JP6784859B1 - Voice recognition system and voice recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice recognition system and a voice recognition device for reducing the recognition time of voice data transmitted by wireless communication. A voice recognition system 100 using wireless communication W includes acquisition means, transmission means, reception means, and recognition means. The acquisition means acquires voice data based on the voice. The transmission means transmits the voice data by wireless communication using the UHF band. The receiving means receives the voice data as a continuous signal at a time. The recognition means uses phoneme recognition to derive a recognition result that recognizes the content of the voice data. For example, the transmission means transmits the voice data that has not been packetized. [Selection diagram] Fig. 1

Description

The present invention relates to a voice recognition system and a voice recognition device.

Conventionally, as a technique related to voice recognition, for example, a voice recognition operation system of Patent Document 1 has been proposed.

In the disclosure technology of Patent Document 1, a voice input / output terminal into which a voice command is input by a user, a voice recognition device installed in a mechanical parking facility and connected to the voice input / output terminal by short-range wireless communication, and voice recognition A control device to which the device is connected and controls the operation of the mechanical parking facility, and a control device connected to the control device via a communication line, to confirm the user's permission to use, analyze voice commands, and to the user's voice input / output terminal. It is equipped with a control device that performs voice guidance notification and output of control commands to the control device.

JP-A-2019-214846

Here, Patent Document 1 discloses a technique for performing voice recognition by transmitting voice data to a voice recognition device using short-range wireless communication such as Wi-Fi (registered trademark). However, when receiving divided voice data using wireless packet communication such as Wi-Fi (registered trademark), it is different from the error detection for each divided voice data and the timing of receiving each voice data. It will be necessary to replace them in order. Therefore, the period from the transmission / reception of voice data to the derivation of the result of voice recognition may be longer than expected. As a result, for example, the user may determine that the voice has not been recognized and emit the voice again, which may cause unnecessary voice recognition processing.

Therefore, the present invention has been devised in view of the above-mentioned problems, and an object of the present invention is a voice recognition system and a voice recognition device for reducing the recognition time of voice data transmitted by wireless communication. To provide.

The voice recognition system according to the first invention is a voice recognition system using wireless communication, and transmits the voice data by an acquisition means for acquiring voice data based on voice and wireless communication using a UHF band. A transmission means, a receiving means for receiving the voice data as a continuous signal at a time, a recognition means for deriving a recognition result of recognizing the contents of the voice data using phonetic element recognition, character string information, and the character string. A start-up database including a plurality of start-up sentences including one or more thresholds associated with information is provided , and the recognition means uses the phonetic element recognition to obtain candidate data corresponding to the contents of the voice data and the candidate data. A generation means for generating a candidate sentence including one or more reliabilitys associated with, and a designation means for designating a first start sentence corresponding to the candidate sentence among a plurality of the start sentences by referring to the start database. It is characterized by having a comparison means for comparing the reliability with the first threshold value included in the first activation sentence, and a derivation means for deriving the recognition result based on the result of the comparison means. ..

The voice recognition system according to the second invention is characterized in that, in the first invention, the transmission means transmits the voice data that has not been packetized.

In the voice recognition system according to the third invention, in the first invention or the second invention , the plurality of the activation sentences are one sentence used for deriving the first recognition result indicating the start of voice recognition, and the first recognition. It is characterized by including a plurality of dummy sentences used for deriving the recognition result different from the result.

In the first invention, the voice recognition system according to the fourth invention further includes one sound collecting device and a plurality of driving devices having the voice recognition device, and the sound collecting device is a conversion unit that performs the acquisition means. And a communication unit that performs the transmission means, and the voice recognition device includes a reception unit that performs the reception means, a recognition unit that performs the recognition means, and a storage unit that stores the activation database. , The driving device is characterized by having a controlled unit controlled based on the recognition result derived by the recognition means.

In the voice recognition system according to the fifth invention, in the fourth invention, the plurality of activation sentences are one sentence used for the first recognition result indicating the start of voice recognition and the recognition different from the first recognition result. It includes a plurality of dummy sentences used in the result, and the content of the one sentence is different for each of the plurality of stored voice recognition devices.

The voice recognition system according to the sixth invention further includes a calibration means for calibrating the voice data in any one of the first to fifth inventions, and the calibration means is based on the sound for calibration. Evaluation sound data generation means for generating a plurality of evaluation sound data, evaluation sound data transmission means for transmitting the plurality of evaluation sound data by the wireless communication, and evaluation for receiving the plurality of evaluation sound data. An evaluation means for generating evaluation results for a plurality of the evaluation sound data using the sound data receiving means, the sound element recognition, an evaluation result transmitting means for transmitting the evaluation results by the wireless communication, and the evaluation. It is characterized by having a setting means for receiving the result and setting the acquisition condition of the voice data based on the evaluation result.

The voice recognition system according to the seventh invention is characterized in that, in the sixth invention, the acquisition condition includes at least one of a microphone gain, an auto gain control, a dynamic range control, and an acoustic echo cancellation.

The voice recognition device according to the eighth invention is a voice recognition device using wireless communication, and is a receiving unit that receives voice data transmitted by wireless communication using the UHF band at once as a continuous signal, and a sound element. A recognition unit that uses recognition to derive a recognition result that recognizes the contents of the voice data, a start-up database that stores a plurality of start-up sentences including character string information and a threshold value associated with the character string information of 1 or more, and a start-up database. The recognition unit includes a generation unit that uses the phonetic element recognition to generate candidate data corresponding to the content of the voice data, and a candidate sentence including one or more reliability associated with the candidate data, and the activation unit. The database is referred to, and among the plurality of the startup sentences, the designation unit that specifies the first startup sentence corresponding to the candidate sentence, the reliability, and the first threshold value included in the first startup sentence are compared. It is characterized by having a comparison unit and a derivation unit for deriving the recognition result based on the result of the comparison unit .

According to the first to seventh inventions, the receiving means receives the voice data as continuous signals at once. Further, the recognition means uses phoneme recognition to derive the recognition result of recognizing the content of the voice data. Therefore, the period from the start of receiving the voice data to the result of the voice recognition can be minimized. This makes it possible to reduce the recognition time of voice data transmitted by wireless communication.

Further, according to the first to seventh inventions, the transmitting means transmits voice data by wireless communication using the UHF band. Therefore, compared to wireless communication using a band with a high penetration rate such as Wi-Fi (registered trademark), the possibility of communication failure between the same bands is low. Further, even when it is used at the same time as wireless communication using Wi-Fi (registered trademark) or the like, wireless communication can be smoothly realized. This makes it possible to expand the applications.

In particular, according to the second invention, the transmitting means transmits voice data that has not been packetized. Therefore, when receiving voice data, it is not necessary to perform error detection for each packet, processing for replacing each packet in a different order from the reception timing, and the like. As a result, it is possible to easily reduce the recognition time of the voice data transmitted by wireless communication.

In particular, according to the first to seventh inventions, the designating means refers to the start-up database and designates the first start-up sentence corresponding to the candidate sentence. Therefore, by changing the content of the startup sentence stored in the startup database, the content of the derived recognition result can be easily changed. This makes it possible to easily realize customization according to the application.

In particular, according to the third invention, the plurality of activation sentences are one sentence used for deriving the first recognition result indicating the start of speech recognition and a plurality of sentences used for deriving a recognition result different from the first recognition result. Includes dummy sentences and. Therefore, the first recognition result can be preferentially generated for the voice data transmitted by wireless communication. This makes it possible to further improve the voice recognition accuracy.

In particular, according to the fourth invention, the driving device has a controlled unit that is controlled based on the recognition result derived by the recognition means. Therefore, by transmitting the voice data by one sound collecting device, it is possible to remotely control a plurality of driving devices. This makes it possible to improve work efficiency.

In particular, according to the fifth invention, the content of one sentence is different for each of the plurality of stored voice recognition devices. Therefore, the timing for controlling the controlled unit can be different for each drive device having the voice recognition device. As a result, when remote control is performed on a plurality of drive devices using one sound collecting device, it is possible to realize independent remote control for each drive device.

