recognition result included in at least one or more speech recognition results corresponding to the sound signal collected in the vicinity of the sound collection means and the partial speech recognition result included in the first speech recognition result are partial speech Partial speech recognition with the same content of recognition result and almost the same time If the result, the speech recognition method comprising a speech recognition result processing step of obtaining the partial speech recognition result obtained by deleting from the first speech recognition result as a voice recognition result of the first speaker, a.

  According to one aspect of the present invention, there is provided a first acoustic signal which is an acoustic signal collected by the first sound collecting means including the voice of a first speaker and a first broadcast which is an acoustic signal of one or more broadcasts. The voice signal of each of the sound signal to the Mth broadcast sound signal (M is an integer of 1 or more) is voice-recognized, and the first voice recognition result and the first broadcast voice recognition as the voice recognition result for each sound signal A voice recognition step for obtaining a result of the Mth broadcast speech recognition, and at least one or more of the first broadcast speech recognition result to the Mth broadcast speech recognition result in which the first sound collecting means is in a broadcast reception target area The partial speech recognition result included in the speech recognition result and the partial speech recognition result included in the first speech recognition result have the same content of the partial speech recognition result, and correspond to acoustic signals at substantially the same time If it is a partial speech recognition result, that part A speech recognition result processing step of obtaining a speech recognition result obtained by deleting from the first speech recognition result as a voice recognition result of the first speaker, a speech recognition method comprising.

  One aspect of the present invention is a speech recognition program for operating a computer as the speech recognition device.

  According to the present invention, unnecessary speech recognition results are included by removing a portion common to the speech recognition result of another input sound signal from the speech recognition result of the sound signal including the speech of the input speaker. By reducing the possibility of being voiced, the effect of being able to improve the speech recognition rate for the speech of the speaker can be obtained.

BRIEF DESCRIPTION OF THE DRAWINGS It is a block diagram which shows the structure of the whole system containing the speech recognition system of 1st, 3rd embodiment of this invention. It is a block diagram which shows the structure by one client side apparatus and the cloud side apparatus of the speech recognition system of 1st Embodiment of this invention. It is a figure which shows the flow of a process of the speech recognition result process part of the 1st operation example of 1st Embodiment of this invention. It is a figure for demonstrating the specific example of the speech recognition result process part of the 1st operation example of 1st Embodiment of this invention. It is a figure for demonstrating the specific example of the speech recognition result process part of the 2nd operation example of 1st Embodiment of this invention. It is a figure for demonstrating the specific example of the speech recognition result process part of the 3rd operation example of 1st Embodiment of this invention. It is a figure which shows the flow of a process of the speech recognition result process part of the 4th operation example of 1st Embodiment of this invention. It is a figure which shows the flow of a process of the modification of the speech recognition result process part of the 4th operation example of 1st Embodiment of this invention. It is a figure which shows the flow of a process of the speech recognition result process part of the 6th operation example of 1st Embodiment of this invention. It is a block diagram which shows the structure of the whole system containing the speech recognition system of 2nd Embodiment of this invention. It is a block diagram which shows the component part by one client side apparatus of the speech recognition system of 2nd Embodiment of this invention, and a cloud side apparatus. It is a figure which shows the flow of a process of the speech recognition result process part of the 1st operation example of 2nd Embodiment of this invention. It is a figure for demonstrating the specific example of the speech recognition result process part of the 1st operation example of 2nd Embodiment of this invention. It is a figure which shows the flow of a process of the speech recognition result process part of the 3rd operation example of 2nd Embodiment of this invention. It is a block diagram which shows the component part by one client side apparatus of 3rd Embodiment of this invention, and a cloud side apparatus. It is a figure which shows the flow of a process of the speech recognition result process part of 3rd Embodiment of this invention. It is a block diagram which shows the component part by one client side apparatus of 4th Embodiment of this invention, and a cloud side apparatus. It is a block diagram which shows the component part by one client side apparatus of the modification of 4th Embodiment of this invention, and a cloud side apparatus. It is a figure for demonstrating the specific example of the speech recognition result process part of the modification of 4th Embodiment of this invention. It is a block diagram which shows the structure of the speech recognition apparatus of this invention.

  Hereinafter, a speech recognition system according to an embodiment of the present invention will be described with reference to the drawings.

  Here, the usage form assumed by the present invention will be described. The present invention assumes that the voice recognition microphone and the speaker are not close to each other, but are used when the distance is relatively far, specifically, in a case where there is a relatively long distance of about 45 cm to several meters. .

  Assumed peripheral conditions are during the recognition period, such as television broadcasts outside this system, and voices (announcements and speeches) of radio broadcasts flowing at intervals of a few seconds to a few tens of seconds, and guidance broadcasts irregularly, etc. There are cases such as cases where voices such as television, radio, and guide broadcasting occasionally enter an audio signal input to the client-side device.

First Embodiment
First, as a first embodiment of the present invention, from the speech recognition result of an acoustic signal including the voice of the speaker input to the client side device, it is common to the speech recognition result of an acoustic signal input to another client side device The form which removes a part is demonstrated. FIG. 1 is a block diagram showing the configuration of the speech recognition system in the first embodiment. In this figure, reference numeral 100 is a speech recognition system, reference numerals 1 ₁ to 1 _N are a plurality of (N, N is an integer of 2 or more) client side devices, and 2 is a cloud side device. The client side devices 1 ₁ to 1 _N are devices used by the user, and for example, a smartphone, a smart TV, an HDMI (registered trademark) dongle STB ("STB" is a "set top box"), a small STB, an advanced home appliance Devices, game consoles, etc. The cloud side device 2 is connected to the client side devices 1 ₁ to 1 _N via the network 3. The network 3 is for enabling the client side devices 1 ₁ to 1 _N of the speech recognition system 100 and the cloud side device 2 to transmit and receive information according to the connection protocol of the Internet, and is, for example, the Internet. Since all the configurations included at least in the client side devices 1 ₁ to 1 _N are the same, in the following, a portion configured by the client side device ₁₁ and the cloud side device 2 in the speech recognition system 100 of the first embodiment This will be described using FIG. 2 which is a detailed block diagram of FIG.

Client device _{1 1} includes an audio input unit 11 _1, the user information acquiring unit 12 _1, the audio output unit 13 _1, the search result receiving unit 14 _1, configured to include at least the screen display unit 15 _1.

  The cloud side device 2 includes at least a voice receiving unit 21, a voice recognition unit 22, a voice recognition result holding unit 23, a voice recognition result processing unit 24, a search processing unit 25, and a search result transmission unit 26.

  Next, the operation of the speech recognition system according to the first embodiment will be described.

[[First operation example of the first embodiment]]
As a first operation example, each user of the 1~N is not using the client-side apparatus 1 ₁ to 1 _N, the sentence to be obtained search results first user to the client-side apparatus 1 ₁ the speaks, an example of operation of displaying a search result corresponding to the utterance to the client-side apparatus 1 ₁ of the screen display unit 15 _1. Here, as a more specific case, N = 3, and the first user is at a place where the sound of a television or radio is flowing or the information broadcasting such as a station or department store is irregularly A second user is in the same place where the same television, radio and information broadcast are flowing as the first user, and the same television, radio and information broadcast as the first user is not flowing and there is a different environmental sound The case where there is a third user is described as an example.

Client device 1 ₁ of the speech input unit 11 ₁ obtains an acoustic signal emitted around the client-side apparatus 1 _1, and outputs the acquired audio signal to the audio transmission unit 13 _1. If the first user has uttered a sentence to be obtained search results to the client-side apparatus 1 _1, obtains and outputs a sound signal including a voice first user uttered. If the client-side apparatus 1 ₁ of environmental sounds such as a television or radio or announcement around has occurred, and outputs the acquired audio signal including the ambient sound. Therefore, in the above specific case, an acoustic signal composed of the voice uttered by the first user and the sound of a television or radio, or an environmental sound such as a guidance broadcast is acquired and output.

User information acquisition section 12 ₁ of the client-side apparatus 1 ₁ obtains the time information voice input unit 11 ₁ of the client-side apparatus 1 ₁ acquires the audio signal, which can specify the time information and the client-side apparatus 1 ₁ identification information (hereinafter, referred to as "ID") to the audio transmission unit 13 ₁ and the user information. The time information, for example, absolute time, for example, if the client-side apparatus 1 ₁ is a smart phone with a built-in GPS receiver, the GPS receiver when the smartphone microphone is a voice input unit 11 ₁ obtains a sound signal The absolute time received by the unit may be used as time information. Further, for example, it may be time information obtained from a base station of a mobile carrier network or a communication server, or may be time information of a local clock held by the OS of the smartphone. In the first embodiment, the time information is used by the voice recognition result processing unit 23 to specify whether the times at which the acoustic signals acquired by each of the plurality of client-side devices are the same are the same. Since the time is a common time among a plurality of client side apparatuses, it may not be the absolute time itself.

Voice output unit 13 ₁ of the client-side apparatus 1 ₁ sends a transmission signal including the user information acoustic signal and the user information acquisition section 12 ₁ speech input unit 11 ₁ is output is output to the cloud side device 2 . More precisely, the audio output unit 13 _1, the transmission signal including the user information acoustic signal and the user information acquisition section 12 ₁ speech input unit 11 ₁ is output is output, to convey to the cloud side device 2, the network Send to 3 Transmission of the transmission signal is performed, for example, at predetermined time intervals such as 10 ms. The audio output unit 13 ₁ obtains a code string acoustic signal is encoded by a predetermined encoding method, it may be sent a transmission signal including a resultant code sequence and the user information. The audio output unit 13 ₁ performs such feature extraction, which is part of the processing of the speech recognition process on the acoustic signals, sends a transmission signal including a feature amount and the user information obtained by the process May be There are many well-known techniques and well-known techniques in the technique of sending transmission signals to another device via a network and the technique of performing speech recognition processing in a distributed manner between the client-side device and the cloud-side device. I omit explanation.

Client device ₁ 2 to 1 _N audio input unit ₁₁ 2 to 11 _N, the user information acquiring unit ₁₂ 2 to 12 _N and the voice sending section ₁₃ 2 to 13 _N be respectively client device _{1 1} of the speech input unit 11 _1, the user information acquiring unit 12 ₁ and the audio output unit 13 ₁ and the same operation. Therefore, if the above specific case, the client-side apparatus 1 _2, and audio second user has uttered, the same environment sound and the first user, to obtain a composed audio signal by, A transmission signal including the sound signal and user information is transmitted. Moreover, the client-side apparatus 1 _3, and audio third user has uttered, and different environmental sound from the first user, to obtain a composed acoustic signal by a corresponding acoustic signal and the user information Send out a transmission signal including

The voice receiving unit 21 of the cloud side device 2 receives the transmission signals respectively sent by the voice sending portions 13 _{1 to} 13 _N of the client side devices 1 ₁ to 1 _N , and generates an acoustic signal and a user from the received transmission signals. Extract and output a pair with information. Reception of the transmission signal is performed, for example, at predetermined time intervals such as 10 ms. When the voice transmitting units 13 _{1 to} 13 _N encode the acoustic signal according to a predetermined coding method to obtain a code string and transmit a transmission signal including the obtained code string, the voice reception of the cloud-side device 2 is performed The unit 21 obtains an acoustic signal by decoding a code string included in the received transmission signal according to a decoding method corresponding to a predetermined encoding method, and outputs a set of the obtained acoustic signal and user information. Good. In addition, the voice transmitting units 13 _{1 to} 13 _N perform feature extraction, which is a part of processing of voice recognition processing, on the sound signal, and transmit signals including the feature and the user information obtained by the processing. In the case of sending, it is sufficient to take out not the acoustic signal but the feature amount from the transmission signal, and output the set of the extracted feature amount and the user information.

  The voice recognition unit 22 of the cloud-side device 2 performs voice recognition processing on each of the sound signals output from the voice reception unit 21, and obtains a voice recognition result that is a character string corresponding to the voice included in the sound signal. And outputting a set of a speech recognition result, time information corresponding to the speech recognition result, and an ID corresponding to the speech recognition result. If there is no time information or an inappropriate value, the server may manage the data based on the approximate time information received without using the received time information.

  Speech recognition processing is performed for each predetermined set of acoustic signals. For example, the voice recognition unit 22 sequentially stores the sound signal output from the voice reception unit 21 in a storage unit (not shown) in the voice recognition unit 22, and performs voice period detection on the stored voice signal. The voice recognition process is performed on the voice signal clusters in each utterance section to obtain a voice recognition result as a character string for the voice signal clusters in each utterance section. Also, for example, even if the speech recognition unit 22 performs speech recognition processing on a group of sound signals in a plurality of speech segments, and obtains a speech recognition result that is a character string to a group of sound signals in a plurality of speech segments. Good.

Therefore, if the above specific case, the speech recognition unit 22, as a speech recognition result to the client-side apparatus 1 ₁ of the acoustic signal, the first user utterance was such as voice and TV or radio or announcement Environment A string consisting of a sound and a speech recognition result is obtained. The speech recognition unit 22, as a speech recognition result to the client-side apparatus 1 _second acoustic signal consists of a speech recognition result of the second user utterance voice and the same environment sound and the first user character Get a row. The speech recognition unit 22, as a speech recognition result to the client-side apparatus 1 _third acoustic signal from a sound third user has uttered the first user and the speech recognition result of the different environmental sound Get a string of

  A well-known speech recognition technology may be used for the speech recognition process. That is, as the details of the speech recognition processing such as an acoustic model, a language model, and how to obtain a feature amount, known ones may be used. Further, an integrated voice recognition process using deep learning or the like may be used as the voice recognition process. In these speech recognition processes, speech may be recognized after the sound signal is analyzed to be a prescribed section by an analysis unit (not shown) or the like. In addition, when the voice receiving unit 21 outputs a feature amount instead of the sound signal, the voice recognition unit 22 may perform the voice recognition process using the feature amount. In any case, the speech recognition process itself includes many well-known techniques and well-known techniques, so detailed description will be omitted.

  Since the speech recognition process is performed for each predetermined set of acoustic signals, the character string of the speech recognition result corresponds to the predetermined set of acoustic signals. Therefore, the speech recognition unit 22 obtains representative time from time information included in a plurality of user information corresponding to a predetermined group of sound signals targeted for speech recognition processing, for example, and the time information representing the representative time is Pair with speech recognition results. The representative time may be any one that represents the time at which the sound signal corresponding to the character string of the speech recognition result is emitted. For example, the time at the beginning of the speech recognition result character string may be used as the representative time. Also, there may be a plurality of representative times in one speech recognition result. For example, for each partial character string such as a word included in the speech recognition result, the time at the beginning of the word or the like may be used as the representative time.

  The ID to be paired with the speech recognition result is included in the user information input from the speech receiving unit 21 in combination with the ID corresponding to the speech recognition result, that is, the acoustic signal from which the speech recognition result is obtained. ID.

  The voice recognition result holding unit 23 of the cloud-side device 2 stores a set of the voice recognition result, the time information, and the ID output from the voice recognition unit 22. The contents stored in the voice recognition result storage unit 23 are processing for determining whether or not the voice recognition result processing unit 24 has a partial character string such as a word having a common time, and a partial character string such as a word having a common time Is used for processing to obtain a processed speech recognition result by removing it from the speech recognition result when there is an error. Therefore, the speech recognition result holding unit 23 stores the combination of the speech recognition result, the time information, and the ID output from the speech recognition unit 22 only for the time required for the process of the speech recognition result processing unit 24. Further, the stored contents held in the speech recognition result holding unit 23 may be deleted when the processing of the speech recognition result processing unit 24 using the stored contents is completed.

  The voice recognition result processing unit 24 of the cloud-side device 2 includes a portion of the combination of at least one voice recognition result, time information, and ID stored in the voice recognition result holding unit 23 in the character string of the voice recognition result. For each character string, if there is a match between the partial character string and the time among the other voice recognition result, the time information, and the ID pair, the one obtained by removing the matched partial character string is As a processed speech recognition result, the processed speech recognition result and the ID are output as a set. Therefore, processed speech recognition results for at least one client-side device will be output. The determination as to whether or not the times coincide may be determined as the same time in consideration of an error in absolute time in each client-side device, an error in time in speech recognition processing, and the like. That is, for at least one processing target client side device, when another client side device has the same partial character string (common partial character string) as the processing target client side device at substantially the same time, What is obtained by removing a common partial character string from the character string of the speech recognition result of the processing target client side device is obtained as a processed speech recognition result.

  Note that this process may be performed on all other client-side devices, or may be performed on at least one other client-side device. In this case, only the speech recognition result necessary for the processing of the speech recognition result processing unit 24 may be obtained in the previous stage. That is, the operations of the voice receiving unit 21, the voice recognition unit 22, and the voice recognition result holding unit 23 for obtaining a voice recognition result unnecessary for the processing of the voice recognition result processing unit 24 may be omitted.

Here, in the above example of N = 3, it will be described with reference to FIGS. 3 and 4 for example at least some one client device is a client-side apparatus 1 _1. FIG. 3 is a view for explaining the processing flow of the speech recognition result processing unit 24 in this operation example, and FIG. 4 is a view for explaining an example of the speech recognition result and the processed speech recognition result in this example. The example of FIG. 3, the subject of each of all the client device other than the client-side apparatus 1 _1, in order from the speech recognition result of the client-side apparatus 1 _2, the speech recognition result of the client-side apparatus 1 ₁ and the partial strings and time If there is a match between the partial character string and the time, there is a partial character string that matches the partial character string and the time pair. it is an example of the client-side apparatus 1 ₁ of the speech recognition result string will remove the common substring.

Speech recognition result processing unit 24 first reads a set of the speech recognition result of the client-side apparatus 1 ₁ and the time information and the ID from the speech recognition result holding unit 23 (step S241). Next, the speech recognition result processing unit 24 sets an initial value x to 2 (step S242). Next, the voice recognition result processing unit 24 reads out a combination of the voice recognition result of the client side device _1x , time information, and an ID from the voice recognition result holding unit 23 (step S243). Speech recognition result processing unit 24, then, in the set of the set and the speech recognition result of the client-side apparatus 1 _x and the time information and the ID of the speech recognition result of the client-side apparatus 1 ₁ and the time information and ID, partial It is searched whether there is a match between the character string and the time (step S244). In this case, since there is an error in the time, the coincidence is determined in approximately time. This is, for example, within a few seconds. Hereinafter, the expression “time matching” in the flow description will be treated in the same manner unless otherwise specified. Next, when there is a partial character string and one whose time coincides with the partial character string at step S244, the speech recognition result processing unit 24 uses the partial character string and all the partial character strings whose time coincides with one another on the client side device 1 _{It is} removed from the character string of the speech recognition result of ₁ (step S245). If there is no match between the partial character string and the time in step S244, the process proceeds to step S246. Next, the speech recognition result processing unit 24 determines whether there remains any client-side device not subject to the processes of steps S243 to S245 (step S246). The voice recognition result processing unit 24 next replaces x with x + 1 if it is determined in step S246 that there is a client-side device not subject to the processing in steps S243 to S245 (step S247). Step If it is determined that there are no remaining client device that is not subject to processing in step S243~ step S245 in S246, the last at step S245 the processed client device 1 ₁ of the speech recognition result of characters column in the ID paired as client-side apparatus 1 ₁ of processed speech recognition result string output (step S248).

Next, an example of the speech recognition result and the processed speech recognition result in this example will be described with reference to FIG. The horizontal axis of FIG. 4 is time, and three above the arrow are the speech recognition results of the client side devices 1 1 _to 1 ₃ which are the input of the speech recognition result processing unit 24 and 1 below the arrow The _{first is} the processed speech recognition result of the client side device 11. The client-side apparatus 1 ₁ of the speech recognition result, the first and the partial string of the speech recognition result of speech 1 and utterance 2 is a speech the user has issued a client device 1 ₁ of the user, the client-side TV around the apparatus 1 ₁ contains a substring of the speech recognition result of the television audio 1 and the television audio 2 is a sound produced by. Moreover, the client-side apparatus 1 ₂ speech recognition result, a partial character string of the speech recognition result of the speech 3 and speech 4 is a utterance second user is a client-side apparatus 1 ₂ user uttered, substring of a speech recognition result of the client-side apparatus 1 ₂ of the television audio 1 is a audio television emitted in and around the television audio 2 are included. Further, the speech recognition result is the client-side apparatus 1 _3, and the client-side apparatus 1 ₃ third user utterance is a spoken 1 and utterance 2 of the speech recognition result emitted partial string is a user of, substring of a speech recognition result of the television audio 3 and video audio 4 is a speech television emitted around the client-side apparatus 1 ₃ are included. Here, partial character strings of speech recognition results of the speech 1 and the speech 2 which are the speech of the first user, and speech recognition results of the speech 1 and the speech 2 which are the speech of the third user. The partial character strings are assumed to be identical to each other.
In order to make the figure easy to understand, an example of an utterance voice example and an example of a character string of television voice will be written together on the voice recognition result of FIG. Utterance examples are usually written in the body, and television speech examples are in italics. In the figure, the partial character strings are written for each word, but in fact, they may be short partial character strings such as one phoneme.

First, the process of step S244 and step S245 of FIG. 3 when x = 2 will be described. The client device 1 ₁ of the substring of the speech recognition result of the television audio 1 and the television audio 2 of the partial character strings included in the speech recognition result, also includes at the same time the speech recognition result of the client-side apparatus 1 ₂ It is therefore removed from the speech recognition result of the client-side apparatus 1 _1. The substring of the utterance 1 and utterance 2 of the speech recognition result of the partial character strings included in the client-side apparatus 1 ₁ for speech recognition result is not included at the same time the speech recognition result of the client-side apparatus 1 ₂ Therefore, it not removed from the speech recognition result of the client-side apparatus 1 _1. In other words, a state in which a substring of the speech recognition result of speech 1 and utterance 2 is left as a partial character string included in the speech recognition result of the client-side apparatus 1 ₁ proceeds to the process in the case of x = 3.

Next, the process of step S244 and step S245 of FIG. 3 when x = 3 will be described. The client device 1 ₁ of the substring of the speech recognition result of speech 1 and utterance 2 of the partial character strings included in the speech recognition result, although included in the voice recognition result of the client-side apparatus 1 _3, client device because it is not included at the same time in the 1 ₃ speech recognition result, it is not removed from the speech recognition result of the client-side apparatus 1 _1. In other words, a state in which a substring of the speech recognition result of speech 1 and utterance 2 is left as a partial character string included in the speech recognition result of the client-side apparatus 1 _1.

  When the processes in steps S244 and S245 in FIG. 3 when x = 3 are completed, it is determined in step S246 that there are no client-side devices that have not completed the processes in steps S243 to S245, and in step S248. A speech recognition result in which partial character strings of speech recognition results of speech 1 and speech 2 are left is output as a processed speech recognition result.

The search processing unit 25 of the cloud-side device 2 uses the processed speech recognition result included in the combination of at least one processed speech recognition result and the ID output by the speech recognition result processing unit 24 as a search query, and performs a predetermined process. The search in the search database or a predetermined information search site is executed, the search result is obtained, and the obtained search result is output to the search result transmitting unit 26 in a pair with the ID. In the above example, using the processed speech recognition result of the client device 1 ₁ speech recognition result processing unit 24 is output as a search query, executes the search at a given search database and predetermined information search site, Produced to obtain a search result corresponding to the speech recognition result, we obtained results in the client device 1 ₁ of the ID and the set, and outputs the search result transmission unit 26. The search processing is a well-known technology, and thus the detailed description is omitted.

The search result sending unit 26 of the cloud-side device 2 transmits a second transmission signal including the search result to the client-side device corresponding to the ID included in the combination of the search result and the ID output from the search processing unit 25. Send a signal. In the example above, it transmits the second transmission signal is a transmission signal including the search results to the client-side apparatus 1 _1. More precisely, the search result transmission unit 26, the second transmission signal is a transmission signal including the search results, to inform the client device 1 ₁ is sent to the network 3.

Search result receiving unit 14 ₁ of the client-side apparatus 1 ₁ receives the second transmission signal cloud side device 2 sends, retrieves the search results from the second transmission signal received, with respect to the screen display unit 15 ₁ Output. That is, the search result retrieval result receiving unit 14 ₁ outputs is a search result corresponding to the processed speech recognition result.

Screen display unit 15 ₁ of the client-side apparatus 1 ₁ displays a search result retrieval result receiving unit 14 ₁ is output to the client-side apparatus 1 ₁ of the screen. That is, the search result screen display section 15 ₁ is displayed is a search result corresponding to the processed speech recognition result.

  By using the speech recognition system according to the first operation example of the first embodiment, it is possible to solve the problem of the problem 1 and it is possible that the speech recognition result different from the speech recognition result desired by the speaker can be obtained. It is possible to reduce the possibility of obtaining a search result different from the search result desired by the speaker in the search, compared to the conventional case.

  The effect of using the speech recognition system according to the first operation example of the first embodiment will be described in more detail in the above specific case.

If the environment sound around the client-side device 1 ₁ such as a television or radio or announcement has occurred, the same television or radio or announcement with the surrounding of the client-side device 1 ₁ is also around the client-side device 1 ₂ Environmental noise such as is generated.

In this case, in the prior art, the speech recognition result of the client-side apparatus 1 _1, in addition to the speech recognition result of the first user's voice, including voice recognition result of the environmental sound, such as a television or radio or announcement It will If the client-side apparatus 1 ₁ but also exist prior art speech recognition processing after applying the noise suppression processing on the sound signal obtained, there is environmental sound that could not be completely suppressed by the noise suppressing process , the speech recognition result of the client-side apparatus 1 _1, will contain the result of speech recognition has not been suppressed ambient sound.

Also present client device 1 ₁ is acquired prior art to remove an acoustic signal that the client-side apparatus 1 ₂ from the acoustic signal acquired was. However, even with the same environmental sound, because different from the transfer characteristic of the transfer characteristic and the client-side device 2 ₁ to the client-side apparatus 1 _1, the acoustic signal and the client-side apparatus 1 by the client-side apparatus 1 ₁ is acquired ₂ is included as a different signal component in the acquired acoustic signal. Therefore, at the removal of the acoustic signals by the client-side apparatus 1 ₂ from sound signals client device 1 ₁ has acquired is acquired, it is impossible to remove all environmental sound from the acoustic signal by the client-side apparatus 1 ₁ is acquired. Therefore, if there is not fully removed environment sound, the speech recognition result of the client-side apparatus 1 _1, it will contain the result of speech recognition could not remove the environmental sound. Further, when the thus removed sound signal which the client-side apparatus 1 ₂ has obtained from the audio signal by the client-side apparatus 1 ₁ is acquired, sound corresponding to the first voice user utters the client-side apparatus 1 ₁ among the components of the signal, for components corresponding to the acoustic signal of the client-side apparatus 1 ₂ will be removed, the component of the acoustic signal corresponding to the first voice user utters the client-side apparatus 1 ₁ becomes a state distorted, there is a possibility that a problem that voice recognition is not performed correctly for the first voice the user has issued the client-side apparatus 1 ₁ occurs.

Even when the transmission characteristics of the client device transfer characteristic and the client-side device 2 ₁ to 1 ₁ are different, if the same environmental sound is present in relatively large volume, the client-side apparatus 1 ₁ in both the speech recognition result for the acquired speech recognition result for the acoustic signals and the client-side apparatus 1 ₂ acquired acoustic signals, portions of the television or radio and announcement is a voice recognition result of the vocal such substring is the same time It is included as a string. Therefore, according to the speech recognition system according to the first operation example of the first embodiment, the speech recognition result for the sound signal by the client-side apparatus 1 ₁ has acquired, to the speech recognition result for the acoustic signal other client-side device is obtained By removing partial strings included at substantially the same time, it is possible to remove speech recognition results of environmental sounds such as television, radio, and information broadcasting.

Meanwhile, the first voice the user has issued the client-side apparatus 1 _1, although the client-side apparatus 1 ₁ are included in the acquired acoustic signal, included in the audio signal by the client-side apparatus 1 ₂ acquires I can not. Therefore, from the speech recognition result for the sound signal by the client-side apparatus 1 ₁ is acquired, also by removing the partial character strings included in substantially the same time to the speech recognition result for the acoustic signal other client device has acquired, first The speech recognition results of the voices uttered by the users of are not removed.

  As described above, according to the speech recognition system according to the first operation example of the first embodiment, the possibility that the speech recognition result of the speech uttered by the utterer, which is the speech recognition result desired by the utterer, is suppressed low It is possible to reduce the possibility of including the speech recognition result of environmental sound such as a television, a radio, and a guidance broadcast, which is a speech recognition result different from the speech recognition result desired by the speaker, than in the past.

[[Second operation example of the first embodiment]]
As a second operation example, for one processing target client side device, a plurality of other client side devices predetermined at substantially the same time same partial character string as the processing target client side device (common partial character string) An example will be described in which, as a processed speech recognition result, one obtained by removing a common partial character string from the character string of the speech recognition result of the processing target client side device when the same time is present is obtained. The second operation example is different from the first operation example in the operation of the speech recognition result processing unit 24 of the cloud-side device 2. Hereinafter, only differences from the first operation example will be described.

  The voice recognition result processing unit 24 of the cloud-side device 2 includes a portion of the combination of at least one voice recognition result, time information, and ID stored in the voice recognition result holding unit 23 in the character string of the voice recognition result. For each character string, among the combinations of other speech recognition results, time information, and ID, there are a plurality of predetermined combinations (K, K is an integer of 2 or more) in which combinations of partial character strings and time match 2.) If there is a matching partial character string removed, it is regarded as a processed speech recognition result, and the processed speech recognition result and the ID are output as a set.

Next, an example of the speech recognition result and the processed speech recognition result in this example will be described with reference to FIG. The horizontal axis in FIG. 5 is time, and three above the arrow are the speech recognition results of the client side devices 1 1 _to 1 ₃ which are the input of the speech recognition result processing unit 24, and 1 below the arrow The _{first is} the processed speech recognition result of the client side device 11. Here, as a more specific case, N = 3 (the number of devices on the client side) and K = 2 (the allowable number of the number of devices whose partial character strings coincide at the same time), and the sound of television or radio is flowing The first user is in a place where information broadcasts such as train stations and department stores flow irregularly, and the same place as the first user such as a television, radio, and a guide broadcast flows with the second user The case where the third user is present will be described as an example.

The client-side apparatus 1 ₁ of the speech recognition result, the first and the partial string of the speech recognition result of speech 1 and utterance 2 is a speech the user has issued a client device 1 ₁ of the user, the client-side TV around the apparatus 1 ₁ contains a substring of the speech recognition result of the television audio 1 and the television audio 2 is a sound produced by. Moreover, the client-side apparatus 1 ₂ speech recognition result, a partial character string of the speech recognition result of the speech 3 and speech 4 is a utterance second user is a client-side apparatus 1 ₂ user uttered, substring of a speech recognition result of the client-side apparatus 1 ₂ of the television audio 1 is a audio television emitted in and around the television audio 2 are included. Further, the speech recognition result is the client-side apparatus 1 _3, the third user utterance and the partial string of the speech recognition result of speech 5 and the speech 2 is emitted in a client device 1 ₃ of the user, substring of client device 1 ₃ TV audio 1 is a audio television emitted around the and television audio 2 speech recognition result are included. Here, the partial character string of the speech recognition result of the speech 2 which is the speech emitted by the first user, and the partial character string of the speech recognition result of the speech 2 which is the speech emitted by the third user It is assumed that they are identical.