In particular, according to the sixth invention, the evaluation sound data transmitting means transmits a plurality of evaluation sound data by wireless communication. Further, the evaluation result transmitting means transmits the evaluation result by wireless communication. Therefore, the number of times data is transmitted and received when calibrating the sound collecting device or the like can be minimized. This makes it possible to significantly reduce the calibration adjustment time.

In particular, according to the seventh invention, the acquisition conditions include at least one of microphone gain, auto gain control, dynamic range control, and acoustic echo cancellation. Therefore, by performing the calibration means once, it is possible to easily identify the acquisition conditions suitable for various situations in which the voice recognition system is used. This makes it possible to suppress restrictions on the usage environment.

According to the eighth invention, the receiving unit receives the voice data as continuous signals at once. In addition, the recognition unit uses phoneme recognition to derive a recognition result that recognizes the content of voice data. Therefore, the period from the start of receiving the voice data to the result of the voice recognition can be minimized. This makes it possible to reduce the recognition time of voice data transmitted by wireless communication.

Further, according to the eighth invention, the receiving unit receives the voice data transmitted by wireless communication using the UHF band. Therefore, compared to wireless communication using a band with a high penetration rate such as Wi-Fi (registered trademark), the possibility of communication failure between the same bands is low. Further, even when it is used at the same time as wireless communication using Wi-Fi (registered trademark) or the like, wireless communication can be smoothly realized. This makes it possible to expand the applications.

FIG. 1 is a schematic diagram showing an example of the configuration of the voice recognition system according to the first embodiment. FIG. 2 is a schematic diagram showing an example of the operation of the voice recognition system according to the first embodiment. FIG. 3A is a schematic diagram showing an example of the configuration of the voice recognition device according to the first embodiment, and FIG. 3B is a schematic diagram showing an example of the function of the voice recognition device according to the first embodiment. is there. FIG. 4A is a schematic diagram showing an example of a sound collecting device according to the first embodiment, and FIG. 4B is a schematic diagram showing an example of a driving device according to the first embodiment. FIG. 5 is a flowchart showing an example of the operation of the voice recognition system according to the first embodiment. FIG. 6 is a schematic diagram showing an example of a startup database. FIG. 7A is a flowchart showing a first modification of the operation of the voice recognition system according to the first embodiment, and FIG. 7B is a schematic view showing a first modification of the recognition unit. FIG. 8 is a schematic diagram showing an example of a character string database, a grammar database, and a reference database. FIG. 9A is a flowchart showing a second modification of the operation of the voice recognition system in the first embodiment, and FIG. 9B is a schematic view showing a second modification of the recognition unit. FIG. 10 is a schematic view showing a second modification of the operation of the voice recognition system according to the first embodiment. FIG. 11 is a schematic diagram showing an example of the operation of the voice recognition system according to the second embodiment. FIG. 12A is a flowchart showing an example of the operation of the voice recognition system in the second embodiment, and FIG. 12B is a schematic diagram showing an example of the function of the voice recognition device in the second embodiment. .. FIG. 13 is a schematic view showing an example of a reference table.

Hereinafter, an example of the voice recognition system and the voice recognition device according to the embodiment of the present invention will be described with reference to the drawings.

(First Embodiment: Configuration of voice recognition system 100)
An example of the configuration of the voice recognition system 100 according to the first embodiment will be described with reference to FIGS. 1 and 2. FIG. 1 is a schematic diagram showing an example of the configuration of the voice recognition system 100 in the present embodiment, and FIG. 2 is a schematic diagram showing an example of the operation of the voice recognition system 100 in the present embodiment.

The voice recognition system 100 is used to recognize the user's voice even when the user is away from the device having the voice recognition function. The voice recognition system 100 can acquire and recognize voice data D based on the user's voice via wireless communication W.

The voice recognition system 100 may include, for example, a voice recognition device 1 and a sound collection device 2, and may include, for example, a drive device 3, as shown in FIG. In the voice recognition system 100, the voice recognition device 1 and the sound collection device 2 are connected by a wireless communication W using a UHF (Ultra High Frequency) band.

For example, one sound collecting device 2 is connected to a plurality of voice recognition devices 1 (voice recognition devices 1a and 1b in FIG. 1) by wireless communication W. Therefore, the user can transmit and recognize the voice data D based on the voice to one or more voice recognition devices 1 via the sound collecting device 2.

The voice recognition device 1 is provided at a place separated from the sound collecting device 2, for example, by about 100 m, and is provided at a place separated from, for example, about 10 to 30 m. Therefore, in the voice recognition system 100, it is possible to realize voice recognition even in an environment in which the user cannot directly emit voice to the voice recognition device 1.

The voice recognition device 1 is connected to one or more drive devices 3 via, for example, wired communication or wireless communication W. Therefore, the voice recognition device 1 can control the drive device 3 based on the recognition result for the voice data D. The drive device 3 may include, for example, a voice recognition device 1, and the voice recognition device 1 may be built in the drive device 3. Further, the voice recognition device 1 can independently perform voice recognition without being connected to a communication network such as the Internet.

The recognition result includes information for controlling the drive device 3, for example, text data that can be confirmed by the user, information related to voice for notifying (replying) to the user (voice information, utterance information), and the like. It may be.

As shown in FIG. 2, for example, the voice recognition system 100 acquires voice data D based on the user's voice via the sound collecting device 2. After that, the sound collecting device 2 transmits the voice data D to the voice recognition device 1 by, for example, the wireless communication W using the UHF band.

The voice recognition device 1 receives the voice data D as continuous signals at once. The voice recognition device 1 uses phoneme recognition to derive a recognition result that recognizes the content of the voice data D. Therefore, in the voice recognition system 100, the period from the start of receiving the voice data D to the result of the voice recognition can be minimized. This makes it possible to reduce the recognition time of the voice data D transmitted by the wireless communication W. For example, the voice recognition device 1 may control the drive device 3 based on the recognition result.

The voice recognition device 1 realizes the derivation of the recognition result of recognizing the contents of the voice data D by using a voice recognition engine including a known phoneme recognition technology such as Julius or a known general-purpose programming language such as Python. can do. The voice recognition device 1 extracts, for example, an array of phonemes (phoneme information) included in the received voice data D, and derives a recognition result based on the phoneme information.

The phoneme information includes a plurality of phoneme sequences (for example, "a / k / a / r / i", etc.) emitted by the user. Phonemes are known including vowels and consonants. Note that the phoneme information includes, for example, a silence section indicating the start of voice (for example, a start silence section indicated by "silB" or the like) and a silence section indicating the end of voice (for example, an end silence section indicated by "silE" or the like). At least any of them may be included. The start silence section and the end silence section can be extracted by a known phoneme recognition technique.

The phoneme information may include, for example, a pause section. The pause section indicates a section shorter than the start silence section and the end silence section, and indicates, for example, a section (length) similar to the phoneme section. The pause section can be extracted by a known phoneme recognition technique.

<Voice recognition device 1>
FIG. 3A is a schematic diagram showing an example of the configuration of the voice recognition device 1. As the voice recognition device 1, a single board computer such as Raspberry Pi (registered trademark) may be used, or a known electronic device such as a personal computer (PC) may be used. The voice recognition device 1 includes, for example, a housing 10, a CPU (Central Processing Unit) 101, a ROM (Read Only Memory) 102, a RAM (Random Access Memory) 103, a storage unit 104, and an I / F 105-108. For example, the communication device 111 may be provided. Each configuration 101 to 108 is connected by an internal bus 110.

The CPU 101 controls the entire voice recognition device 1. The ROM 102 stores the operation code of the CPU 101. The RAM 103 is a work area used during the operation of the CPU 101. The storage unit 104 stores various information such as a character string database. As the storage unit 104, for example, in addition to an SD memory card, a known data storage medium such as an HDD (Hard Disk Drive) or SSD (Solid State Drive) is used.

The I / F 105 is a known interface for transmitting and receiving various information to and from the communication device 111. The communication device 111 is used to perform wireless communication W with the sound collecting device 2 and the like. As the communication device 111, for example, a known communication control device that can use the UHF band is used. The communication device 111 may be retrofitted to the voice recognition device 1 or may be built into the voice recognition device 1.

The communication device 111 has, for example, an antenna and a receiving mechanism, and may have, for example, a transmitting mechanism. The receiving mechanism includes, for example, a digital demodulation unit that demodulates the modulated data and a decoding unit that decodes the encoded data. The transmission mechanism includes, for example, a conversion unit that converts voice into digital data (voice data D) and a communication unit that processes voice data D as needed. The communication unit includes a coding unit that encodes the voice data D and a modulation unit that modulates the voice data D. The modulation unit modulates the data by using a phase modulation method such as DQPSK (Differential Quadrature Phase-Shift Keying). The receiving unit 11 may perform at least a part of the functions of the communication device 111 described above.

The I / F 106 is a known interface for transmitting and receiving various information to and from the input unit 112 connected according to the application. For example, a keyboard is used as the input unit 112, and an administrator or the like who manages the voice recognition system 100 inputs or selects various information or a control command of the voice recognition device 1 via the input unit 112.

The I / F 107 is a known interface for transmitting and receiving various information to and from the display unit 113 connected according to the application. The display unit 113 outputs various information stored in the storage unit 104, the processing status of the voice recognition device 1, and the like. As the display unit 113, for example, a display is used, and for example, a touch panel type may be used. In this case, the display unit 113 may be configured to include the input unit 112.

The I / F 108 is a known interface for transmitting and receiving various information with an external device such as a drive device 3. A plurality of I / F 108s may be provided, for example, and may be used for transmitting and receiving various information via a communication network such as the Internet.

As I / F105 to I / F108, for example, the same one may be used, and as each I / F105 to I / F108, for example, a plurality of ones may be used. Further, at least one of the input unit 112 and the display unit 113 may be removed depending on the situation.

FIG. 3B is a schematic diagram showing an example of the function of the voice recognition device 1. The voice recognition device 1 includes a receiving unit 11, a recognition unit 12, a storage unit 13, and an output unit 14. Each function shown in FIG. 3B is realized by the CPU 101 executing a program stored in the storage unit 104 or the like using the RAM 103 as a work area.

<< Receiver 11 >>
The receiving unit 11 receives the voice data D via, for example, the communication device 111. The receiving unit 11 receives, for example, the voice data D transmitted by the wireless communication W using the UHF band as a continuous signal at one time. That is, the receiving unit 11 does not perform processing such as error detection for each received packet and sorting of each packet, as in a communication method using known wireless packet communication, for example. Therefore, the receiving unit 11 can minimize the communication period when receiving the voice data D.

<< Recognition unit 12 >>
The recognition unit 12 uses phoneme recognition to derive a recognition result that recognizes the content of the voice data D. The recognition unit 12 can be realized by using a voice recognition engine including a known phoneme recognition technique such as Julius, or a known general-purpose programming language such as Python.

<< Memory 13 >>
The storage unit 13 stores various information in the storage unit 104, or retrieves various information from the storage unit 104. The storage unit 13 stores or retrieves various types of information according to the processing contents of the receiving unit 11, the recognition unit 12, and the output unit 14, for example.

<< Output unit 14 >>
The output unit 14 outputs various information to the display unit 113 and the like. The output unit 14 outputs, for example, the recognition result derived by the recognition unit 12 to the display unit 113 or the like. The output unit 14 generates control information for controlling the drive device 3 based on, for example, the recognition result, and outputs the control information to the drive device 3. The output unit 14 may output the recognition result as control information to the drive device 3, for example.

<Sound collecting device 2>
FIG. 4A is a schematic view showing an example of the configuration of the sound collecting device 2. The sound collecting device 2 may include, for example, a microphone M, a control unit 21, a conversion unit 22, a communication unit 23, and an antenna A, and may include, for example, a storage unit 24. As the sound collecting device 2, an intercom (transceiver with a headset) is used, and a known sound collecting device having a wireless communication function is used.

<< Microphone M >>
The microphone M collects the user's voice. A known sound collecting device is used as the microphone M.

<< Control unit 21 >>
The control unit 21 controls the entire sound collecting device 2, and controls operations related to transmission / reception of voice data D, for example. The control unit 21 controls, for example, the modulation method of the voice data D to be transmitted, and also controls the timing of transmitting the voice data D and the like. As the control unit 21, a known processor such as a DSP (Digital Signal Processor) is used.

<< Conversion unit 22 >>
The conversion unit 22 converts the voice (analog signal) picked up by the microphone M into a digital signal (voice data D). The conversion unit 22 converts an analog signal into a digital signal and acquires it as voice data D by using a known technique such as PCM (Pulse Code Modulation). An A / D converter is used as the conversion unit 22.

<< Communication Unit 23 >>
The communication unit 23 transmits voice data D by wireless communication W using the UHF band. The communication unit 23 transmits the voice data D via, for example, the antenna A. The communication unit 23 includes, for example, a coding unit that encodes the voice data D and a modulation unit that modulates the voice data D. The modulation unit modulates the data by using a phase modulation method such as DQPSK.

The communication unit 23 transmits, for example, voice data D that has not been packetized. That is, the sound collecting device 2 does not perform packetization processing on the voice data D, for example. Therefore, the communication unit 23 can transmit the voice data D as a continuous signal at once. As a result, the receiving unit 11 of the voice recognition device 1 can receive the voice data D as continuous signals at once.

The sound collecting device 2 may have a receiving mechanism, for example, like the above-mentioned communication device 111. The conversion unit and the communication unit of the communication device 111 can correspond to the conversion unit 22 and the communication unit 23.

<< Memory unit 24 >>
The storage unit 24 stores various information in a storage unit (not shown), or retrieves various information from the storage unit. As the storage unit, for example, a known data storage medium such as an SD memory card is used as in the storage unit 104 described above.

<Drive device 3>
FIG. 4B is a schematic view showing an example of the configuration of the drive device 3. The drive device 3 may have, for example, a voice recognition device 1, and may have, for example, a control unit 31 and a controlled unit 32.

As the drive device 3, for example, home appliances such as lighting and air conditioners are used, and for example, a configuration having a driveable portion such as a movable bed, a door, a window, and a curtain is used. As the drive device 3, for example, a robot, an automated guided vehicle (AGV), a production machine, or the like may be used.

<< Control unit 31 >>
The control unit 31 controls the entire drive device 3. The control unit 31 controls the controlled unit 32 based on, for example, the recognition result acquired from the voice recognition device 1 or the control information generated based on the recognition result. As the control unit 31, for example, a processor such as a CPU is used.

<< Controlled Unit 32 >>
The controlled unit 32 is controlled by the control unit 31. The controlled unit 32 shows, for example, a configuration in which it is driven by using an electric signal, and may show, for example, a pneumatic mechanism, a hydraulic mechanism, a motor, or the like, or a control circuit or the like.

(First Embodiment: An example of the operation of the voice recognition system 100)
Next, an example of the operation of the voice recognition system 100 in the first embodiment will be described. FIG. 5 is a flowchart showing an example of the operation of the voice recognition system 100 in the present embodiment.

The voice recognition system 100 includes acquisition means S110, transmission means S120, reception means S130, and recognition means S140. The acquisition means S110 and the transmission means S120 can be carried out by using, for example, the sound collecting device 2, and the receiving means S130 and the recognition means S140 can be carried out by using, for example, the voice recognition device 1.

<Acquisition means S110>
The acquisition means S110 acquires voice data D based on the voice. For example, the microphone M of the sound collecting device 2 collects the user's voice. After that, for example, the conversion unit 22 converts the audio of the analog data into digital data and acquires it as the audio data D. The conversion unit 22 sequentially acquires the voice data D in accordance with the timing at which the voice is picked up by the microphone M, for example.

<Transmission means S120>
The transmission means S120 transmits voice data D by wireless communication W using the UHF band. For example, the communication unit 23 performs coding processing and modulation processing on the voice data D, and then transmits the voice data D to the voice recognition device 1 via the antenna A. The communication unit 23 sequentially transmits the voice data D, for example, at the timing when the conversion unit 22 acquires the voice data D. The communication unit 23 may transmit one voice data D to, for example, a plurality of voice recognition devices 1.