Client for the substring of the speech recognition result of the speech one of device 1 ₁ of the partial character strings included in the voice recognition result, the speech recognition result of the client-side apparatus 1 _second client-side device to the speech recognition result 1 ₃ because it is not included at the same time also it is not removed from the speech recognition result of the client-side apparatus 1 _1. The substring of the speech recognition result of the television audio 1 of the partial character strings included in the client-side apparatus 1 ₁ of the speech recognition result, the client-side apparatus 1 ₃ of the speech recognition in the client-side apparatus 1 ₂ speech recognition result to be included in the result to be the same time, i.e., because it contains at the same time also on the result of the speech recognition of the other two client devices, the number of matching becomes K greater than 3, the client-side apparatus 1 ₁ speech It is removed from the recognition result. The client device 1 ₁ of the substring of the speech recognition result of the speech 2 of the partial character strings included in the speech recognition result, the speech recognition result of the client-side apparatus 1 ₂ will not be included in the same time, the client-side apparatus 1 _Since the voice recognition result of ₃ is included at the same time, that is, the voice recognition result of another client side apparatus is also included at the same time, the number of matches is 2 and does not exceed K. not removed from the speech recognition result of the device 1 _1. The substring of the speech recognition result of the television audio 2 of the partial character strings included in the client-side apparatus 1 ₁ of the speech recognition result, the client-side apparatus 1 ₃ of the speech recognition in the client-side apparatus 1 ₂ speech recognition result since also included at the same time to the result, that is, since the speech recognition result of the other two client device included in the same time, be removed from the speech recognition result of the client-side apparatus 1 _1. Therefore, speech 1 and the client-side apparatus 1 ₁ for speech recognition result is a state in which the partial character string is left in the speech recognition result of the speech 2 is outputted as a processed speech recognition result.

  Note that this process may be performed on all other client-side devices, or may be performed on other partial (a plurality of) client-side devices. In this case, only the speech recognition result necessary for the processing of the speech recognition result processing unit 24 may be obtained in the previous stage. That is, the operations of the voice receiving unit 21, the voice recognition unit 22, and the voice recognition result holding unit 23 for obtaining a voice recognition result unnecessary for the processing of the voice recognition result processing unit 24 may be omitted.

  In the first operation example, when two users utter the same content at the same time by chance, the voice recognition result of the voice emitted by the user is removed. On the other hand, in the second operation example, the voice recognition result of the voice emitted by the user is not removed unless three or more people (K + 1 or more) utter the same content at the same time. It is extremely unlikely that the utterances of multiple users are the same content at the same time, as compared to the environmental sound such as television, radio, and guide broadcasting, which always have the same content at the same time, three people It is even less likely to be above. Therefore, according to the speech recognition system according to the second operation example of the first embodiment, the possibility that the speech recognition result of the speech emitted by the utterer, which is the speech recognition result desired by the utterer is lost may be higher than the first operation example. It is possible to reduce the possibility of including the speech recognition result of environmental sounds such as television, radio, and guidance broadcasting, which is a speech recognition result different from the speech recognition result desired by the speaker, than the conventional technology, while keeping it low.

[[Third operation example of the first embodiment]]
As a third operation example, only with respect to one processing target client-side device, among other client-side devices, only a client-side device in which the same partial character string as the processing target client-side device appears multiple times at the same time If the same partial character string (common partial character string) as the process target client device is present at the same time in another client device for the same target character string from the character recognition result of the process target client device An example will be described in which a partial character string removed is obtained as a processed speech recognition result. The third operation example is different from the first operation example in the operation of the speech recognition result processing unit 24 of the cloud device 2. Hereinafter, only differences from the first operation example will be described.

  The voice recognition result processing unit 24 of the cloud-side device 2 mixes other voice recognition results, time information, and ID with respect to a combination of at least one voice recognition result, time information, and ID stored in the voice recognition result holding unit 23. If there are a predetermined number (L pieces, L is an integer of 2 or more) in which the combination of the partial character string and the time match is a predetermined number in the group of, the one from which the matched partial character string is removed is As a processed speech recognition result, the processed speech recognition result and the ID are output as a set.

Next, with reference to FIG. 6, an example of the speech recognition result and the processed speech recognition result in this example will be described. The horizontal axis in FIG. 6 is a time, three at the top of the arrow is an input and is the client-side apparatus 1 ₁ to 1 ₃ each speech recognition result of the speech recognition result processing section 24, the bottom of the arrow 1 The _{first is} the processed speech recognition result of the client side device 11. Here, as a more specific case, it is N = 3 and L = 2 and the first user is at a place where the sound of a television or radio is flowing or the guidance broadcast of a station or department store is irregularly flowing. In the same place as the first user, the second user is in the place where the same television, radio and information broadcast are flowing, and the second user is in the place where the television, radio and information broadcast are different from the first user. The case where the third user is present will be described as an example.

The client-side apparatus 1 ₁ of the speech recognition result, the first and the partial string of the speech recognition result of speech 1 and utterance 2 is a speech the user has issued a client device 1 ₁ of the user, the client-side TV around the apparatus 1 ₁ contains a substring of the speech recognition result of the television audio 1 and the television audio 2 is a sound produced by. Moreover, the client-side apparatus 1 ₂ speech recognition result, a partial character string of the speech recognition result of the speech 3 and speech 4 is a utterance second user is a client-side apparatus 1 ₂ user uttered, substring of a speech recognition result of the client-side apparatus 1 ₂ of the television audio 1 is a audio television emitted in and around the television audio 2 are included. Further, the speech recognition result is the client-side apparatus 1 _3, the third user utterance and the partial string of the speech recognition result of speech 5 and the speech 2 is emitted in a client device 1 ₃ of the user, substring of a speech recognition result of the television audio 3 and video audio 4 is a speech television emitted around the client-side apparatus 1 ₃ are included. Here, the partial character string of the speech recognition result of the speech 2 which is the speech emitted by the first user, and the partial character string of the speech recognition result of the speech 2 which is the speech emitted by the third user It is assumed that they are identical.

The client-side apparatus 1 ₂ speech recognition results, as a partial character string is the same substring as a speech recognition result of the client-side apparatus 1 ₁ are included at the same time, the speech recognition result of the television audio 1 substring There are two substrings of and the substring of the speech recognition result of TV voice 2. Client device 1 _2, since the same sub-string and the client-side apparatus 1 ₁ is a client device in a plurality of times may appear at the same time, the object of removing processing substrings. The client-side apparatus 1 ₃ of the speech recognition results, as a partial character string is the same substring as a speech recognition result of the client-side apparatus 1 ₁ are included at the same time, the partial character string of the speech recognition result of the speech 2 is there. Client device _1-3, since the same sub-string and the client-side apparatus 1 ₁ is a client device without a plurality of times may appear at the same time, is not used for removing process of the substring. A portion for character client device 1 ₂ to be processed in the subject removing the column only in its client-side apparatus 1 ₂ speech recognition result and the client-side apparatus 1 ₁ speech recognition result, the same character string is the same time Search and get everything that is included in. That is, the partial character string of the speech recognition result of television sound 1 and the partial character string of the speech recognition result of television sound 2 are obtained. Then, all the partial character string is searched, i.e., removing sub-string of the speech recognition result of the partial strings and TV voice second speech recognition result of the television audio 1, from the speech recognition result of the client-side apparatus 1 ₁ things, namely, obtaining speech 1 and the client-side apparatus 1 ₁ for speech recognition result is a state in which the partial character string is left in the speech recognition result of the speech 2, the resulting processed speech recognition.

  Note that this process may be performed on all other client-side devices, or may be performed on some other client-side devices. In this case, only the speech recognition result necessary for the processing of the speech recognition result processing unit 24 may be obtained in the previous stage. That is, the operations of the voice receiving unit 21, the voice recognition unit 22, and the voice recognition result holding unit 23 for obtaining a voice recognition result unnecessary for the processing of the voice recognition result processing unit 24 may be omitted.

  In the first operation example, when two users utter the same content at the same time by chance, the voice recognition result of the voice emitted by the user is removed. On the other hand, in the third operation example, unless the two users utter the same content a plurality of times at the same time, the voice recognition result of the voice emitted by the user is removed. Absent. It is extremely unlikely that two users' utterances will be the same content at the same time, as compared to that the environmental sound such as a television, radio, and guide broadcast always have the same content at the same time, Even less likely to be. Therefore, according to the speech recognition system according to the third operation example of the first embodiment, the possibility that the speech recognition result of the speech uttered by the utterer, which is the speech recognition result desired by the utterer, may be lost is more than the first operation example. It is possible to reduce the possibility of including the speech recognition result of environmental sounds such as television, radio, and guidance broadcasting, which is a speech recognition result different from the speech recognition result desired by the speaker, than the conventional technology, while keeping it low.

[[The fourth operation example of the first embodiment]]
As a fourth operation example, an example using position information in addition to the time information of the first operation example will be described. Fourth operation example is different from the first operation example, the client-side apparatus ₁ 1 to 1 _N user information acquisition unit ₁₂ 1 to 12 _N, the speech recognition unit 22 of the cloud side device 2, the speech recognition result holding unit 23, This is the operation of the speech recognition result processing unit 24. Hereinafter, only differences from the first operation example will be described.

User information acquisition section 12 ₁ of the client-side apparatus 1 _1, the client-side apparatus 1 ₁ obtains location information and time information voice input unit 11 ₁ obtains an acoustic signal, the position information and the time information as the user information and outputs to the audio transmission unit 13 _1. Position information is, for example, information indicating an absolute position such as latitude and longitude, if the client-side device is a smart phone with a built-in GPS receiver, when the microphone is a voice input unit 11 ₁ obtains a sound signal The latitude and longitude measured by the GPS receiver may be used as the position information. Further, in the case of a smartphone having an auxiliary positioning function using a Wifi base station or a beacon, the latitude and longitude measured by the auxiliary positioning unit may be used as the position information. The position information is used by the voice recognition result processing unit 24 to specify whether or not the position at which the sound signal acquired by each of the plurality of client-side devices is emitted is in the vicinity. It may be information representing the relative positional relationship between client-side devices of For example, in the case of smart TV or STB, a region-specific code that indicates the same value in a mesh area of a certain latitude and longitude, such as a region code, a zip code, a beacon code received from a proximity beacon, or a geohash ID The ID may be used as information representing the relative positional relationship of the position information. Client device ₁ 2 to 1 _N user information acquisition unit ₁₂ 2 to 12 _N also operates in the same manner as the user information acquisition section 12 ₁ of the client-side apparatus _{1 1.}

  The voice recognition unit 22 of the cloud-side device 2 performs voice recognition processing on each of the sound signals output from the voice reception unit 21, and obtains a voice recognition result that is a character string corresponding to the voice included in the sound signal. A set of a speech recognition result, time information corresponding to the speech recognition result, position information corresponding to the speech recognition result, and an ID corresponding to the speech recognition result is output. The voice recognition process, the voice recognition result, the time information corresponding to the voice recognition result, and the ID corresponding to the voice recognition result are the same as those in the first operation example. The position information to be paired with the speech recognition result is the position information corresponding to the speech recognition result, that is, the user information input from the speech receiving unit 21 in combination with the acoustic signal from which the speech recognition result is obtained. Location information included in When there is a plurality of position information included in the user information input from the voice receiving unit 21 in combination with an acoustic signal from which the voice recognition result is obtained, for one voice recognition result, a plurality of position information One piece of position information representing position information is paired with the speech recognition result. The one position information representing a plurality of position information may be anything as long as the position where the sound signal corresponding to the speech recognition result can be substantially identified, for example, any one of a plurality of position information The position information may be an average value of latitudes included in a plurality of position information and an average value of longitudes included in the plurality of position information.

  The voice recognition result holding unit 23 of the cloud-side device 2 stores a set of the voice recognition result, the time information, the position information, and the ID output from the voice recognition unit 22. The contents stored in the voice recognition result storage unit 23 are processing for the voice recognition result processing unit 24 to determine whether or not there is a partial character string such as a word having the same position as the time, and a word having the same position as the time Is used for processing to obtain a processed speech recognition result by removing it from the speech recognition result when there is a partial character string of. Therefore, the stored contents held in the speech recognition result holding unit 23 may be deleted when the processing of the speech recognition result processing unit 24 using the stored contents is over and a predetermined time has elapsed. This is, for example, dozens of seconds in consideration of the time required for internal processing of the client side device, the time and error required for data transmission to the cloud side device, and the time length of each partial character string.

  The voice recognition result processing unit 24 of the cloud-side device 2 uses the characters of the voice recognition result for the combination of at least one ID-added voice recognition result, time information, position information, and ID stored in the voice recognition result holding unit 23. A combination of partial character string, time, and position among sets of other speech recognition results, time information, positional information, and ID for each of the partial character string, time, and position in the string If there is a match, the processed speech recognition result is obtained by removing the matched partial character string, and the processed speech recognition result and the ID are output as a set. Therefore, processed speech recognition results for at least one client-side device will be output. Note that the determination as to whether or not the positions coincide is not to determine whether the client-side devices are in exactly the same position, but the client-side devices may use the same television or radio sound, such as a guide broadcast, etc. Since it is determined whether or not the environmental sound may be acquired as an acoustic signal, it is determined whether or not the positions coincide with each other depending on whether they are within a predetermined distance range. Used as a judgment. That is, for at least one client-side device, if another client-side device at a near position at substantially the same time has the same partial character string (common partial character string) as the client-side device, What is obtained by removing the common partial character string from the character string of the speech recognition result of the client side device is obtained as a processed speech recognition result.

  Note that this process may be performed on all other client-side devices, or may be performed on at least one other client-side device. In this case, only the speech recognition result necessary for the processing of the speech recognition result processing unit 24 may be obtained in the previous stage. That is, the operations of the voice receiving unit 21, the voice recognition unit 22, and the voice recognition result holding unit 23 for obtaining a voice recognition result unnecessary for the processing of the voice recognition result processing unit 24 may be omitted.

  The processing flow of the speech recognition result determination unit 26 of the cloud device 2 described above is as shown in FIG. 7 differs from the processing flow of FIG. 3 in that step S241A is performed instead of step S241 in FIG. 3 and step S243A is performed instead of step S243 in FIG. 3 and step S244 in FIG. Step S244A is performed, and step S245A is performed instead of step S245 of FIG. 3.

Speech recognition result processing unit 24 first reads a set of the position information and the ID and the client device 1 ₁ of the speech recognition result and the time information from the speech recognition result holding unit 23 (step S241A). Next, the speech recognition result processing unit 24 sets an initial value x to 2 (step S242). Next, the speech recognition result processing unit 24 reads out a combination of the speech recognition result, time information, position information, and ID of the client side device _1x from the speech recognition result holding unit 23 (step S243A). Speech recognition result processing unit 24, then the position information and the ID and the speech recognition result and the time information of the set and the client-side apparatus 1 _x between the position information and the ID and the client device 1 ₁ of the speech recognition result and the time information It is searched whether or not there is a partial character string and its approximate time (for example, several seconds) and position match among the group of (step S 244 A). Next, when there is a partial character string that matches the time and position at step S244A, the speech recognition result processing unit 24 selects the partial character string and all partial strings that match the time and position. removed from the client-side apparatus _{1 1} of the speech recognition result string (step S245A). If it is determined in step S244A that there is no partial character string whose time and position match, the process proceeds to step S246. Next, the speech recognition result processing unit 24 determines whether there remains any client-side device not subject to the processes of step S243, step S244A, and step S245A (step S246). The voice recognition result processing unit 24 next replaces x with x + 1 if it is determined in step S246 that there remains a client-side device not subject to the processing in steps S243, S244A, and S245A (step S246). S247). Step S243 In step S246, step S244A, when it is determined that there are no remaining client device that is not subject to the process of step S245A is end processed speech recognition client device _{1 1} in step S245A Been results of the string in the ID and the set output as client-side apparatus 1 ₁ of processed speech recognition result string (step S248).