In the transmission means S120, for example, the communication unit 23 transmits the voice data D in a state where the voice data D is not packetized. Therefore, when receiving the voice data D, it is not necessary to perform an error detection for each packet or a process of replacing each packet in a different order from the reception timing.

When receiving voice data D using wireless packet communication such as Wi-Fi (registered trademark), some packets may be lost due to a problem such as an environment. At this time, the information included in the received voice data D is lost in packet units. Therefore, a loss in the amount of data corresponding to a period longer than the period in which the defect affects the voice data D may occur, which may affect the accuracy of voice recognition. Therefore, when the voice data D is transmitted and received using the wireless communication W, the voice recognition accuracy may be significantly reduced.

On the other hand, in the voice recognition system 100, the transmission means S120 transmits the voice data D without performing packetization processing on the voice data D. That is, when the voice data D is transmitted and received by the wireless communication W, the amount of data loss due to the defect corresponds to the period during which the defect affects the voice data D. Therefore, the amount of data loss due to a defect can be reduced as compared with the case where the voice data D is transmitted using wireless packet communication. As a result, it is possible to suppress a decrease in voice recognition accuracy in the voice data D transmitted by the wireless communication W.

<Receiving means S130>
The receiving means S130 receives the voice data D as a continuous signal at one time. For example, the receiving unit 11 of the voice recognition device 1 receives the voice data D via the communication device 111. The receiving unit 11 sequentially receives the voice data D, for example, at the timing when the communication unit 23 transmits the voice data D. The receiving unit 11 may perform at least one of demodulation and decoding of the voice data D, for example, via the communication device 111.

<Recognition means S140>
The recognition means S140 uses phoneme recognition to derive a recognition result that recognizes the content of the voice data D. For example, the recognition unit 12 uses a known voice recognition engine to derive a recognition result for the content of the voice data D. The recognition unit 12 derives the recognition result for the content of the voice data D, for example, according to the timing when the reception unit 11 receives the voice data D.

For example, when the voice data D includes a pause section, the recognition unit 12 derives a recognition result based on the pause section. Therefore, in the receiving means S130, by receiving the voice data D as continuous signals at once, the possibility of loss of the pause section can be suppressed. This makes it possible to improve the voice recognition accuracy.

In particular, in the transmission means S120, the voice recognition accuracy can be further improved by transmitting the voice data D in a state where the voice data D is not packetized. For example, when the conventional packetization processing is performed on the voice data D, the possibility that the pause section is lost is greatly increased. On the other hand, in the voice recognition system 100, the recognition means S140 derives the recognition result for the content of the voice data D that has not been packetized. Therefore, voice recognition can be realized while suppressing the loss of the pause section.

After that, for example, the recognition result is output as needed. For example, the output unit 14 may output the recognition result to the display unit 113 via the I / F 107, or may output the recognition result to the drive device 3 via the I / F 108.

As a result, the operation of the voice recognition system 100 ends.

(First Embodiment: First modification of voice recognition system 100)
Next, a first modification of the voice recognition system 100 according to the first embodiment will be described. In the first modification of the voice recognition system 100, the voice recognition device 1 refers to the activation database and derives the recognition result for the voice data D.

<Startup database>
FIG. 6 is a schematic diagram showing an example of a startup database. The startup database is stored in advance in the storage unit 104 of the voice recognition device 1.

The activation database is referred to when recognizing a keyword (wake word) required for starting voice recognition in the voice recognition system 100. The startup database may be stored in the storage unit 104 in a state of being compatible with, for example, a known phoneme recognition technique used.

A plurality of boot sentences acquired in advance are stored in the boot database. The activation sentence includes one or more character string information and a threshold value associated with each character string information. The activation sentence may include, for example, phoneme information associated with the character string information, and may include, for example, phoneme information instead of the character string information. In addition, one threshold value may be associated with each activation sentence, for example.

The activation sentence is used to estimate the content of the voice data D. For example, one activation sentence is selected for one voice data D.

As the character string information, an expression element having a meaning such as a word or a morpheme may be used, or a character string having no meaning may be used. In addition to Japanese, the character string information may include, for example, languages of two or more countries, and may include character strings such as numbers and abbreviations used in places of use.

As the phoneme information, for example, an array of phonemes corresponding to character string information is used. The phoneme information may include, for example, a pause section (“*” in FIG. 6). As the phoneme information, for example, an array of different phonemes may be associated with the same character string information.

The threshold value is used when comparing with the reliability associated with the voice data D. The reliability indicates a value calculated from the voice data D using phoneme recognition. The threshold value is set in advance, for example, in the range of 0 or more and 1 or less.

For example, the recognition information associated with each startup sentence may be stored in the startup database. In this case, by selecting one activation sentence for the voice data D, the recognition information associated with the selected activation sentence can be derived as the recognition result.

For example, a plurality of activation sentences may include one sentence (first activation sentence) and a plurality of dummy sentences. The first activation sentence is used to derive a recognition result (first recognition result) indicating the start of speech recognition. The dummy sentence is used to derive a recognition result different from the first recognition result.

For example, the first activation sentence is stored in the activation database in association with the information indicating that voice recognition is started (first recognition information). Therefore, when it is determined that the voice data D corresponds to the first activation sentence, the first recognition information is derived as the recognition result, and the voice recognition can be started.

For example, a dummy sentence is stored in the startup database in association with information indicating that speech recognition is not started (for example, "do nothing"). Therefore, when it is determined that the voice data D corresponds to the dummy sentence, the voice recognition can be prevented from starting.

Here, the inventor has discovered that the voice data D received by the above-mentioned wireless communication (wireless communication W using the UHF band) is easier to recognize the characteristics of voice than the voice data D acquired by wire or the like. did. Therefore, it is sufficient to use one activation sentence as the activation sentence used for estimating the voice data D, and it is not necessary to provide, for example, an activation sentence similar to one activation sentence. This makes it possible to reduce the amount of data to be stored. Further, by setting a plurality of dummy sentences for one first activation sentence, it is possible to increase the possibility that the first activation sentence is adopted. This makes it possible to suppress malfunction of the voice recognition device 1.

For example, the threshold value included in the first activation sentence may be set lower than the threshold value included in the dummy sentence. In this case, the first activation sentence can be preferentially adopted as the sentence corresponding to the voice data D.

For example, when the voice recognition system 100 includes a plurality of voice recognition devices 1, the content of one sentence (first activation sentence) is different for each of the plurality of stored voice recognition devices 1. Therefore, the timing of recognition can be different for each voice recognition device 1. Further, for example, the timing of controlling the controlled unit 32 can be different for each drive device 3 having the voice recognition device 1.

(First Embodiment: First modification of the operation of the voice recognition system 100)
Next, a first modification of the operation of the voice recognition system 100 in the first embodiment will be described. FIG. 7A is a flowchart showing a part of the first modification in the operation of the voice recognition system 100 in the present embodiment.

In the first modification of the voice recognition system 100, for example, the recognition means S140 has a generation means S141, a designation means S142, a comparison means S143, and a derivation means S144. In the first modification of the voice recognition system 100, for example, as shown in FIG. 7B, the recognition unit 12 may have a generation unit 12a, a designation unit 12b, and a comparison unit 12c.

<Generation means S141>
The generating means S141 is implemented after the receiving means S130 described above. In the generation means S141, for example, the generation unit 12a uses phoneme recognition to generate a candidate sentence corresponding to the content of the voice data D. The candidate sentence includes one or more candidate data and one or more reliability.

The candidate data indicates at least one of the phoneme information extracted from the voice data D by phoneme recognition and the character string corresponding to the phoneme information. The reliability indicates the accuracy (extraction accuracy) of the phoneme information extracted by the phoneme recognition in association with the candidate data. For example, the generation unit 12a calculates the reliability in the range of 0 or more and 1 or less. The generation unit 12a can generate a candidate sentence and calculate the reliability by using a known speech recognition engine such as Julius.

<Designation means S142>
In the designation means S142, for example, the designation unit 12b refers to the start-up database and designates a start-up sentence (first start-up sentence) corresponding to the candidate start-up sentence among the plurality of start-up sentences. For example, the designation unit 12b compares the candidate data included in the candidate sentence with the character string information or phoneme information included in each activation sentence, and designates the same or similar activation sentence. The designation unit 12b specifies, for example, an activation sentence that most closely resembles the candidate sentence.

<Comparison means S143>
In the comparison means S143, for example, the comparison unit 12c compares the reliability included in the candidate sentence with the threshold value (first threshold value) included in the designated start-up sentence. For example, when the reliability is equal to or higher than the threshold value, the comparison unit 12c determines that the candidate sentence corresponds to the activation sentence. For example, when the reliability is less than the threshold value, the comparison unit 12c determines that the candidate sentence does not correspond to the activation sentence, and may perform the designation means S142 again and specify another activation sentence, for example.

When the candidate sentence and the activation sentence include a plurality of reliabilitys, the comparison unit 12c compares each reliability with each threshold value. At this time, as a criterion for determining that the candidate sentence corresponds to the activation sentence, at least a part of the reliability may be equal to or higher than the threshold value, and can be arbitrarily set.

<Derivation means S144>
In the derivation means S144, for example, the recognition unit 12 derives the recognition result based on the result of the comparison means S143. For example, when the recognition information is stored in the startup database, the recognition unit 12 derives the recognition information associated with the startup sentence corresponding to the candidate sentence as the recognition result.

As a result, for example, when a recognition result indicating that voice recognition is started is derived, the voice recognition operation in the voice recognition system 100 is started, and for example, the above-mentioned acquisition means S110 or the like is implemented. For example, when a recognition result indicating that voice recognition is not started is derived, the acquisition means S110 or the like is executed, and then the designation means S142 or the like that refers to the startup database is executed again.

(First Embodiment: Second modification of the voice recognition system 100)
Next, a second modification of the voice recognition system 100 according to the first embodiment will be described. In the second modification of the voice recognition system 100, the voice recognition device 1 refers to the character string database and the grammar database, and derives the recognition result for the voice data D.

In the second modification of the voice recognition system 100, for example, the voice recognition device 1 refers to a character string database and a grammar database constructed according to the user's use, and derives a recognition result corresponding to the user's voice. To do. The voice recognition device 1 may refer to a reference database instead of the startup database described above, for example.

<Database>
FIG. 8 shows an example of a character string database, a grammar database, and a reference database. Each database is stored in advance in the storage unit 104 of the voice recognition device 1.

<< Character string database >>
The character string database stores a character string (character string information) that is expected to be emitted by the user and a phoneme (phoneme information) corresponding to the character string information. Therefore, by accumulating the above character strings and phonemes, it is possible to derive recognition according to the application, and it is possible to develop it for various applications.

By classifying the phoneme array (phoneme information) stored in the character string database based on the pause section included in the voice, the accuracy of deriving the recognition result for the voice data D received by the wireless communication W is dramatically improved. It is possible to improve. Further, in the voice recognition device 1, for example, the recognition result for the voice is derived only for the character string information stored in the character string database. Therefore, by changing the contents of the grammar database for each application, it is possible to derive a recognition result suitable for the application with high accuracy.

The character string database stores the character string information acquired in advance, the phonetic information associated with the character string information, and the class ID assigned to the character string information. The character string database is used, for example, when the detection unit 12e, which will be described later, detects candidate data.

The class ID indicates an arrangement location (for example, first class ID "1" or the like) in which words or the like of the character string information are assumed to be used grammatically in association with the character string information. For example, when the speech grammar (sentence) can be expressed as "target" + "particle" + "action", "1" is used for the character string information that is the "target" of the speech as the class ID. "2" is used for the character string information that becomes the "particle" of the voice, and "3" is used for the character string information that becomes the "action" of the voice.

The phoneme information stored in the character string database may indicate, for example, an array in which at least one consonant is removed from an array of phonemes corresponding to the character string information. The simple phoneme information is a basic phoneme array "a / k / a / r" corresponding to the character string information, such as "a / a / i" associated with the character string information "light" in FIG. An array of phonemes consisting only of vowels excluding consonants from "/ i" is shown. In addition, the simple phoneme information may indicate an array based on a combination of vowels and pause sections, such as "i / * / i / e" linked to the character string information "play", and consonants. An array excluding at least one consonant, such as "i / * / i / t / e" excluding one, may be shown.

The simple phoneme information may indicate, for example, an array in which at least one of a vowel corresponding to character string information, a voiced sound, a semi-voiced sound, a sokuon, and a yoon is combined. In this case, the simple phoneme information shows an array excluding consonants other than the above, for example, an array excluding unvoiced consonants corresponding to the character string information.

For example, a muddy sound indicates a phoneme "g / a" or the like corresponding to the character string "ga", a handakuon indicates a phoneme "p / a" or the like corresponding to the character string "pa", and a prompting sound indicates a character string "p / a" or the like. The phoneme "k / a / q / t / a" and the like corresponding to "win" are shown, and the phoneme "s / h / o" and the like corresponding to the character string "sho" are shown.

By including at least one of voiced sound, semi-voiced sound, sokuon, and yoon in the arrangement indicated by the simple phoneme information, it is possible to narrow down the target of voice recognition to characteristic voices and further improve the recognition accuracy. It will be possible. For example, by including at least one of voiced sound and semi-voiced sound in the array of simple phoneme information, voice recognition based on characteristics such as dialect can be realized. Further, for example, by including at least one of a sokuon and a yoon in the array of simple phoneme information, it is possible to suppress a decrease in recognition accuracy for a voice that is difficult to recognize.

Further, the phoneme information stored in the character string database may include phoneme candidate information indicating an array of two or more different phonemes associated with one character string information. Each phoneme candidate information is a single character string such as "k / e / s / i / t / e" and "e / i / e" linked to the character string information "erase" in FIG. Corresponds to the information and shows a different arrangement of phonemes. For example, the phonetic candidate information may include the above-mentioned simple phonetic information. In this case, it is possible to suppress a decrease in recognition accuracy even in a situation where a user who has difficulty in producing a consonant among a plurality of users is included.

For example, the phoneme array "k / e / s / i / t / e" in the standard language and the phoneme array "k / e / s / u / t / e" in the dialect are two phoneme candidate information. May be associated with one character string information. Further, different languages may be associated with one character string information as two phoneme candidate information.

Hereinafter, the phoneme information may include at least one of the above-mentioned simple phoneme information and phoneme candidate information, and the description thereof will be omitted as appropriate.

<< Grammar database >>
The grammar database stores the grammar information required to generate a sentence that combines character string information. The grammatical information includes a plurality of information indicating the arrangement order of the class IDs associated with each character string information. By referring to the grammar database, it is possible to easily combine each character string information after detecting the character string information based on the arrangement of phonemes classified based on the pause interval. As a result, it is possible to derive a recognition result in consideration of the grammar for speech. As a result, it becomes possible to realize highly accurate voice recognition based on the content of the voice emitted by the user or the like.

The grammar database stores grammar information indicating the sequence order of a plurality of class IDs acquired in advance. The grammar database is used, for example, when the generation unit 12a calculates the reliability. When, for example, the first grammatical information "1, 2, 3" is used as the grammatical information, a sentence indicating "object" + "particle" + "action" can be generated as a voice candidate. The grammar information includes a plurality of class ID arrangement orders such as the first grammar information "1, 2, 3", the second grammar information "4, 5, 6", and the third grammar information "2, 1, 3". ..

<< Reference database >>
The reference database stores the character string information acquired in advance, the reference sentence that combines the character strings, and the threshold value assigned to each character string information. For example, the phonetic information associated with the character string information is stored. May be good. The reference database is used in the same manner as the startup database described above. The reference database is used in place of the startup database when, for example, the designated means S142 or the like is implemented after starting voice recognition by referring to the startup database. The data capacity can be reduced by making the character string information and the phonetic element information stored in the reference database equal to, for example, the character string information and the phonetic element information stored in the character string database.