Note that steps S244A11 and S244A12 described in (1) of FIG. 8 may be performed instead of step S244A of FIG. 7. By performing the step S244A11 steps S244A12 in (1) described in FIG. 8 in place of step S244A of FIG. 7, only if the set of substrings and time step S244A2 match, the client-side apparatus ₁ 2 to 1 _N since each is well be carried out of the search in the vicinity of the client-side apparatus 1 _1, when matched substring is small, it is possible to reduce the amount of computation.

In addition, step S244A21 and step S244A22 described in (2) of FIG. 8 may be performed instead of step S244A of FIG. 7. By performing the step S244A21 steps S244A22 (2) described in FIG. 8 in place of step S244A of FIG. 7, step only if the respective client-side apparatus ₁ 2 to 1 _N is close to the client device _{1 1} S244A2 Since it is sufficient to search whether the partial character string and the time pair match, it is possible to reduce the amount of operation processing when there are few client devices in the vicinity.

  By using the speech recognition system according to the fourth operation example of the first embodiment, it is possible to solve the problem of the problem 2 and it is possible that the speech recognition result different from the speech recognition result desired by the speaker can be obtained. It is possible to reduce the possibility of obtaining a search result different from the search result desired by the speaker in the search, compared to the conventional case.

[[Fifth operation example of the first embodiment]]
Also in the fourth operation example, the same change as the change of the operation from the first operation example to the second operation example can be made. This will be described as a fifth operation example. That is, in the fifth operation example, the same partial character string as the process target client side apparatus (a plurality of predetermined other client side apparatuses located near at substantially the same time) for one process target client side apparatus This is an example in which, when there is a common partial character string, a character string of the speech recognition result of the processing target client side device from which the common partial character string is removed is obtained as a processed speech recognition result. The fifth operation example is different from the fourth operation example in the operation of the speech recognition result processing unit 24 of the cloud device 2. Hereinafter, only differences from the fourth operation example will be described.

  The voice recognition result processing unit 24 of the cloud-side device 2 sets the combination of at least one voice recognition result, time information, position information, and ID stored in the voice recognition result holding unit 23 to a character string of the voice recognition result. For each of the included partial character strings, there are a plurality of (K, predetermined ones in which combinations of partial character strings and time among the combinations of other speech recognition results, time information, position information, and ID match When K is an integer greater than or equal to 2), the one from which the matched partial character string has been removed is taken as a processed voice recognition result, and the processed voice recognition result and the ID are output in pairs.

  In the fourth operation example, when two users utter the same content at the same time near each other at the same time, the speech recognition result of the speech emitted by the user is removed. On the other hand, in the fifth operation example, as long as three or more people (K + 1 or more) do not utter the same content at the same position at the same time, the result of the speech recognition of the speech uttered by the user is removed Absent. It is extremely unlikely that the utterances of multiple users will be the same content at the same time in the vicinity, as compared to the environmental sound such as television, radio, and guide broadcasting always having the same content at the same time. Is less likely to be three or more. Therefore, according to the speech recognition system according to the fifth operation example of the first embodiment, the possibility that the speech recognition result of the speech emitted by the utterer, which is the speech recognition result desired by the utterer, may be lost is better than the second operation example. It is possible to reduce the possibility of including the speech recognition result of environmental sounds such as television, radio, and guidance broadcasting, which is a speech recognition result different from the speech recognition result desired by the speaker, than the conventional technology, while keeping it low.

[[Sixth operation example of the first embodiment]]
The first to third operation examples not using position information and the fourth to fifth operation examples using position information may be combined to operate, and an example thereof will be described as a sixth operation example. In the sixth operation example, with regard to one processing target client side device, the client side device determined to be near the processing target client side device from the position information, and being near the processing target client side device Regarding a plurality of partial character strings in the character string of the speech recognition result although not determined, with respect to the client side device in which the same partial character string as the character string of the speech recognition result of the processing target client side device appears at the same time In this example, a character string of the speech recognition result of the processing target client side device from which a common partial character string is removed is obtained as a processed speech recognition result. The sixth operation example is different from the fourth operation example in the operation of the speech recognition result processing unit 24 of the cloud device 2. Hereinafter, the operation of the speech recognition result processing unit 24 different from the fourth operation example will be described with reference to FIG.

Speech recognition result processing unit 24 first reads a set of the position information and the ID and the client device 1 ₁ of the speech recognition result and the time information from the speech recognition result holding unit 23 (step S241). Next, the speech recognition result processing unit 24 sets an initial value x to 2 (step S242). Next, the speech recognition result processing unit 24 reads a combination of the speech recognition result of the client side device _1x , time information, position information, and an ID from the speech recognition result holding unit 23 (step S243). Speech recognition result processing unit 24, then the position information and the ID and the speech recognition result and the time information of the set and the client-side apparatus 1 _x between the position information and the ID and the client device 1 ₁ of the speech recognition result and the time information It is searched whether or not there are a plurality of partial character strings that match the time in the group of (step S244B1). Next, when there are a plurality of partial character strings that match the time at step S 244 B 1, the speech recognition result processing unit 24 sets all the partial strings that match the partial character and the time to the client side. device _{1 1} removed from the string of the speech recognition result (step S245B1). In step S 244 B 1, if there are not a plurality of partial character strings that match the time, that is, if there is only one partial character string that matches the time and a partial character string that matches the time If not, the process proceeds to step S244B2. Speech recognition result processing unit 24, then the position information and the ID and the speech recognition result and the time information of the set and the client-side apparatus 1 _x between the position information and the ID and the client device 1 ₁ of the speech recognition result and the time information It is searched whether or not there is a partial character string and one whose time and position coincide with each other (step S244B2). Next, when there is a partial character string that matches the time and position at step S244B2, the speech recognition result processing unit 24 selects the partial character string and all partial strings that match the time and position. removed from the client-side apparatus _{1 1} of the speech recognition result string (step S245B2). If it is determined in step S244B2 that there is no partial character string and its time and position match, the process proceeds to step S246. Next, the voice recognition result processing unit 24 determines whether any client-side device not to be processed remains in any of step S243, step 244B1, step 244B2, step 245B1 and step 245B2 (step S246). Next, if it is determined in step S246 that any client-side device not to be processed remains in any of step S243, step 244B1, step 244B2, step 245B1 and step 245B2 in step S246. , X to x + 1 (step S247). If it is determined in step S246 that there is no client-side device not to be processed in any of step S243, step 244B1, step 244B2, step 245B1 and step 245B2, the process in step S245B1 or S245B2 last performed already client device 1 ₁ of the ID paired outputs a string of the speech recognition result as a client-side device 1 ₁ of the processed speech recognition result string (step S248).

  By using the voice recognition system according to the sixth operation example of the first embodiment, a plurality of client-side devices are not in the vicinity, but the same television or radio is flowing, and a plurality of client-side devices are in the vicinity It is possible to remove character strings of speech recognition results of environmental sounds when the same television and radio are flowing, and speech recognition results different from the speech recognition results desired by the speaker can be obtained. It is possible to reduce the gender than in the past and to reduce the possibility of obtaining a search result different from the search result desired by the speaker in the search than in the past.

Second Embodiment
Next, as a second embodiment of the present invention, a form in which a portion common to the speech recognition result of the sound signal of the public broadcast is removed from the speech recognition result of the sound signal including the speech of the speaker input to the client side device explain. FIG. 10 is a block diagram showing the configuration of the speech recognition system in the second embodiment. Among the components of FIG. 10, the same components as those of FIG. 1 are denoted by the same reference numerals. The code 100 is a speech recognition system, the codes 1 ₁ to 1 _N are one or more (N, N is an integer of 1 or more) client side apparatuses, and the code 2 is a cloud side apparatus. Reference numerals 5 _{1 to} 5 _M denote public broadcast stations of one or more stations (M stations, M is an integer of 1 or more). The client side devices 1 ₁ to 1 _N are devices used by the user, and are exemplified in the description of the first embodiment, for example. The public broadcast stations 5 _{1 to} 5 _M exist outside the speech recognition system 100. The cloud side device 2 is connected to the client side devices 1 ₁ to 1 _N via the network 3. The network 3 is for enabling the client side devices 1 ₁ to 1 _N of the voice recognition system 100 and the cloud side device 2 to transmit and receive information according to the connection protocol of the Internet, and is, for example, the Internet. The client side devices 1 ₁ to 1 _N and the cloud side device 2 can transmit and receive information according to the connection protocol of the Internet.
Cloud-side device 2 is connected to the public broadcasting station ₅ 1 to 5 _M via the network 4. Network 4, cloud-side device 2 and the public broadcasting station 5 ₁ to 5 M of the speech recognition system _100, a and Internet connection protocol is intended to allow transmission and reception of information in accordance with a dedicated protocol for video relay For example, closed-type dedicated line Internet.

In the configuration of FIG. 10, the public broadcast stations 5 _{1 to} 5 _M and the cloud side device 2 are connected via the network 4 according to the connection protocol of the Internet, and the cloud side device 2 is a public broadcast of the public broadcast stations 5 _{1 to} 5 _M It is made possible to receive a signal. However, the cloud-side device 2 may be provided with a receiver (not shown) so that the public broadcast signal can be received without going through the network 4. Further, for example, since the program configuration and the broadcast time of public broadcasting broadcast by each area, such as Tokyo, Osaka, Nagoya, Fukuyama, etc., the acoustic signal will also differ by area. Since installing many cloud side devices 2 all over the country is expensive, receivers of public broadcast signals are installed all over the country, processed signals and recognition results are network 3 and network 4 via cloud side devices 2 It may be sent to

The public broadcast stations 5 _{1 to} 5 _M are, for example, all or main public broadcast stations in the area where the client side device may be located, for example, the residence area or movement of the user using the client side device These include satellites, terrestrial and major IP broadcasts / streaming / CATV / wired broadcasts in areas including the area. The cloud side device 2 is connected to necessary facilities such as the network 3 and the network 4 and a receiving facility and a receiving device so that all desired public broadcast stations can be received.

Since all the configurations included at least in the client side devices 1 ₁ to 1 _N are the same, in the following, a part configured by the client side device ₁₁ and the cloud side device 2 in the speech recognition system 100 of the second embodiment This will be described using FIG. 11 which is a detailed block diagram of FIG. Among the components shown in FIG. 11, the components assigned the same reference numerals as those in FIG. 2 perform the same operations as those in FIG.

Client device _{1 1} includes an audio input unit 11 _1, the user information acquiring unit 12 _1, the audio output unit 13 _1, the search result receiving unit 14 _1, configured to include at least the screen display unit 15 _1.

  The cloud side device 2 includes a voice receiving unit 21, a voice recognition unit 22, a broadcast receiving unit 41, a broadcast voice recognition unit 42, a voice recognition result holding unit 43, a voice recognition result processing unit 44, a search processing unit 25, a search result sending unit It comprises at least 26. The voice receiving unit 21, voice recognition unit 22, search processing unit 25 and search result sending unit 26 of the cloud side device 2 are the voice receiving unit 21, voice recognition unit 22, search processing portion of the cloud side device 2 according to the first embodiment. The same operation is performed with the search result sending unit 26 and the search result sending unit 26.

  Next, the operation of the speech recognition system according to the second embodiment will be described.

[[First operation example of the second embodiment]]
As a first operation example, each user of the 1~N is not using the client-side apparatus 1 ₁ to 1 _N, the sentence to be obtained search results first user to the client-side apparatus 1 ₁ the speaks, an example of operation of displaying a search result corresponding to the utterance to the client-side apparatus 1 ₁ of the screen display unit 15 _1. Here, as a more specific case, there are first user to the location where only broadcast public broadcasting station 5 ₁ in the area that can receive broadcast of the two public broadcasting stations 5 ₁ to 5 ₂ flows The case will be described as an example.

Client device 1 ₁ of the speech input unit 11 ₁ obtains an acoustic signal emitted around the client-side apparatus 1 _1, and outputs the acquired audio signal to the audio transmission unit 13 _1. If the first user has uttered a sentence to be obtained search results to the client-side apparatus 1 _1, obtains and outputs a sound signal including a voice first user uttered. If the client-side apparatus 1 ₁ of environmental sounds such as a television or radio or announcement around has occurred, and outputs the acquired audio signal including the ambient sound. Therefore, if the above specific case, and audio first user has uttered, and sound public broadcasting station 5 ₁ broadcast, obtains and outputs the composed audio signal by.

User information acquisition section 12 ₁ of the client-side apparatus 1 ₁ obtains the time information voice input unit 11 ₁ of the client-side apparatus 1 ₁ acquires the audio signal, which can specify the time information and the client-side apparatus 1 ₁ identification information (hereinafter, referred to as "ID") to the audio transmission unit 13 ₁ and the user information. The time information, for example, absolute time, for example, if the client-side apparatus 1 ₁ is a smart phone with a built-in GPS receiver, the GPS receiver when the smartphone microphone is a voice input unit 11 ₁ obtains a sound signal The absolute time received by the unit may be used as time information. Further, for example, time information may be received from a base station of a mobile network or a communication server, or may be time information of a local clock held by the OS.

Voice output unit 13 ₁ of the client-side apparatus 1 ₁ sends a transmission signal including the user information acoustic signal and the user information acquisition section 12 ₁ speech input unit 11 ₁ is output is output to the cloud side device 2 . More precisely, the audio output unit 13 _1, the transmission signal including the user information acoustic signal and the user information acquisition section 12 ₁ speech input unit 11 ₁ is output is output, to convey to the cloud side device 2, the network Send to 3 Transmission of the transmission signal is performed, for example, at predetermined time intervals such as 10 ms. The audio output unit 13 ₁ obtains a code string acoustic signal is encoded by a predetermined encoding method, it may be sent a transmission signal including a resultant code sequence and the user information. The audio output unit 13 ₁ performs such feature extraction, which is part of the processing of the speech recognition process on the acoustic signals, sends a transmission signal including a feature amount and the user information obtained by the process May be There are many well-known techniques and well-known techniques in the technique of sending transmission signals to another device via a network and the technique of performing speech recognition processing in a distributed manner between the client-side device and the cloud-side device. I omit explanation.

Client device ₁ 2 to 1 _N audio input unit ₁₁ 2 to 11 _N, the user information acquiring unit ₁₂ 2 to 12 _N and the voice sending section ₁₃ 2 to 13 _N be respectively client device _{1 1} of the speech input unit 11 _1, the user information acquiring unit 12 ₁ and the audio output unit 13 ₁ and the same operation. In addition, when it is not necessary to obtain a search result corresponding to the utterance of any of the second to N users, the client side apparatus of the user who does not need to obtain the search result may not be provided, and the search result is Even if the client-side device of the user who does not need to obtain is provided, the voice input unit, the user information acquisition unit and the voice sending unit of the client-side device need not be operated.

The voice receiving unit 21 of the cloud side device 2 receives the transmission signals respectively sent by the voice sending portions 13 _{1 to} 13 _N of the client side devices 1 ₁ to 1 _N , and generates an acoustic signal and a user from the received transmission signals. Extract and output a pair with information. Reception of the transmission signal is performed, for example, at predetermined time intervals such as 10 ms. When the voice transmitting units 13 _{1 to} 13 _N encode the acoustic signal according to a predetermined coding method to obtain a code string and transmit the obtained code string, the voice receiving unit 21 of the cloud-side device 2 A sound signal may be obtained by decoding a code string included in the received transmission signal by a decoding method corresponding to a predetermined coding method, and a set of the obtained sound signal and user information may be output. In addition, the voice transmitting units 13 _{1 to} 13 _N perform feature extraction, which is a part of processing of voice recognition processing, on the sound signal, and transmit signals including the feature and the user information obtained by the processing. In the case of sending, it is sufficient to take out not the acoustic signal but the feature amount from the transmission signal, and output the set of the extracted feature amount and the user information. In addition, when it is not necessary to obtain a search result corresponding to the utterance of any of the second to N users, it is not necessary to receive an acoustic signal of a user who does not need to obtain a search result. Even if a user's acoustic signal that does not need to be obtained is received, it is not necessary to output the set of the acoustic signal and the user information. In addition, it is not necessary to obtain search results corresponding to the utterances of all the second to N users, and when not outputting a set of acoustic signals other than the first user and user information, the client side device 1 _It may be output without including the ID in ₁ user information.

The voice recognition unit 22 of the cloud-side device 2 performs voice recognition processing on each of the sound signals output from the voice reception unit 21, and obtains a voice recognition result that is a character string corresponding to the voice included in the sound signal. And outputting a set of a speech recognition result, time information corresponding to the speech recognition result, and an ID corresponding to the speech recognition result. If there is no time information or an inappropriate value, the server may manage the data based on the approximate time information received without using the received time information. When not the output of the other first user, without including the ID of the client-side apparatus 1 _1, and the speech recognition result, and outputs the set by the time information corresponding to the speech recognition result It is also good.

  Speech recognition processing is performed for each predetermined set of acoustic signals. For example, the voice recognition unit 22 sequentially stores the sound signal output from the voice reception unit 21 in a storage unit (not shown) in the voice recognition unit 22, and performs voice period detection on the stored voice signal. The voice recognition process is performed on the voice signal clusters in each utterance section to obtain a voice recognition result as a character string for the voice signal clusters in each utterance section. Also, for example, even if the speech recognition unit 22 performs speech recognition processing on a group of sound signals in a plurality of speech segments, and obtains a speech recognition result that is a character string to a group of sound signals in a plurality of speech segments. Good.

Therefore, if the above specific case, the speech recognition unit 22, as a speech recognition result to the client-side apparatus 1 ₁ of the acoustic signal, the first user utterance voice and sound of the public broadcasting station 5 ₁ for broadcasting To obtain and output a character string composed of the speech recognition result with the speech contained in and.

  As in the first embodiment, a known speech recognition technology may be used for the speech recognition process. When the voice receiving unit 21 outputs a feature amount instead of the sound signal, the voice recognition unit 22 may perform voice recognition processing using the feature amount.

  Since the speech recognition process is performed for each predetermined set of acoustic signals, the character string of the speech recognition result corresponds to the predetermined set of acoustic signals. Therefore, the speech recognition unit 22 obtains representative time from time information included in a plurality of user information corresponding to a predetermined group of sound signals targeted for speech recognition processing, for example, and the time information representing the representative time is Pair with speech recognition results. The representative time may be any one that represents the time at which the sound signal corresponding to the character string of the speech recognition result is emitted. For example, the time at the beginning of the speech recognition result character string may be used as the representative time. Also, there may be a plurality of representative times in one speech recognition result. For example, for each partial character string such as a word included in the speech recognition result, the time at the beginning of the word or the like may be used as the representative time.

  The ID to be paired with the speech recognition result is included in the user information input from the speech receiving unit 21 in combination with the ID corresponding to the speech recognition result, that is, the acoustic signal from which the speech recognition result is obtained. ID.

Broadcast receiving unit 41 of the cloud side device 2, the time information public broadcasting station 5 ₁ to 5 _M has received a public broadcast signal transmitted respectively correspond to the acoustic signal and the acoustic signal from each of the public broadcasting signals received Take out the pair with and output. At that time, identification information (hereinafter referred to as "broadcasting station ID") that can specify a public broadcasting station may be output as a set of acoustic signals and time information. The reception of the public broadcast signal is performed, for example, at predetermined time intervals such as 10 ms. Public broadcasting station 5 ₁ to 5 _N can obtain a code string by encoding with a predetermined encoding method the acoustic signal, when sending the obtained code string, the broadcast receiving unit 41 of the cloud side device 2, A sound signal is obtained by decoding a code string included in the received public broadcast signal by a decoding method corresponding to a predetermined coding method, and a set of the obtained sound signal and time information corresponding to the sound signal is obtained. You can output it. The broadcast reception unit 41 is provided with a clock (not shown) so as to be able to output an absolute time, and when the public broadcast signal is an analog broadcast and time information can not be extracted from the public broadcast signal, etc. The absolute time obtained from the clock provided at 41 may be output from the public broadcast signal in combination with the acoustic signal. As for the broadcast reception unit 41, for example, since the sound signal group of public broadcasts varies depending on the area, such as Tokyo, Osaka, Nagoya, Fukuyama, etc., the broadcast reception unit 41 is not located in a region different from the cloud side device 2. , And may be connected to the broadcast speech recognition unit 42 via a network. Also, when a direct signal can be obtained from a public broadcasting station via a network, the acoustic signal of the public broadcast obtained thereby may be used directly for input.

The broadcast speech recognition unit 42 of the cloud-side device 2 performs speech recognition processing on each of the acoustic signals output from the broadcast reception unit 41, and obtains a speech recognition result that is a character string corresponding to the speech included in the acoustic signal. A set of the speech recognition result and time information corresponding to the speech recognition result is output. At this time, the broadcast station ID may be output as a set of acoustic signal and time information. The broadcast reception unit 41 and the broadcast speech recognition unit 42 may be configured to connect the output thereof to the device 43 via a network in a region different from the cloud-side device 2.
Voice recognition processing is performed for each predetermined set of acoustic signals, and time information and an ID are added to the character string of the obtained voice recognition result, and voice recognition technology and the like used for voice recognition processing are voice recognition. As it is the same as part 22, detailed description is omitted.

  The voice recognition result holding unit 43 of the cloud-side device 2 sets the combination of the voice recognition result, the time information, and the ID output from the voice recognition unit 22 and the voice recognition result and time information output from the broadcast voice recognition unit 42. And remember. When the broadcast speech recognition unit 42 outputs a combination of speech recognition result, time information and broadcasting station ID, the speech recognition result is substituted for the combination of speech recognition result and time information output from the broadcast speech recognition unit 42. And stores time information and a station ID. The contents stored in the voice recognition result storage unit 43 are processes of the voice recognition result processing unit 44 determining whether or not there is a partial character string such as a word having a common time, and a partial character string such as a word having a common time Is used for processing to obtain a processed speech recognition result by removing it from the speech recognition result when there is an error. Therefore, in the voice recognition result holding unit 43, a combination of the voice recognition result outputted by the voice recognition unit 22, time information and ID, the voice recognition result outputted by the broadcast voice recognition unit 42, time information and the broadcasting station ID The set is stored only for the time required for the processing of the speech recognition result processing unit 44. Further, the stored contents held in the speech recognition result holding unit 43 may be deleted when the processing of the speech recognition result processing unit 44 using the stored contents is completed.

  The voice recognition result processing unit 44 of the cloud side device 2 stores the combination of the voice recognition result of at least one client side device stored in the voice recognition result holding unit 43, the time information, and the ID in the voice recognition result holding unit 43. If there is a match between the partial character string and the time set among the stored speech recognition results of each public broadcast and the time information, the matched partial character string is removed and processed As a speech recognition result, the processed speech recognition result and the ID are output as a set. Therefore, processed speech recognition results for at least one client-side device will be output. It should be noted that the determination as to whether or not the times coincide is determined to be the same time in consideration of an error in absolute time in the client side device and the public broadcasting station or the cloud side device or an error in time in speech recognition processing. You may That is, for at least one client-side device, if there is a partial character string (a common partial character string) in the public broadcast at approximately the same time in any public broadcast, the client-side device What removed the common substring from the character string of the speech recognition result is obtained as a processed speech recognition result.

  Note that this process may be performed on all public broadcast stations in an area where the client-side device may be located, or may be performed on at least one public broadcast. In this case, only the speech recognition result required by the processing of the speech recognition result processing unit 44 may be obtained in the previous stage. That is, the operations of the broadcast reception unit 41, the broadcast speech recognition unit 42, and the speech recognition result holding unit 43 for obtaining speech recognition results unnecessary for the processing of the speech recognition result processing unit 44 may be omitted.

Here, in the above example of M = 2, it will be described with reference to FIGS. 12 and 13 for example at least some one client device is a client-side apparatus 1 _1. FIG. 12 is a diagram for explaining the processing flow of the speech recognition result processing unit 44 in this operation example, and FIG. 13 is a diagram for explaining an example of the speech recognition result and the processed speech recognition result in this example. Example of FIG. 12, as the target speech recognition results of all public broadcasting station, in order from the speech recognition result of the public broadcasting station 5 _1, the set of the speech recognition result of the client-side apparatus 1 ₁ and the partial strings and time It is searched whether there is a match, and if there is a match between the partial character string and the time set, the partial character string matching the partial character string and the time set is used as the client-side device 1 It is an example which removes the said common partial character string from the character string of the speech recognition result of _1. FIG.

Speech recognition result processing unit 44 first reads a set of the speech recognition result of the client-side apparatus 1 ₁ and the time information and the ID from the speech recognition result holding unit 43 (step S441). Next, the speech recognition result processing unit 44 sets an initial value y to 1 (step S442). Next, the speech recognition result processing unit 44 reads out the combination of the speech recognition result of the public broadcasting station 5 _y and the time information from the speech recognition result holding unit 43 (step S443). Speech recognition result processing unit 44, then, in a set of client-side apparatus 1 ₁ for speech recognition result and the time information and the ID as a set and public broadcasting stations 5 _y speech recognition result and the time information of the partial strings It is searched whether there is a match between the time and the time (step S444). Next, when there is a partial character string that matches the time at step S444, the speech recognition result processing unit 44 uses the client-side device 1 for all the partial strings that match the partial character string and the time. _{It is} removed from the character string of the speech recognition result of ₁ (step S445). If there is no match between the partial character string and its time in step S444, the process proceeds to step S446. Next, the speech recognition result processing unit 44 determines whether there are any public broadcast stations not subject to the processes of steps S443 to S445 (step S446). Next, the voice recognition result processing unit 44 replaces y with y + 1 if it is determined in step S446 that there remains a public broadcasting station not subjected to the processing in steps S443 to S445 (step S447). Step If it is determined that there are no remaining steps S443~ public broadcasting station that is not subject to processing in step S445 in S446, the last in step S445 the processed client device 1 ₁ of the speech recognition result of characters column in the ID paired as client-side apparatus 1 ₁ of processed speech recognition result string output (step S448).

Next, an example of the speech recognition result and the processed speech recognition result in this example will be described with reference to FIG. The horizontal axis of FIG. 13 is a time, three each of the speech recognition of the speech recognition result the client device is an input of the processing unit 44 1 ₁ and public broadcasting stations 5 ₁ and public broadcasting stations 5 ₂ above the arrow the result, one below the arrow is processed speech recognition result of the client-side apparatus 1 _1. The client-side apparatus 1 ₁ of the speech recognition result, the first and the partial string of the speech recognition result of speech 1 and utterance 2 is a speech the user has issued a client device 1 ₁ of the user, the client-side TV around the apparatus 1 ₁ contains a substring of the speech recognition result of the television audio 1 and the television audio 2 is a sound produced by. Further, the speech recognition result is a public broadcasting station 5 _1, public broadcasting station 5 ₁ contains a substring of the speech recognition result of the television audio 1 and the television audio 2 is a sound included in the sound signal broadcast . Further, the speech recognition result is a public broadcasting station 5 _2, contains a substring of the speech recognition result of the television audio 3 and video audio 4 is a sound public broadcasting station 5 ₂ is included in the acoustic signal broadcast .

First, the process of step S444 and step S445 of FIG. 12 when y = 1 will be described. The client device 1 ₁ of the substring of the speech recognition result of the television audio 1 and the television audio 2 of the partial character strings included in the speech recognition result, also includes at the same time the speech recognition result of the public broadcasting station 5 ₁ It is therefore removed from the speech recognition result of the client-side apparatus 1 _1. The client device 1 ₁ of the substring of the speech recognition result of speech 1 and utterance 2 of the partial character strings included in the speech recognition result, not included at the same time the speech recognition result of the public broadcasting station 5 ₁ Therefore, it not removed from the speech recognition result of the client-side apparatus 1 _1. In other words, a state in which a substring of the speech recognition result of speech 1 and utterance 2 is left as a partial character string included in the speech recognition result of the client-side apparatus 1 ₁ proceeds to the process in the case of y = 2.

Next, the process of step S444 and step S445 of FIG. 12 when y = 2 will be described. The substring of the utterance 1 and utterance 2 of the speech recognition result of the partial character strings included in the client-side apparatus 1 ₁ for speech recognition result is not included at the same time the speech recognition result of the public broadcasting station 5 ₂ Therefore, it not removed from the speech recognition result of the client-side apparatus 1 _1. In other words, a state in which a substring of the speech recognition result of speech 1 and utterance 2 is left as a partial character string included in the speech recognition result of the client-side apparatus 1 _1.

  When the processes in steps S444 and S445 in FIG. 12 when y = 2 are finished, it is determined in step S446 that there are no public broadcast stations that have not completed the processes in steps S443 to S445, and in step S448. A speech recognition result in which partial character strings of speech recognition results of speech 1 and speech 2 are left is output as a processed speech recognition result.

  By using the speech recognition system according to the first operation example of the second embodiment, it is possible to solve the problem of the problem 1 and it is possible that the speech recognition result different from the speech recognition result desired by the speaker can be obtained. It is possible to reduce the possibility of obtaining a search result different from the search result desired by the speaker in the search, compared to the conventional case.

[[Second operation example of the second embodiment]]
As a second operation example, with regard to one processing target client side device, among public broadcasting stations, targeting only the public broadcasting stations in which the same partial character string as the processing target client side device appears multiple times at the same time When the same partial character string (common partial character string) as the processing target client device is present at the same time in the public broadcasting station, the common partial character string is removed from the character string of the speech recognition result of the processing target client device An example will be described in which one is obtained as a processed speech recognition result. The second operation example is different from the first operation example in the operation of the speech recognition result processing unit 44 of the cloud device 2. Hereinafter, only differences from the first operation example will be described.

  The voice recognition result processing unit 44 of the cloud side device 2 stores the combination of the voice recognition result of at least one client side device stored in the voice recognition result holding unit 43, the time information, and the ID in the voice recognition result holding unit 43. The voice recognition result of the client-side device for only the public broadcast in which there are a plurality of sets of partial character string and time match among the stored speech recognition result of public broadcast and time information A character string obtained by removing a partial character string in which a combination of partial character string and time matches the speech recognition result of the public broadcast from the character string is taken as a processed speech recognition result, and the processed speech recognition result and ID are paired. Output. As in the first operation example, when determining whether the times match, the error of the absolute time in the client device and the public broadcasting station or the cloud device, the error of the time in the speech recognition process, etc. are taken into consideration. It may be determined that it is the same time. That is, in the second operation example, when there is a plurality of partial character strings (common partial character strings) the same as the client-side device in any public broadcast at substantially the same time for at least one client-side device In the character string of the speech recognition result of the client-side device, the common speech of the public broadcast having a plurality of common partial character strings is removed as a processed speech recognition result.

  In the first operation example, when the user accidentally utters the same content at the same time of the television or radio that is not present as environmental sound around the user, the speech recognition result of the voice emitted by the user is It will be removed. On the other hand, in the second operation example, as long as the user does not utter the same content multiple times at the same time on the television or radio that is not present as environmental sound around the user, There is no need to remove the speech recognition result. The possibility of the user accidentally uttering the same content at the same time on a television or radio that is not present as environmental sound around the user is extremely low, and the possibility of multiple occurrences is even lower. Therefore, according to the speech recognition system according to the second operation example of the second embodiment, the possibility that the speech recognition result of the speech emitted by the utterer, which is the speech recognition result desired by the utterer, may be lost is more than the first operation example. It is possible to reduce the possibility of including the speech recognition result of environmental sounds such as television, radio, and guidance broadcasting, which is a speech recognition result different from the speech recognition result desired by the speaker, than the conventional technology, while keeping it low.