The phoneme information stored in the reference database may include, for example, at least one of the above-mentioned simple phoneme information and phoneme candidate information. In this case, for example, as shown in FIG. 8, the two phoneme candidate information "a / k / a / r / i" and "a / a / i" associated with one character string information "light" are different. In addition to classifying the sentences (see the first and third sentences in FIG. 8), the same sentences may be used. Further, for example, a different threshold value "0." For each of the two phoneme candidate information "h / i / * / i / t / e" and "i / * / i / e" linked to one character string information "play". In addition to setting "800" and "0.890", the same threshold value may be set.

In addition, for example, the threshold value associated with the simple phoneme information may be lowered so that the simple phoneme information is preferentially selected. In this case, a voice including a voiced sound, a semi-voiced sound, a sokuon, a yoon, etc. can be preferentially recognized as a characteristic voice.

(First Embodiment: Second modification of the operation of the voice recognition system 100)
Next, a second modification of the operation of the voice recognition system 100 in the first embodiment will be described. FIG. 9A is a flowchart showing a part of the second modification in the operation of the voice recognition system 100 in the present embodiment.

In the second modification of the voice recognition system 100, for example, the recognition means S140 has an extraction means S145 and a detection means S146. In the second modification of the voice recognition system 100, for example, as shown in FIG. 9B, the recognition unit 12 may have an extraction unit 12d and a detection unit 12e.

<Extraction means S145>
The extraction means S145 is implemented after the reception means S130 described above. In the extraction means S145, for example, the extraction unit 12d extracts the start silence section and the end silence section included in the voice data D by phoneme recognition. Further, the extraction unit 12d extracts an array of phonemes and pause sections sandwiched between the start silence section and the end silence section as recognition target data by phoneme recognition. That is, the basic function of the extraction unit 12d can be realized by using the phoneme recognition technique.

The extraction unit 12d extracts, for example, a non-utterance state (silence section) of 100 milliseconds or more and 1 second or less as a start silence section and an end silence section. The extraction unit 12d allocates phonemes and pause sections to the sections (voice sections) sandwiched between the start silence section and the end silence section. The extraction unit 12d extracts the array of the assigned phonemes and pause sections as the recognition target data.

The extraction unit 12d determines, for example, the length of each phoneme or the length of the entire recognition target data, sets the length of the pause section, and then uses an array to which the phonemes and the pause section are assigned as the recognition target data. It may be extracted. That is, the extraction unit 12d may set the length of the pause section according to the length of the phoneme or the length of the entire recognition target data.

As shown in the extraction means S145 of FIG. 10, the extraction unit 12d extracts the start silence section “silB” and the end silence section “silE”, and arranges them in the voice section “a / k / a / r / i / *”. / w / o / * / ts / u / k / e / t / e ”(* indicates a pause section) is extracted as target recognition data. The extraction unit 12d may, for example, extract a plurality of target recognition data having different sequences from one voice data D. In this case, it is possible to carry out voice recognition in consideration of variations due to the allocation of phonemes and pause sections in the extraction unit 12d. For example, the extraction unit 12d can improve the recognition accuracy while suppressing the processing time by extracting one or more and five or less target recognition data. The extraction unit 12d may extract, for example, an array including at least one of a start silence section and an end silence section as target recognition data.

The rest interval may include, for example, at least one of breath sounds and lip noise. That is, the extraction unit 12d may extract, for example, at least one of the breath sounds and the lip noise included in the pause section as the recognition target data. In this case, by including at least one of breath sounds and lip noise in the phoneme information stored in the character string database or the like described later, it is possible to derive a more accurate recognition result.

<Detecting means S146>
In the detection means S146, for example, the detection unit 12e refers to the character string database and selects phoneme information corresponding to the phoneme arrangement of the recognition target data. Further, the detection unit 12e detects a plurality of character string information and class IDs associated with the selected phoneme information as candidate data.

As shown in FIG. 10, for example, the detection unit 12e has phoneme information "a / k / a / r / i", "w / o", and "ts / u / k / e / t /" corresponding to the recognition target data. Select "e" and detect the character string information associated with each phoneme information and the class IDs "light / 1", "o / 2", and "attach / 3" as candidate data, respectively. At this time, the number of candidate data increases according to the number of recognition target data. The arrangement of each phoneme may be classified in advance by dividing it into pause sections, or may be classified based on phoneme information including phonemes and pause sections.

<Modification example of generation means S141>
After the detection means S146, for example, the generation means S141 is implemented. The generation means S141 can generate a candidate sentence by referring to the grammar database, for example, in addition to the contents described above.

In the generation means S141, for example, the generation unit 12a refers to the grammar database and generates a sentence in which a plurality of candidate data are combined based on the grammar information. In addition, the generation unit 12a calculates the reliability of the character string information for each candidate data included in the sentence by using the grammar database. To calculate the reliability, for example, a stack decoding search with reference to a grammar database is performed.

As shown in FIG. 10, for example, the generation unit 12a attaches each candidate data “light / 1”, “o / 2”, and “attach” for each class ID included in the first grammatical information “1, 2, 3”. Corresponds to the class ID of "/ 3" and generates sentences "light / 1", "to / 2", and "attach / 3". At this time, for example, when the grammatical information is "3, 1, 2", "Turn on / 3", "Light / 1", and "O / 2" are generated as sentences.

The generation unit 12a has reliabilitys "0.982", "1.000", and "0" corresponding to each candidate data "light / 1", "o / 2", and "attach / 3" included in the sentence. .990 "is calculated. The generation unit 12a calculates the reliability of the character string information of each candidate data in the range of 0 or more and 1 or less.

The generation unit 12a can generate a plurality of sentences according to, for example, the type of grammar information in the grammar database. Further, the generation unit 12a can perform voice recognition suitable for the situation with high accuracy by selecting the type of grammatical information.

For example, the generation unit 12a may set ranks indicating priorities for each sentence (ranks 1 to 5 in FIG. 10). By setting the rank, sentences ranked at any lower rank (for example, rank 6 or lower) can be excluded from the evaluation target. Therefore, the number of candidate data selected as the evaluation data described later can be reduced, and the processing speed can be improved.

For example, when the same candidate data is included in sentences having different contents, the generation unit 12a may calculate different reliability for each candidate data. For example, the reliability "0.982", "1.000", "0.990" corresponding to each candidate data "light / 1", "o / 2", and "attach / 3" included in the first sentence. Is calculated, the reliability corresponding to each candidate data "light / 1", "o / 2", and "play / 3" included in the second sentence is "0.942", "1.000". , "0.023" is calculated. That is, even if the same candidate data "light" is used, different reliability may be calculated depending on the content of the sentence and the order of combinations.

For example, as the reliability, a preset value may be used for each character string information. In this case, the preset value is stored in the grammar database, for example. In addition to the above, for example, as the reliability, a relative value according to the type and number of candidate data (or character string information of the candidate data) detected by the detection unit 12e may be used. For example, in a plurality of candidate data detected by the detection unit 12e, as the number of types of character string information to which one class ID is assigned increases, a low reliability may be calculated for each character string information. ..

After that, for example, the generation unit 12a selects one or more candidate data from a plurality of candidate data based on the reliability, and generates a candidate sentence including the reliability associated with the selected candidate data. The generation unit 12a selects, for example, the candidate data for which the highest reliability is calculated for each class ID from a plurality of candidate data, and generates it as a candidate sentence. For example, the generation unit 12a has the highest reliability of the candidate data "attach / 3 / 0.990" and "play / 3 / 0.023" in the same class ID "3". Select "/ 3 / 0.990" and generate it as a candidate sentence.

After that, the above-mentioned designation means S142 and the like are implemented (for example, the designation means S142, the comparison means S143, and the derivation means S144 shown in FIG. 7A and FIG. 10). In the designating means S142 and the like, the reference database or the startup database is referred to depending on the situation.

In the generation means S141, for example, after generating a candidate sentence, the recognition result may be derived based on the candidate sentence. In that case, the designation means S142 and the like need not be implemented.