[[Third operation example of the second embodiment]]
As a third operation example, an example using position information in addition to the time information of the first operation example will be described.
The third operation example differs from the first operation example, the client-side apparatus ₁ 1 to 1 _N user information acquisition unit ₁₂ 1 to 12 _N, the speech recognition unit 22 of the cloud side device 2, the broadcast receiving unit 41, broadcast audio The operations of the recognition unit 42, the speech recognition result holding unit 43, and the speech recognition result processing unit 44. Hereinafter, only differences from the first operation example will be described.

User information acquisition section 12 ₁ of the client-side apparatus 1 ₁ of the third operation example, the client-side apparatus 1 ₁ obtains location information and time information voice input unit 11 ₁ obtains an acoustic signal, the position and the time information and outputs to the audio transmission unit 13 ₁ of the information as the user information. Position information is, for example, information indicating an absolute position such as latitude and longitude, if the client-side device is a smart phone with a built-in GPS receiver, GPS when the microphone is a voice input unit 11 ₁ obtains a sound signal The latitude and longitude determined by the receiving unit may be used as the position information. In the case of a smartphone having an auxiliary positioning function using a Wifi base station or a beacon, the latitude and longitude measured by the auxiliary positioning unit may be used as the position information. The position information may be information representing the relative positional relationship between a plurality of client-side devices. For example, in the case of smart TV or STB, a region-specific code that indicates the same value in a mesh area of a certain latitude and longitude, such as a region code, a zip code, a beacon code received from a proximity beacon, or a geohash ID The ID may be used as information representing the relative positional relationship of the position information. Client device ₁ 2 to 1 _N user information acquisition unit ₁₂ 2 to 12 _N also operates in the same manner as the user information acquisition section 12 ₁ of the client-side apparatus _{1 1.}

  The voice recognition unit 22 of the cloud-side device 2 of the third operation example performs voice recognition processing on each of the acoustic signals output from the voice reception unit 21 and generates a voice that is a character string corresponding to the voice included in the acoustic signal. The recognition result is obtained, and a combination of the speech recognition result, time information corresponding to the speech recognition result, position information corresponding to the speech recognition result, and an ID corresponding to the speech recognition result is output. The voice recognition process, the voice recognition result, the time information corresponding to the voice recognition result, and the ID corresponding to the voice recognition result are the same as those in the first operation example. The position information to be paired with the speech recognition result is the position information corresponding to the speech recognition result, that is, the user information input from the speech receiving unit 21 in combination with the acoustic signal from which the speech recognition result is obtained. Location information included in When there is a plurality of position information included in the user information input from the voice receiving unit 21 in combination with an acoustic signal from which the voice recognition result is obtained, for one voice recognition result, a plurality of position information One piece of position information representing position information is paired with the speech recognition result. The one position information representing a plurality of position information may be anything as long as the position where the sound signal corresponding to the speech recognition result can be substantially identified, for example, any one of a plurality of position information The position information may be an average value of latitudes included in a plurality of position information and an average value of longitudes included in the plurality of position information.

  Among the operations performed by the broadcast reception unit 41 of the cloud-side device 2 of the third operation example, the operation different from the operations performed by the broadcast reception unit 41 of the first operation example is that the broadcasting station ID is always output, that is, an acoustic signal , Time point information and a station ID. The other operations are the same as in the first operation example.

  The operation performed by the broadcast speech recognition unit 42 of the cloud-side device 2 of the third operation example differs from the operation performed by the broadcast speech recognition unit 42 of the first operation example in that the broadcasting station ID is always input / output, A set of an acoustic signal, time information and a broadcasting station ID is input, and a set of a character string of speech recognition result, time information and a broadcasting station ID is output. The other operations are the same as in the first operation example.

  The speech recognition result holding unit 43 of the cloud-side device 2 in the third operation example is a combination of the speech recognition result output from the speech recognition unit 22, time information, position information, and ID, and the speech output from the broadcast speech recognition unit 42. A recognition result, a set of time information and a broadcasting station ID are stored. The contents stored in the voice recognition result storage unit 43 are processes for determining whether the voice recognition result processing unit 44 has a client-side device in the public broadcast reception target area, and there is a partial character string such as a word having a common time. It is used for the process of determining whether or not it is a process of obtaining a processed speech recognition result by removing it from the speech recognition result when there is a partial character string such as a word having a common time. Therefore, in the voice recognition result holding unit 43, a combination of the voice recognition result outputted by the voice recognition unit 22, time information, position information and ID, the voice recognition result outputted by the broadcast voice recognition unit 42, time information and the broadcasting station The combination with the ID is stored for the time required for the processing of the speech recognition result processing unit 44. Further, the stored contents held in the speech recognition result holding unit 43 may be deleted when the processing of the speech recognition result processing unit 44 using the stored contents is completed.

  The voice recognition result processing unit 44 of the cloud device 2 of the third operation example is a combination of the voice recognition result, time information, position information, and ID of at least one client device stored in the voice recognition result storage unit 43. Among the combination of speech recognition result of each public broadcast and time information and broadcasting station ID stored in the speech recognition result holding unit 43, there is a client-side device in the reception target area of the public broadcast, partial character string and time If there is a match in the pair, the result of removing the matched partial character string is taken as a processed speech recognition result, and the processed speech recognition result and the ID are output as a set. Therefore, processed speech recognition results for at least one client-side device will be output.

  Whether the client-side device is in a public broadcast reception target area may be determined by matching known absolute positions with identifiable information. For example, the storage unit (not shown) in the voice recognition result processing unit 44 stores in advance information on the country, prefecture, city, town, etc. for which the broadcast station ID of public broadcasting is received, and a map with latitude and longitude attached. Every country, prefecture, municipality, etc. where the client-side device is located is determined from the latitude and longitude specified by the location information of the client-side device, and the requested country, prefecture, municipality, etc. It may be determined depending on whether it corresponds to etc. Alternatively, for example, information on the range of latitude and longitude of the reception target area of the broadcast station ID of the public broadcast is stored in advance in a storage unit (not shown) in the voice recognition result processing unit 44, and specified by the position information of the client side device It may be determined based on whether the latitude and longitude of the position where the acoustic signal corresponding to the voice recognition result is generated is within the range of the latitude and longitude of the reception target area.

  It should be noted that the determination as to whether or not the times coincide is determined to be the same time in consideration of an error in absolute time in the client side device and the public broadcasting station or the cloud side device or an error in time in speech recognition processing. You may

  That is, the voice recognition result processing unit 44 of the cloud device 2 of the third operation example relates to public broadcasting in which the client device is in a reception target area at substantially the same time for at least one client device. When there is the same partial character string (common partial character string) as the client side device, a character string of the speech recognition result of the client side device from which the common partial character string is removed is obtained as a processed speech recognition result .

  Note that the process of removing the common partial character string may not necessarily be performed on all the public broadcast stations in which the client-side device is the reception target area, and may be performed on at least one public broadcast. In this case, only the speech recognition result required by the processing of the speech recognition result processing unit 44 may be obtained in the previous stage. That is, the operations of the broadcast reception unit 41, the broadcast speech recognition unit 42, and the speech recognition result holding unit 43 for obtaining speech recognition results unnecessary for the processing of the speech recognition result processing unit 44 may be omitted.

Here it will be described with reference to the processing flow in FIG. 14 for example at least some one client device is a client-side apparatus 1 _1. Example of FIG. 14, as the target speech recognition results of all public broadcasting station, in order from the public broadcasting station 5 _1, it is determined whether the received target area of the public broadcasting stations have client-side apparatus 1 _1, the public if the received target area of the broadcast station is a client-side device 1 _1, the speech recognition result of the public broadcasting station, those set of the speech recognition result of the client-side apparatus 1 ₁ and the partial strings and time matches searching whether, if the combination of the partial character string and time there is a match, the set of client-side apparatus 1 ₁ for speech recognition a substring that matches the partial character string and time It is an example which removes the said common substring from the character string of a result.

Speech recognition result processing unit 44 first reads a set of the position information and the ID and the client device 1 ₁ of the speech recognition result and the time information from the speech recognition result holding unit 43 (step S441A). Next, the speech recognition result processing unit 44 sets an initial value y to 1 (step S442). Next, the speech recognition result processing unit 44 reads out the combination of the speech recognition result of the public broadcast station 5 _y , the time information, and the broadcasting station ID from the speech recognition result holding unit 43 (step S443A). Speech recognition result processing unit 44, then, a set of client-side apparatus 1 ₁ speech recognition result and the time information and the position information and the speech recognition result sets and public broadcasting stations 5 _y of the ID and the time information and the broadcast station ID And the position information included in the combination of the voice recognition result of the client-side device 11 and the time information and the position information is included in the reception target area of the public broadcast station 5 _y , and the partial character string and the time match Search for (step S444A). Speech recognition result processing unit 44, then, when conditions are satisfied in step S444A removes all partial strings substring and its time matches the client-side apparatus 1 ₁ of the speech recognition result string (Step S445A). If the condition of step S444A is not satisfied, the process proceeds to step S446. Next, the speech recognition result processing unit 44 determines whether there are any public broadcast stations not subject to the processes of step S443, step S444A, and step S445A (step S446). Next, when it is determined in step S446 that there are still public broadcasting stations that are not targets of the processes of step S443, step S444A, and step S445A, the voice recognition result processing unit 44 replaces y with y + 1 (step S447). Step S443 In step S446, step S444A, when it is determined that there are no remaining public broadcasting station that is not subject to the process of step S445A is end processed speech recognition client device _{1 1} in step S445A Been results of the string in the ID and the set output as client-side apparatus 1 ₁ of processed speech recognition result string (step S448).

  By using the speech recognition system according to the third operation example of the second embodiment, the possibility that the speech recognition result of the speech emitted by the utterer, which is the speech recognition result desired by the utterer, is lower than in the first operation example It is possible to solve the problem 1 of the problem 1 while suppressing the possibility that the speech recognition result different from the speech recognition result desired by the utterer can be obtained than before, and the search result desired by the utterer in the search It is possible to reduce the possibility of obtaining different search results as compared to the prior art.

Third Embodiment
Next, as a third embodiment of the present invention, the speech recognition result of the acoustic signal including the voice of the speaker input to the client side device is common to the speech recognition result of the acoustic signal reproduced by the client side device. The form which removes a part is demonstrated. The configuration of the speech recognition system in the third embodiment is the same as that of the speech recognition system in the first embodiment, and a block diagram showing the configuration of the speech recognition system is FIG. The code 100 is a speech recognition system, the codes 1 ₁ to 1 _N are one or more (N, N is an integer of 1 or more) client side apparatuses, and the code 2 is a cloud side apparatus. In the third embodiment, the client side devices 1 ₁ to 1 _N have a function of reproducing the content stored in the client side device and / or a function of reproducing the content while the client side device downloads via the network 3 It is. Media may be attached to the "stored content". The content includes an audio signal including voice of at least Japanese and foreign languages such as serifs. And so on.

Since all the configurations included at least in the client side devices 1 ₁ to 1 _N are the same, in the following, a part configured by the client side device ₁₁ and the cloud side device 2 in the speech recognition system 100 of the third embodiment Will be described with reference to FIG. 15 which is a detailed block diagram of FIG. Among the components shown in FIG. 15, the components assigned the same reference numerals as those in FIG. 11 perform the same operations as those in FIG.

Client device _{1 1} includes an audio input unit 11 _1, the user information acquiring unit 12 _1, the audio output unit 13 _1, the search result receiving unit 14 _1, the screen display unit 15 _1, the content information obtaining unit 16 _1, the contents information sending part It comprises 17 ₁ at least. Client device _{1 1} of the speech input unit 11 _1, the user information acquiring unit 12 _1, the audio output unit 13 _1, the search result receiving unit 14 _1, the screen display unit 15 _1, the client of the first operation example of the second embodiment side device _{1 1} of the speech input unit 11 _1, the user information acquiring unit 12 _1, the audio output unit 13 _1, the search result receiving unit 14 _1, and the screen display unit 15 _1, respectively the same operation.

  The cloud side device 2 includes a voice receiving unit 21, a voice recognition unit 22, a content voice recognition result storage unit 60, a content information receiving unit 61, a content voice recognition result obtaining unit 62, a voice recognition result holding unit 63, and a voice recognition result processing unit 64 includes at least a search processing unit 25 and a search result sending unit 26. The voice receiving unit 21, the voice recognition unit 22, the search processing unit 25 and the search result sending unit 26 of the cloud side device 2 are the voice receiving unit 21 and the voice recognition portion of the cloud side device 2 according to the first operation example of the second embodiment. 22. The search processing unit 25 and the search result sending unit 26 operate in the same manner.

  Hereinafter, parts different from the first operation example of the second embodiment will be described.

Client device 1 ₁ of the content information acquisition unit 16 _1, the content that the client-side apparatus 1 ₁ is currently reproduced, identification information capable of identifying the content (hereinafter, referred to as "content ID". In the same movies Even though there are different words depending on the choice of language such as Japanese and foreign language, in the following explanation, in the case of content corresponding to voice of multiple languages such as Japanese and foreign language, they have different content ID respectively treated as possible to, and omitted), the relative time that represents the position currently being played during the content (a so-called playback position), to obtain the outputs to the content information sender 17 _1. The relative time indicating the portion currently being reproduced in the content is, for example, the number of seconds required to reproduce the portion when reproducing at a standard speed from the start point of the content, or It is a time stamp or the like given in advance.

The content information sender 17 _first client-side apparatus 1 _1, identifiable identification information client device 1 ₁ (hereinafter, referred to as "ID") and the content ID and the relative time that the content information acquisition unit 16 ₁ is output If, by the time information which the user information acquisition section 12 ₁ has output, and to set, to the ID and the content ID and the third transmission signal cloud side device 2 is a transmission signal including a set of the relative time and the time information Send out. The identification information may be specified, for example, using an existing external cloud service that can acquire a content ID and a relative time (reproduction position) from an optical media information database or fragmentary audio data.

  The content voice recognition result storage unit 60 of the cloud-side device 2 compares, in advance, the content ID for a content such as a movie, and the character string of the voice recognition result obtained by speech recognition of the sound signal of the content. The time is associated with each other and stored. The character string of the speech recognition result and the relative time are stored by combining the partial character string and the relative time for each partial character string included in the character string of the speech recognition result.

The content information reception unit 61 of the cloud side device 2 receives the third transmission signal content information sender 17 _first client device 1 ₁ is transmitted, ID and the content ID and the relative included in the third transmission signal A set of time and time information is obtained and output to the content voice recognition result acquisition unit 62.

  The content voice recognition result acquisition unit 62 of the cloud-side device 2 searches the content voice recognition result storage unit 60 using the content ID and relative time output by the content information receiving unit 61, and the voice of the content corresponding to the content ID Obtain a pair of the partial character string and relative time for each partial character string included in the character string of the recognition result, replace the relative time with the time information output by the corresponding content information receiving unit 61, and A pair of the partial character string and time information for each partial character string included in the character string of the speech recognition result is generated, and the partial character string for each partial character string included in the character string of the generated speech recognition result A set of time information and an ID are output to the voice recognition result holding unit 63 as a set.

  The voice recognition result holding unit 63 of the cloud-side device 2 uses the combination of the voice recognition result output by the voice recognition unit 22, the time information and the ID, and the character string of the voice recognition result output by the content voice recognition result acquisition unit 62. A combination of the partial character string, time information pair, and ID for each partial character string included is stored. The stored contents of the voice recognition result holding unit 63 are processes of the voice recognition result processing unit 64 determining whether or not there is a partial character string such as a word having a common time, and a partial character string such as a word having a common time Is used for processing to obtain a processed speech recognition result by removing it from the speech recognition result when there is an error. Therefore, the stored contents held in the speech recognition result holding unit 63 may be deleted when the processing of the speech recognition result processing unit 64 using the stored contents is completed.