For example, the generation unit 12a converts the candidate sentence into a data format such as text or numerical value and derives it as a recognition result, and also converts the candidate sentence into a voice data format or a control data format for controlling the drive device 3, for example. However, it may be derived as a recognition result. As a method of converting to a data format such as text or numerical value, a voice data format, or a control data format based on a candidate sentence, a known technique can be used, and a database or the like accumulating each data format can be used as needed. You may use it.

According to the present embodiment, the receiving means S130 receives the voice data D as continuous signals at once. Further, the recognition means S140 uses phoneme recognition to derive the recognition result of recognizing the content of the voice data D. Therefore, the period from the start of receiving the voice data D to the result of the voice recognition can be minimized. This makes it possible to reduce the recognition time of the voice data D transmitted by the wireless communication W.

Further, according to the present embodiment, the transmission means S120 transmits the voice data D by the wireless communication W using the UHF band. Therefore, compared to wireless communication using a band with a high penetration rate such as Wi-Fi (registered trademark), the possibility of communication failure between the same bands is low. Further, even when it is used at the same time as wireless communication using Wi-Fi (registered trademark) or the like, wireless communication W can be smoothly realized. This makes it possible to expand the applications.

In addition to the above, by using the wireless communication W using the UHF band, the authentication cost is not required and the protocol stack is not required as compared with Wi-Fi (registered trademark) and Bluetooth (registered trademark). Therefore, the processing load can be reduced and the real-time performance can be improved.

Further, according to the present embodiment, the transmission means S120 transmits the voice data D that has not been packetized. Therefore, when receiving the voice data D, it is not necessary to perform error detection for each packet, processing for replacing each packet in a different order from the reception timing, and the like. This makes it possible to easily reduce the recognition time of the voice data D transmitted by the wireless communication W.

Further, according to the present embodiment, the designating means S142 refers to the startup database and designates the first startup sentence corresponding to the candidate sentence. Therefore, by changing the content of the startup sentence stored in the startup database, the content of the derived recognition result can be easily changed. This makes it possible to easily realize customization according to the application.

Further, according to the present embodiment, the plurality of activation sentences are one sentence used for deriving a first recognition result indicating the start of speech recognition and a plurality of sentences used for deriving a recognition result different from the first recognition result. Includes dummy sentences and. Therefore, the first recognition result can be preferentially generated for the voice data D transmitted by the wireless communication W. This makes it possible to further improve the voice recognition accuracy.

Further, according to the present embodiment, the drive device 3 has a controlled unit 32 controlled based on the recognition result derived by the recognition means S140. Therefore, by transmitting the voice data D by one sound collecting device 2, it is possible to remotely control the plurality of driving devices 3. This makes it possible to improve work efficiency.

Further, according to the present embodiment, the content of one sentence is different for each of the plurality of stored voice recognition devices 1. Therefore, the timing for controlling the controlled unit 32 can be different for each drive device 3 having the voice recognition device 1. As a result, when remote control is performed on a plurality of drive devices 3 using one sound collecting device 2, it is possible to realize independent remote control for each drive device 3.

Further, according to the present embodiment, the receiving unit 11 receives the voice data D as continuous signals at once. Further, the recognition unit 12 uses phoneme recognition to derive the recognition result of recognizing the content of the voice data D. Therefore, the period from the start of receiving the voice data D to the result of the voice recognition can be minimized. This makes it possible to reduce the recognition time of the voice data D transmitted by the wireless communication W.

Further, according to the present embodiment, the receiving unit 11 receives the voice data D transmitted by the wireless communication W using the UHF band. Therefore, compared to wireless communication using a band with a high penetration rate such as Wi-Fi (registered trademark), the possibility of communication failure between the same bands is low. Further, even when it is used at the same time as wireless communication using Wi-Fi (registered trademark) or the like, wireless communication W can be smoothly realized. This makes it possible to expand the applications.

(Second embodiment: voice recognition system 100)
Next, an example of the voice recognition system 100 according to the second embodiment will be described. The difference between the above-described embodiment and the second embodiment is that the voice data D is calibrated. The description of the same contents as those of the above-described embodiment will be omitted.

In the voice recognition system 100, for example, as shown in FIG. 11, the sound for calibration is picked up via the sound collecting device 2, and a plurality of evaluation sound data Das are generated. At this time, in addition to generating evaluation sound data Da for each of a plurality of calibration sounds, for example, a plurality of evaluation sound data Das may be generated for one calibration sound. After that, the sound collecting device 2 transmits a plurality of evaluation sound data Das to the voice recognition device 1 by, for example, wireless communication W using the UHF band.

The voice recognition device 1 receives a plurality of evaluation sound data Da. The voice recognition device 1 recognizes the content of each evaluation sound data Da by using phoneme recognition, and derives an evaluation result R based on each recognition result. After that, the voice recognition device 1 transmits the evaluation result R to the sound collecting device 2 by, for example, wireless communication W using the UHF band.

The sound collecting device 2 receives the evaluation result R and sets the setting conditions based on the evaluation result R. Therefore, in the voice recognition system 100, the evaluation results R of the plurality of evaluation sound data Das required for calibration can be transmitted and received by one wireless communication W. As a result, the evaluation result for one evaluation sound data as in the conventional case is received, the setting is changed based on the evaluation result, one evaluation sound data is transmitted again, the evaluation result is received, and the setting is changed. Compared with the case of repeating the operation of ,, it is possible to significantly reduce the adjustment time of transmission and calibration.

(Second Embodiment: Example of operation of voice recognition system 100)
Next, an example of the operation of the voice recognition system 100 in the second embodiment will be described. FIG. 12A is a flowchart showing a part of an example in the operation of the voice recognition system 100 according to the present embodiment.

The voice recognition system 100 further includes a calibration means S200 in addition to the means of the above-described embodiment. The calibration means S200 is mainly performed before the acquisition means S110 described above, and can lead to an improvement in voice recognition accuracy.

The calibration means S200 includes a sound collecting means S210, an evaluation sound data generating means S220, an evaluation sound data transmitting means S230, an evaluation sound data receiving means S240, an evaluation means S250, and an evaluation result transmitting means S260. , And setting means S270. The voice recognition device 1 may further include an evaluation unit 15 in addition to the configurations of the above-described embodiments, for example, as shown in FIG. 12 (b).

<Sound collecting means S210>
In the sound collecting means S210, for example, the microphone M of the sound collecting device 2 picks up the sound for calibration. The microphone M may pick up the user's sound as a calibration sound, or may, for example, pick up a pre-recorded sound or a calibration sound source as a calibration sound.

<Evaluation sound data generation means S220>
In the evaluation sound data generation means S220, for example, the conversion unit 22 generates a plurality of evaluation sound data Da based on the calibration sound. The conversion unit 22 converts the sound for calibration of analog data into digital data and generates it as evaluation sound data Da.

The conversion unit 22 generates a plurality of evaluation sound data Das by converting a plurality of different calibration sounds picked up by the sound collecting means S210 into digital data. In this case, for example, different conversion conditions may be set in advance by the user or the like for the sound of each calibration.

In addition to the above, the conversion unit 22 refers to, for example, a reference table as shown in FIG. 13 and generates a plurality of evaluation sound data Das having different digital data conversion conditions for one calibration sound. You may. In addition, instead of the reference table, any conversion condition saved in advance by the user or the like may be referred to.

In the reference table, conversion conditions from analog data to digital data (acquisition conditions for audio data D) are stored in advance. Known conversion parameters are used as conversion conditions, for example, as shown in FIG. 13, microphone gain (MIC Gain), auto gain control (AGC: Auto Gain Control), dynamic range control (DRC: Dynamic Range Control), and A plurality of sets (No.) in which at least one of the conditions of Acoustic Echo Cancellation (AEC) is included and each condition is set are stored in the reference table, for example. Therefore, the conversion unit 22 can generate a plurality of evaluation sound data Das having different conversion conditions for one calibration sound by referring to the reference table.