The voice recognition result processing unit 64 of the cloud side device 2 stores the combination of the voice recognition result of at least one client side device stored in the voice recognition result holding unit 63, the time information, and the ID in the voice recognition result holding unit 43. A combination of a partial character string and a time in a combination of a partial character string for each partial character string included in the character string of the stored voice recognition result and a set of time information and an ID If there is a match, the processed speech recognition result is obtained by removing the matched partial character string, and the processed speech recognition result and the ID are output as a set. The determination as to whether or not the times coincide may be determined as the same time in consideration of an error of the time in the speech recognition process. Processing Flow in at least some one client device is a client-side apparatus 1 ₁ is as shown in FIG 16.

  Thus, according to the third embodiment, when playing a movie or VOD, for example, the content ID and the number of seconds of the playback position are also sent to the cloud-side device 2 together with the acoustic signal acquired by the client-side device. As a result, the cloud device 6 searches the DB in which the voice recognition result of the content is accumulated according to the content ID and the number of seconds of the playback position, and obtains the voice recognition result (announcement or speech character string) of the voice of the content. The user of the client-side device by excluding the speech recognition result (an announcement or speech character string) of the obtained voice of the content as the noise from the character string of the speech recognition result of the acoustic signal acquired by the client-side device It is possible to lower the probability of misrecognition of speech recognition with respect to the speech emitted by.

  That is, by using the speech recognition system according to the third embodiment, it is possible to solve the problem 3 of the problem.

Fourth Embodiment
Next, as a fourth embodiment of the present invention, an embodiment in which an input of a search instruction is specified will be described. Here, a form in which the input of the search instruction is specified in the second embodiment of FIG. 11 will be described with reference to FIG. FIG. 17 is a detailed block diagram of a portion constituted by the client side device ₁₁ and the cloud side device 2 in the speech recognition system 100 of the fourth embodiment corresponding to the second embodiment. To indicate configuration differs from the structure shown in FIG. 11 FIG. 17 is that the client-side apparatus 1 ₁ is configured to include at least also search instruction input unit 18 _1. Search instruction input unit 18 ₁ except for components of FIG. 17 is the same as FIG. 11. The differences from the description of the second embodiment will be described below.

[[Operation example of the fourth embodiment]]
4 As an example of the operation of the embodiment, speaks a sentence to be obtained search results first user to the client-side apparatus 1 _1, the screen display unit of the search result corresponding to the utterance client device 1 ₁ An example of the operation in the case of displaying in 15 ₁ will be described.

Search instruction input unit 18 ₁ of the client-side apparatus 1 _1, when the first user utters a sentence to be obtained search results, accepts the input of a search start instruction, the voice input an instruction received search start part 11 and outputs to ₁ and the user information acquisition section 12 ₁ and the audio output unit 13 _1. The instruction to start the search can be said to be an instruction to start both the speech recognition and the search. For example, if the client-side apparatus 1 ₁ is a smart phone, a voice search start button displayed on a screen, detecting means for detecting that the voice search start button is touched, the client device 1 ₁ Search is an instruction input unit 18 _1.

Client device 1 ₁ of the speech input unit 11 ₁ in accordance with the instruction of the search instruction input unit 18 ₁ is output search start obtains the acoustic signal, and outputs the acquired audio signal to the audio output unit 13 _1. For example, the speech input unit 11 _1, the acquisition of the acoustic signal at the time when the search start instruction starts acquisition of acoustic signals at the time of the input, the time search start instruction is predetermined from the time input has elapsed Finish. Point addition, for example, the speech input unit 11 _1, which comprises a speech detecting means (not shown), an instruction to start searching starts acquisition of acoustic signals at the time of the input is determined that the speech detecting means has disappeared speech End acquisition of the acoustic signal.

User information acquisition section 12 ₁ of the client-side apparatus 1 ₁ search instruction in accordance with an instruction input portion 18 ₁ is the search start outputting, to obtain time information voice input unit 11 ₁ of the client-side apparatus 1 ₁ acquires the audio signal Te, the time information and the client-side apparatus 1 ₁ can specify identification information (hereinafter, referred to as "ID") to the audio output unit 13 ₁ and the user information. For example, the user information acquiring unit 12 _1, the audio while the input unit 11 ₁ is output by obtaining the acoustic signal, obtains time information, the audio output unit 13 and the time information obtained and the ID as the user information ₁ Output to

User information sound output unit 13 ₁ of the client-side apparatus 1 ₁ according to an instruction of the search instruction input unit 18 searches _{started 1} is output, the acoustic signal and the user information acquisition section 12 ₁ speech input unit 11 ₁ is output is output And transmits to the cloud device 2.

  The subsequent operation of the speech recognition system 100 of the operation example of the fourth embodiment is the same as that of the first operation example of the second embodiment.

With this configuration, in response to the retrieval instruction input unit 18 ₁ of the client-side apparatus 1 ₁ accepts the input of a search start instruction, corresponding to the sentence to be obtained search results first user has uttered it is possible to display the search results to the client-side apparatus 1 ₁ of the screen display unit 15 _1.

  In the operation examples other than the first operation example of the second embodiment, the operation of the speech recognition system 100 which clearly indicates the input of the search instruction is the same as that described above in the first embodiment and the third embodiment as well. Although the description is omitted, when the search instruction input unit of the client side device receives the input of the search start instruction, the search result corresponding to the sentence for which the user wants to obtain the search result spoken by the user is It becomes possible to display on the screen display unit.

Modification of Fourth Embodiment
Next, as a modified example of the fourth embodiment of the present invention, an embodiment using an acoustic signal before the input time of a search instruction will be described using FIG. FIG. 18 is a detailed block diagram of a portion constituted by the client side device ₁₁ and the cloud side device 2 in the speech recognition system 100 according to the modification of the fourth embodiment shown in FIG. Configuration shown in Figure 18 differs from the structure shown in FIG. 17 is that the client device 1 ₁ is configured to include at least also voice holding unit 19 _1. The components other than the voice holding unit _{1 of} FIG. 18 are the same as those of FIG. Below, the difference with 4th Embodiment is demonstrated.

[[Operation Example of Modification of Fourth Embodiment]]
As an example of the operation of the modification of the fourth embodiment, an example of the same case as the operation of the fourth embodiment, i.e., speaks a sentence to be obtained search results first user to the client-side apparatus 1 ₁ , an example of operation of displaying a search result corresponding to the utterance to the screen display unit 15 ₁ of the client-side apparatus 1 _1.

Speech input unit 11 ₁ of the client-side apparatus 1 ₁ always acquires an acoustic signal. Speech input unit 11 ₁ is searched if the instruction input unit 18 instruction to start searching from ₁ is input, searches in accordance with the instruction input unit 18 instruction search start input _1, speech sending unit acoustic signals acquired 13 Output to ₁ For example, ₁ speech input unit 11, when the search instruction input unit 18 search start instruction from ₁ is input in advance from the time the search start instruction has been input, from the time the instruction is input in the search start up to the point where an agreement time elapses, and outputs a sound signal to the sound output unit 13 _1. The voice input unit 11 ₁ outputs all the acoustic signals acquired on the speech holding unit 19 _1.

User information acquisition section 12 ₁ of the client-side apparatus 1 ₁ always obtains the time information of time at which the speech input unit 11 ₁ obtains an acoustic signal. User information acquisition unit 12 ₁ is searched if the instruction input unit 18 search start instruction from ₁ is input, according to an instruction of the search instruction input unit 18 ₁ is input from the search start, the client-side apparatus 1 ₁ speech and time information of an acoustic signal input unit 11 ₁ outputs the voice output unit 13 _1, the speech the time information and the client-side apparatus 1 ₁ can specify identification information (hereinafter, referred to as "ID") and a user information and outputs to the transmission unit 13 _1. The user information acquisition section 12 ₁ outputs all the time information acquired in the speech holding unit 19 _1.

Client device 1 ₁ of the speech holding unit 19 ₁ stores in the storage unit (not shown) by the time information input from the sound signal and the user information acquisition unit 12 ₁ that is input from the speech input unit 11 ₁ in the set, A set of an acoustic signal and time information for which a predetermined time has elapsed from the latest one is deleted from the storage unit. That is, the speech holding unit 19 ₁ stores the time information corresponding to the acoustic signal and its acoustic signal input from the speech input unit 11 ₁ to the most recent one predetermined time period. The predetermined time is a time set in advance, and is, for example, about ten seconds to several minutes. Further, ₁ voice holding unit 19, when the search instruction input unit 18 search start instruction from ₁ is inputted, a set of the acoustic signal and the time information stored in the storage unit to the audio output unit 13 ₁ Output. That is, the speech holding unit 19 ₁ is searched when an instruction search start instruction from the input unit 18 ₁ is input, the output audio signals of the predetermined time from the latest one and its time information to the audio output unit 13 ₁ Do.

Voice output unit 13 ₁ of the client-side apparatus 1 ₁ search in accordance with an instruction of search start input from the instruction input unit 18 ₁ is input from the audio signal and the user information acquisition unit 12 ₁ that is input from the speech input unit 11 ₁ sending user information and the acoustic signal input from the voice holding portion 19 ₁ and a transmission signal including its time information to the cloud side device 2. That is, the voice output unit 13 _1, the acoustic signal up to the point of time when the search instruction starts when an instruction to start searching has been input a predetermined from the time input has elapsed and its time information, initiating searches instructions than when it is inputted sends a past audio signal in a predetermined time duration and its time information, and the ID of the client-side apparatus 1 _1, a transmission signal including the cloud side device 2.

  The operations of the voice reception unit 21, voice recognition unit 22, broadcast reception unit 41, and broadcast voice recognition unit 42 of the cloud side device 2 are the same as the voice reception unit 21, voice recognition unit 22, and broadcast reception unit 41 of the second embodiment. , The operation of the broadcast speech recognition unit 42 is the same.

  The voice recognition result processing unit 44 of the cloud side device 2 first makes a voice recognition result holding unit for the combination of the voice recognition result of at least one client side device stored in the voice recognition result holding unit 43, time information, and an ID. The voice of the client-side device is targeted only for public broadcasts in which there are a plurality of combinations of partial character string and time in the set of speech recognition result of public broadcast and time information stored in 43. From the character string of the recognition result, the partial character string in which the combination of partial character string and time matches the speech recognition result of the public broadcast is removed, and the character string of the speech recognition result of the client-side device after removal is obtained. The voice recognition result processing unit 44 further performs voice recognition on the character string of the voice recognition result of the client-side device after removal from which the partial character string in the past from the time when the search start instruction is input is removed. As a result, the processed speech recognition result and the ID are output in pairs.

Next, an example of the speech recognition result and the processed speech recognition result in this example will be described with reference to FIG. The horizontal axis in FIG. 19 represents time, and the time when a search instruction is input is T ₀ , the time when a predetermined time has elapsed from time T ₀ when a search instruction is input is T _A , and Let T _{B be} the time of a predetermined time past the input time T ₀ . Three above the thick arrow in the upper is an input client device 1 ₁ and each of the speech recognition result of the public broadcasting station 5 ₁ and public broadcasting station 5 ₂ of the speech recognition result processing section 44, the lower one below the thick arrow in the are processed speech recognition result of the client-side apparatus 1 _1.

The client-side apparatus 1 ₁ of the speech recognition results, search instruction as the result time of the speech recognition until the time T _A the time determined in advance from the time T ₀ that is input has elapsed, the client-side apparatus 1 ₁ of the user there first user and a sub-string of the speech recognition result of the speech 2 is a speech uttered substring of a speech recognition result of the television audio 2 is a audio television emitted around the client-side apparatus 1 ₁ And is included. Moreover, the client-side apparatus 1 ₁ of the speech recognition results, as the predetermined time speech recognition result of time from past time T _B until time T ₀ the search instruction is input at time T ₀ the search instruction is input, first user and a sub-string of the speech recognition result of the speech 1 is a speech uttered, TV voice is the voice of television emitted around the client-side apparatus 1 ₁ is a client-side apparatus 1 ₁ of the user 1 and a partial string of the speech recognition result are included.

A public broadcasting station 5 ₁ for speech recognition result, the search instruction as the result time of the speech recognition until the time T _A the time determined in advance from the time T ₀ that is input has elapsed, public broadcasting station 5 ₁ is broadcast sound A partial character string of the speech recognition result of the television speech 2 which is the speech included in the signal is included. Further, the speech recognition result of the public broadcasting station 5 _1, as a predetermined time speech recognition result of time from past time T _B until time T ₀ the search instruction is input at time T ₀ the search instruction is input, public broadcasting station 5 ₁ contains a substring of the speech recognition result of the television audio 1 is a sound included in the sound signal broadcast.

A public broadcasting station 5 ₂ of the speech recognition results, search instruction as the result time of the speech recognition until the time T _A the time determined in advance from the time T ₀ that is input has elapsed, public broadcasting station 5 ₂ is broadcast sound A partial character string of the speech recognition result of the television speech 4 which is the speech included in the signal is included. Also, the public broadcasting station 5 ₂ of the speech recognition results, as the predetermined time speech recognition result of time from past time T _B until time T ₀ the search instruction is input at time T ₀ the search instruction is input, public broadcasting station 5 ₂ contains a substring of the speech recognition result of the television audio 3 is a sound included in the sound signal broadcast.

The client-side apparatus 1 of the _first speech recognition result string and public broadcasting stations 5 ₁ speech recognition result string, while from the time T _B at time T _A, which substring and its time matches There are two partial character strings of the speech recognition result of the television sound 1 and the speech recognition result of the television sound 2. Thus, public broadcasting station 5 ₁ for a plurality of substrings, for partial strings and their time are the same, the removed target. Then, the speech recognition result processing unit 44, the client-side apparatus 1 ₁ of the speech recognition result string, to the public broadcasting station 5 ₁ for speech recognition result string same substring are present at the same time The partial strings of the speech recognition result of television voice 1 and the partial strings of the speech recognition result of television voice 2 which are all partial strings are removed. The client-side apparatus 1 of the _first speech recognition result string and public broadcasting station 5 ₂ of the speech recognition result string, while from the time T _B at time T _A, the partial character string and its time matches parts There is no string. Thus, public broadcasting stations 5 _2, for a plurality of substrings, for partial strings and their time do not match, not a removed object. Client device 1 ₁ of a result of removing the processing up to this point for the character string of the speech recognition result is an illustration between the upper thick arrow and the lower thick arrow in FIG. 19.

Then, the speech recognition result processing unit 44, the client-side apparatus 1 ₁ of the speech recognition result string, the partial character is a sequence of speech first speech recognition result sub-string between times T ₀ from the time T _B Get rid of As a result, those only partial string of the speech recognition result of the speech 2 is left is outputted as processed speech recognition result of the client-side apparatus 1 _1.

  The subsequent operation of the speech recognition system 100 of the operation example of the modification of the fourth embodiment is the same as the operation example of the fourth embodiment.

Similarly to the modification of the fourth embodiment, also all embodiments operation example of the first to third embodiments, the like comprises a speech holding unit 19 _1, than search start instruction of the past acoustic The signal may be used to operate the speech recognition system 100.

  As in the modification of the fourth embodiment, the configuration is made to operate using an acoustic signal in the past rather than a search start instruction, in particular, the third operation example of the first embodiment and the second operation example of the second embodiment. In a configuration in which processing for removing a speech recognition result is performed on another client-side device or a public broadcast station having a plurality of partial character strings in common as in the operation example, a configuration not using acoustic signals in the past rather than a search start instruction The response speed can be faster than in the case of

<Embodiment of Speech Recognition Device>
Although the voice recognition system described above has a configuration in which the client side devices 1 ₁ to 1 _N and the cloud side device 2 are connected by the network 3, even if the cloud side device 2 is configured by a plurality of server devices etc. Good. Further, the speech recognition system may not be a cloud type system, but may be a stand-alone type speech recognition device. That is, it may be a voice recognition device in which the configuration of the cloud side device 2 is provided in the client side devices 1 ₁ to 1 _N.

In the above description, an example in which the speech recognition result is applied to information search has been described, but the speech recognition result may be used in any way. That may be a speech recognition device including an only essential part of the portion constituted by the client-side apparatus 1 ₁ and the cloud side device 2 shown in FIGS. 2 and 11. These speech recognition devices will be described with reference to FIG.

[[First Example of Speech Recognition Device]]
FIG. 20A is a block diagram showing a first example of the speech recognition apparatus. The speech recognition apparatus 700 of the first example includes at least a speech recognition unit 710 and a speech recognition result 720.

  The speech recognition unit 710 of the speech recognition apparatus 700 corresponds to the speech recognition unit 22 of FIG. For example, the voice recognition unit 710 is a first sound signal which is a sound signal collected including a first speaker voice by a first sound collecting means such as a microphone of a smartphone of a first speaker of a voice recognition target. Second to N sound collecting means different from the signal and the first sound collecting means such as microphones of second to N (N is an integer of 2 or more) users different from the first speaker The voice signal of each of the second to N-th sound signals which are the collected sound signals is voice-recognized, and the first voice recognition result to the N-th voice which is the voice recognition result for each of the sound signals Get recognition results. Here, the first to Nth acoustic signals are, for example, acoustic signals including the same time. For example, the first sound signal is a sound signal including a voice uttered by the first speaker as a voice recognition target, and the second sound signal to the Nth sound signal have the same beginning and end as the first sound signal. Or it is an acoustic signal which is the absolute time of the neighborhood.

  The speech recognition result processing unit 720 of the speech recognition apparatus 700 corresponds to the speech recognition result holding unit 23 and the speech recognition result processing unit 24 in FIG. For example, the speech recognition result processing unit 720 may generate a partial speech recognition result included in at least one of the second speech recognition result to the Nth speech recognition result and a partial speech recognition result included in the first speech recognition result. When the content of the partial speech recognition result is the same and the partial speech recognition result corresponds to the sound signal at substantially the same time, the partial speech recognition result is deleted from the first speech recognition result Obtained as a speech recognition result of the first speaker. In addition to the fact that the contents of the partial speech recognition result are the same and the time is substantially the same, the position of the second to N sound pickup means is in the vicinity of the first sound pickup means. In some cases, the partial speech recognition result may be deleted from the first speech recognition result.

[[Second example of speech recognition device]]
FIG. 20B is a block diagram showing a second example of the speech recognition device. The speech recognition apparatus 701 of the second example includes at least a speech recognition unit 711 and a speech recognition result 721.

  The speech recognition unit 711 of the speech recognition device 701 corresponds to the speech recognition unit 22 and the broadcast speech recognition unit 42 in FIG. For example, the voice recognition unit 711 is a first sound that is an acoustic signal collected including a first speaker voice by a first sound collecting unit such as a microphone of a smartphone of a first speaker of a voice recognition target. The audio signal of each of the first broadcast audio signal to the M-th broadcast audio signal (M is an integer of 1 or greater) which is one or more broadcast audio signals is voice-recognized, and the voice for each audio signal is voice-recognized. The first speech recognition result and the first broadcast speech recognition result to the Mth broadcast speech recognition result, which are recognition results, are obtained. Here, the first acoustic signal and the first to M-th broadcast acoustic signals are, for example, acoustic signals including the same time. For example, the first acoustic signal is an acoustic signal including a voice uttered by the first speaker as a voice recognition target, and the first to Mth broadcast acoustic signals have a first acoustic signal at the beginning and the end, respectively. Is an acoustic signal that is the same as or near an absolute time.

  The speech recognition result processing unit 721 of the speech recognition device 701 corresponds to the speech recognition result holding unit 43 and the speech recognition result processing unit 44 in FIG. For example, the voice recognition result processing unit 721 may select a partial voice recognition result included in at least one or more of the first broadcast voice recognition result to the Mth broadcast voice recognition result and a partial voice included in the first voice recognition result. When the recognition result is the same as the partial speech recognition result and the partial speech recognition result corresponding to the sound signal at substantially the same time, the partial speech recognition result is deleted from the first speech recognition result One is obtained as the speech recognition result of the first speaker. The voice recognition result processing unit 721 targets the voice recognition results of the second to M broadcasts in which the first sound collecting means is in the reception target area, and the contents of the partial voice recognition results are the same and the time is substantially the same. In this case, the partial speech recognition result may be deleted from the first speech recognition result.

  According to these voice recognitions, even if the user utters in an environment where environmental sounds such as television, radio, and guide broadcasting exist at relatively large volume, voice recognition of environmental sounds with high accuracy is realized. The result can be removed, and by reducing the possibility that unnecessary speech recognition results are included, the speech recognition rate for the speech of the speaker can be improved.

  In the above description, the speech recognition result is described as a character string, but when the speech recognition result is represented by a string of symbols representing phonemes, a string of phoneme symbols is used instead of the character string. May be That is, the character string or partial character string of the speech recognition result in the above description is an example of the speech recognition result such as a phoneme symbol string or partial phoneme character string of the speech recognition result or a part of the content thereof.

  All or part of the speech recognition system in the above-described embodiment may be realized by a computer. In that case, a program for realizing this function may be recorded in a computer readable recording medium, and the program recorded in the recording medium may be read and executed by a computer system. Here, the “computer system” includes an OS and hardware such as peripheral devices. The term "computer-readable recording medium" refers to a storage medium such as a flexible disk, a magneto-optical disk, a ROM, a portable medium such as a ROM or a CD-ROM, or a hard disk built in a computer system. Furthermore, “computer-readable recording medium” dynamically holds a program for a short time, like a communication line in the case of transmitting a program via a network such as the Internet or a communication line such as a telephone line. It may also include one that holds a program for a certain period of time, such as volatile memory in a computer system that becomes a server or a client in that case. Further, the program may be for realizing a part of the functions described above, or may be realized in combination with the program already recorded in the computer system. It may be realized using hardware such as PLD (Programmable Logic Device) or FPGA (Field Programmable Gate Array).

  Although the embodiments of the present invention have been described above with reference to the drawings, it is apparent that the above embodiments are merely examples of the present invention, and the present invention is not limited to the above embodiments. is there. Therefore, additions, omissions, substitutions, and other modifications of the components may be made without departing from the technical spirit and scope of the present invention.

  The present invention recognizes and recognizes an acoustic signal including the voice of the input speaker even if a voice of relatively large volume is present in the background of the voice intended to be recognized by the speaker. Unnecessary speech is obtained by obtaining a result and performing speech recognition on another acoustic signal input to obtain a recognition result, and removing a common one among the recognition results from the speech recognition result of the input acoustic signal. By reducing the possibility of including a recognition result, the speech recognition rate for the speech of the speaker is improved. Therefore, in various applications in which it is essential to obtain the result of speech recognition of only the speech of the speaker even if there is a relatively large volume of speech in the background of the speech that the speaker wants to recognize. Is also applicable.

  For example, in the living room of a general home, according to the present invention, it is not necessary to move away the sound input unit and the position of a speaker sound source such as a TV or AV amplifier, and a remote control device with a voice recognition microphone becomes unnecessary. It will be possible to recognize voice with only the sensitivity microphone. Therefore, convenience is improved by cost reduction of the terminal system severe to the device cost, weight reduction of the remote control, and reduction of power consumption of the remote control.

  In addition, the present invention is a voice in an Olympic hall, a station, an airport, a lecture hall, a train, a public viewing hall, etc. where television and radio with little silent and silent state are broadcasted and on-site guidance broadcasting is repeated in multiple languages. It is useful for the utilization of recognition.

  Further, according to the present invention, even in a car, speech recognition commands are uttered at any time without concern for a stand-alone car TV, traffic information or car radio voices (announcements, speeches) and related information by speech recognition. Search can be done. As a result, for example, an effect of increasing the convenience of the cloud-linked type voice agent service for cars is expected.

  In voice command recognition of a telephone automatic response system in a call center of a company, the influence of lifelike voice noise such as television and radio voice which often flows in the background of the user's home can be suppressed, thereby improving the accuracy Cost reduction by reducing operator intervention can also be expected as a secondary factor.

1 ₁ 1 ₂ 1 ₃ 1 1 _N ... Client side device 2 ... Cloud side device 3 ... Network 100 ... Speech recognition system 11 ₁ ... Speech input unit 12 DESCRIPTION OF SYMBOLS ₁ ... User information acquisition part, 13 ₁ ... Voice transmission part, 14 ₁ ... Search result receiving part, 15 ₁ ... Screen display part, 21 ... Voice reception part, 22 ... Voice Recognition part, 23 ... voice recognition result holding part, 24 ... voice recognition result processing part, 25 ... search processing part, 26 ... search result sending part, 5 ₁ , 5 _M ... public broadcast Bureau, 41 ... broadcast receiving unit, 42 ... broadcast audio recognition unit, 43 ... speech recognition result holding portion, 44 ... speech recognition result processing unit, ₁₆ 1 ... content information acquisition unit, 17 ₁ ... content information sending unit, 60 ... content speech recognition result accumulating section, 61 · The content information receiving section, 62 ... content speech recognition result obtaining unit, 63 ... speech recognition result holding portion, 64 ... speech recognition result processing unit, ₁₈ 1 ... retrieval instruction input section, 19 ₁ ... voice holding unit, 700 ... voice recognition device, 710 ... voice recognition unit, 720 ... voice recognition result processing unit, 701 ... voice recognition device, 711 ... voice recognition unit, 721 ... Speech recognition result processing unit

Claims (6)

A first sound signal which is a sound signal collected by the first sound collecting means including the voice of the first speaker and at least one sound collecting means different from the first sound collecting means; Each of the second to N-th sound signals, which are sound signals collected by the second to N-th (N is an integer of 2 or more) sound-pickup means, is voice-recognized, Voice recognition means for obtaining a first voice recognition result to a N-th voice recognition result as a voice recognition result for an audio signal;
The partial speech recognition result included in at least one of the second speech recognition result to the Nth speech recognition result and the partial speech recognition result included in the first speech recognition result are partial speech recognition results When the content is the same and the partial speech recognition result corresponds to the sound signal at substantially the same time, the partial speech recognition result deleted from the first speech recognition result is the first speaker's Voice recognition result processing means obtained as a voice recognition result;
Equipped with
The voice recognition result processing means
The partial speech recognition result included in at least one of the second speech recognition result to the Nth speech recognition result and the partial speech recognition result included in the first speech recognition result are partial speech recognition results The second speech recognition result to the Nth speech recognition result in which the contents are the same and there are a plurality of partial speech recognition results corresponding to the acoustic signals at substantially the same time (1) The partial speech recognition result contained in the speech recognition result, the content of the partial speech recognition result is the same, and all partial speech recognition results corresponding to the acoustic signals at substantially the same time are obtained to obtain partial speech A voice recognition apparatus for obtaining a result of removing a recognition result from the first voice recognition result as a voice recognition result of the first speaker . A first acoustic signal which is an acoustic signal collected including a voice of a first speaker by a first sound collecting means, and a first broadcast acoustic signal to an M-th broadcast acoustic which is an acoustic signal of one or more broadcasts Speech recognition is performed on the respective sound signals of the signal (M is an integer of 1 or more), and the first speech recognition result and the first broadcast speech recognition result which are speech recognition results for the respective speech signals ~ Mth broadcast speech recognition Voice recognition means for obtaining a result;
The partial speech recognition result included in at least one or more of the first broadcast speech recognition result to the Mth broadcast speech recognition result and the partial speech recognition result included in the first speech recognition result are partial speech recognition When the content of the result is the same and the partial speech recognition result corresponds to the acoustic signal at substantially the same time, the first speech in which the partial speech recognition result is deleted from the first speech recognition result Voice recognition result processing means to be obtained as the voice recognition result of the person,
Equipped with
The voice recognition result processing means
The partial speech recognition result included in at least one or more of the first broadcast speech recognition result to the Mth speech recognition result and the partial speech recognition result included in the first speech recognition result are partial speech recognition results Are the same, and there are a plurality of partial speech recognition results corresponding to acoustic signals at substantially the same time, and only the first to the Mth broadcast speech recognition results,
The partial speech recognition result is identical to the partial speech recognition result included in the first speech recognition result, and all partial speech recognition results corresponding to acoustic signals at substantially the same time are obtained. A voice recognition apparatus for obtaining, as the voice recognition result of the first speaker, a result obtained by deleting the obtained partial voice recognition result from the first voice recognition result . Voice recognition instruction input means for receiving an instruction to start voice recognition processing;
And sound holding means for holding the sound signal for a predetermined time.
The voice recognition means
When the voice recognition instruction input unit receives an instruction to start voice recognition processing, the voice signal including the sound signal held in the voice holding unit is voice-recognized and the voices including the past are voice-recognized. Get recognition results,
The voice recognition result processing means
A process of deleting the partial speech recognition result including the past from the first speech recognition result is performed using each speech recognition result including the past speech, and the speech recognition result of the first speaker is The speech recognition apparatus according to claim 1 or 2, which can be obtained. A first sound signal which is a sound signal collected by the first sound collecting means including the voice of the first speaker and at least one sound collecting means different from the first sound collecting means; Each of the second to N-th sound signals, which are sound signals collected by the second to N-th (N is an integer of 2 or more) sound-pickup means, is voice-recognized, Obtaining a first voice recognition result to a N-th voice recognition result, which are voice recognition results for an audio signal;
The partial speech recognition result included in at least one of the second speech recognition result to the Nth speech recognition result and the partial speech recognition result included in the first speech recognition result are partial speech recognition results When the content is the same and the partial speech recognition result corresponds to the sound signal at substantially the same time, the partial speech recognition result deleted from the first speech recognition result is the first speaker's A speech recognition result processing step obtained as a speech recognition result;
I have a,
In the voice recognition result processing step,
The partial speech recognition result included in at least one of the second speech recognition result to the Nth speech recognition result and the partial speech recognition result included in the first speech recognition result are partial speech recognition results The second speech recognition result to the Nth speech recognition result in which the contents are the same and there are a plurality of partial speech recognition results corresponding to the acoustic signals at substantially the same time (1) The partial speech recognition result contained in the speech recognition result, the content of the partial speech recognition result is the same, and all partial speech recognition results corresponding to the acoustic signals at substantially the same time are obtained to obtain partial speech A voice recognition method for obtaining a result of removing a recognition result from the first voice recognition result as a voice recognition result of the first speaker . A first acoustic signal which is an acoustic signal collected including a voice of a first speaker by a first sound collecting means, and a first broadcast acoustic signal to an M-th broadcast acoustic which is an acoustic signal of one or more broadcasts Speech recognition is performed on the respective sound signals of the signal (M is an integer of 1 or more), and the first speech recognition result and the first broadcast speech recognition result which are speech recognition results for the respective speech signals ~ Mth broadcast speech recognition A speech recognition step to obtain a result;
The partial speech recognition result included in at least one or more of the first broadcast speech recognition result to the Mth broadcast speech recognition result and the partial speech recognition result included in the first speech recognition result are partial speech recognition When the content of the result is the same and the partial speech recognition result corresponds to the acoustic signal at substantially the same time, the first speech in which the partial speech recognition result is deleted from the first speech recognition result A speech recognition result processing step obtained as a speech recognition result of the person,
Only including,
In the voice recognition result processing step,
The partial speech recognition result included in at least one or more of the first broadcast speech recognition result to the Mth speech recognition result and the partial speech recognition result included in the first speech recognition result are partial speech recognition results Are the same, and there are a plurality of partial speech recognition results corresponding to acoustic signals at substantially the same time, and only the first to the Mth broadcast speech recognition results,
The partial speech recognition result is identical to the partial speech recognition result included in the first speech recognition result, and all partial speech recognition results corresponding to acoustic signals at substantially the same time are obtained. A voice recognition method for obtaining a result obtained by deleting the obtained partial voice recognition result from the first voice recognition result as a voice recognition result of the first speaker . A voice recognition program for operating a computer as the voice recognition device according to any one of claims 1 to 3 .

Images