<Evaluation sound data transmission means S230>
In the evaluation sound data transmission means S230, for example, the communication unit 23 transmits a plurality of evaluation sound data Das by, for example, wireless communication W using the UHF band. For example, the communication unit 23 performs coding processing and modulation processing on the plurality of evaluation sound data Das, and then transmits the plurality of evaluation sound data Das to the voice recognition device 1 via the antenna A. To do.

For example, the communication unit 23 transmits the evaluation sound data Da without packetizing the plurality of evaluation sound data Das. Therefore, when receiving a plurality of evaluation sound data Das, it is not necessary to perform error detection for each packet or processing for replacing each packet in a different order from the reception timing.

<Evaluation sound data receiving means S240>
In the evaluation sound data receiving means S240, for example, the receiving unit 11 of the voice recognition device 1 receives a plurality of evaluation sound data Da via the communication device 111. The receiving unit 11 sequentially receives a plurality of evaluation sound data Das, for example, in accordance with the timing at which the communication unit 23 transmits the plurality of evaluation sound data Das. The receiving unit 11 may perform at least one of demodulation and decoding of the evaluation sound data Da via, for example, the communication device 111.

<Evaluation means S250>
In the evaluation means S250, for example, the evaluation unit 15 uses phoneme recognition to generate evaluation results R for a plurality of evaluation sound data Da. For example, the evaluation unit 15 evaluates the contents of a plurality of evaluation sound data Das using a known voice recognition engine, and selects and selects the evaluation sound data Das for which the most accurate evaluation is obtained. The evaluation result R including the information of the evaluation sound data Da (for example, the first data) is generated.

<Evaluation result transmission means S260>
In the evaluation result transmitting means S260, for example, the output unit 14 transmits the evaluation result R via the communication device 111. The output unit 14 transmits the evaluation result R to the sound collecting device 2 by, for example, the wireless communication W using the UHF band. For example, the output unit 14 executes the coding process and the modulation process on the evaluation result R, and then transmits the evaluation result R to the sound collecting device 2.

<Setting means S270>
In the setting means S270, for example, the communication unit 23 of the sound collecting device 2 receives the evaluation result R. The sound collecting device 2 receives the evaluation result R via, for example, the antenna A, and demodulates and decodes the evaluation result R.

After that, for example, the control unit 21 sets the acquisition condition of the voice data D based on the evaluation result R. The control unit 21 selects, for example, the evaluation sound data Da associated with the first data included in the evaluation result R from among the plurality of transmitted evaluation sound data Das. After that, the control unit 21 can set the acquisition condition of the voice data D by specifying the conversion condition of the digital data used for generating the selected evaluation sound data Da. At this time, the control unit 21 may refer to the reference table for setting, or may refer to, for example, any preset parameter.

After that, the above-mentioned acquisition means S110 and the like are implemented, and the voice recognition system 100 ends.

According to this embodiment, it is possible to obtain the same effect as that of the above-described embodiment.

Further, according to the present embodiment, the evaluation sound data transmission means S230 transmits a plurality of evaluation sound data Das by wireless communication W. Further, the evaluation result transmitting means S260 transmits the evaluation result R by the wireless communication W. Therefore, the number of times data is transmitted and received when calibrating the sound collecting device 2 and the like can be minimized. This makes it possible to significantly reduce the calibration adjustment time.

Further, according to the present embodiment, the acquisition condition includes at least one of the conditions of microphone gain, auto gain control, dynamic range control, and acoustic echo cancellation. Therefore, by performing the calibration means S200 once, it is possible to easily specify the acquisition conditions suitable for various situations in which the voice recognition system 100 is used. This makes it possible to suppress restrictions on the usage environment.

Although the embodiments of the present invention have been described, the above-described embodiments are presented as examples and are not intended to limit the scope of the invention. The novel embodiment described above can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the gist of the invention. The above-described embodiments and modifications thereof are included in the scope and gist of the invention, and are also included in the scope of the invention described in the claims and the equivalent scope thereof.

1: Voice recognition device 10: Housing 11: Reception unit 12: Recognition unit 13: Storage unit 14: Output unit 101: CPU
102: ROM
103: RAM
104: Preservation unit 105: I / F
106: I / F
107: I / F
108: I / F
110: Internal bus 111: Communication device 112: Input unit 113: Display unit 2: Sound collecting device 21: Control unit 22: Conversion unit 23: Communication unit 24: Storage unit 3: Drive device 31: Control unit 32: Controlled unit 100: Voice recognition system A: Antenna M: Microphone S110: Acquisition means S120: Transmission means S130: Reception means S140: Recognition means W: Wireless communication

Claims (8)

A voice recognition system that uses wireless communication
An acquisition method for acquiring voice data based on voice,
A transmission means for transmitting the voice data by wireless communication using the UHF band, and
A receiving means for receiving the voice data as a continuous signal at once, and
A recognition means that uses phoneme recognition to derive a recognition result that recognizes the content of the voice data,
A start-up database that stores a plurality of start-up sentences including character string information and a threshold value associated with the character string information of 1 or more, and a start-up database.
Equipped with a,
The recognition means
A generation means that uses the phoneme recognition to generate candidate data corresponding to the content of the voice data and a candidate sentence including one or more reliability associated with the candidate data.
A designation means for designating the first startup sentence corresponding to the candidate sentence among the plurality of startup sentences by referring to the startup database.
A comparison means for comparing the reliability with the first threshold value included in the first activation sentence, and
A derivation means for deriving the recognition result based on the result of the comparison means,
A voice recognition system characterized by having . The voice recognition system according to claim 1, wherein the transmission means transmits the voice data that has not been packetized. The plurality of said activation sentences
One sentence used to derive the first recognition result indicating the start of speech recognition,
The voice recognition system according to claim 1 or 2, wherein a plurality of dummy sentences used for deriving the recognition result different from the first recognition result are included. One sound collecting device and
Multiple drive devices with voice recognition devices and
With more
The sound collecting device is
The conversion unit that performs the acquisition means and
The communication unit that performs the transmission means and
Have,
The voice recognition device is
The receiving unit that performs the receiving means and
A recognition unit that performs the recognition means and
The storage unit in which the startup database is stored and
Have,
The driving device, a speech recognition system according to claim 1, characterized in that it has a controlled unit that is controlled on the basis of the derived recognition result by the recognizing means. The plurality of said activation sentences
One sentence used for the first recognition result indicating the start of speech recognition,
It includes a plurality of dummy sentences used for the recognition result different from the first recognition result.
The voice recognition system according to claim 4 , wherein the content of the one sentence is different for each of the plurality of stored voice recognition devices. Further provided with a calibration means for calibrating the voice data,
The calibration means is
Evaluation sound data generation means for generating a plurality of evaluation sound data based on calibration sound, and
Evaluation sound data transmitting means for transmitting a plurality of the evaluation sound data by the wireless communication, and
Evaluation sound data receiving means for receiving a plurality of the evaluation sound data, and
An evaluation means that uses the phoneme recognition to generate evaluation results for a plurality of the evaluation sound data,
An evaluation result transmitting means for transmitting the evaluation result by the wireless communication,
The voice recognition system according to any one of claims 1 to 5 , further comprising a setting means for receiving the evaluation result and setting acquisition conditions for the voice data based on the evaluation result. The voice recognition system according to claim 6 , wherein the acquisition condition includes at least one of a microphone gain, an auto gain control, a dynamic range control, and an acoustic echo cancellation condition. A voice recognition device that uses wireless communication
A receiver that receives voice data transmitted by wireless communication using the UHF band as continuous signals at once, and
A recognition unit that uses phoneme recognition to derive a recognition result that recognizes the contents of the voice data,
A start-up database in which a plurality of start-up sentences including character string information and one or more threshold values associated with the character string information are stored, and
Equipped with a,
The recognition unit
A generation unit that uses the phoneme recognition to generate candidate data corresponding to the content of the voice data and a candidate sentence including one or more reliability associated with the candidate data.
A designation unit that refers to the startup database and specifies the first startup sentence corresponding to the candidate sentence among the plurality of startup sentences.
A comparison unit that compares the reliability with the first threshold value included in the first activation sentence.
A derivation unit that derives the recognition result based on the result of the comparison unit,
A voice recognition device characterized by having .

Images