JP2009122598A - Electronic device, control method of electronic device, speech recognition device, speech recognition method and speech recognition program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To recognize a whole speech part without missing a head part of the speech part which is required for speech recognition. <P>SOLUTION: When it is detected by a detecting means 9 that a sound volume level of sound data 4 exceeds a threshold sound volume level, a speech recognition means 11 discriminates the speech part from the recorded speech data 4, by going back to time of a predetermined period Tw2 before the detection time Tp by the detecting means 9, in the recorded sound data 4 in an overwrite record means 7, and performs speech recognition of the speech part. A control means 13 controls operation based on a control content expressed by the speech part which is discriminated by the speech recognition means 11. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to an electronic device or the like that recognizes sound from ambient environmental sounds.

  In recent electronic devices, there is a form in which a user gives a desired operation content from the outside by voice, and performs an operation according to the desired operation content recognized by recognizing the voice. As such an electronic device, in order to start such voice recognition, for example, a method triggered by an operator operating the utterance button may be considered, but the utterance button is operated every time the operation is desired. If necessary, the willingness to use may be reduced.

  Therefore, in a conventional electronic device, for example, it may be possible to always enter a standby state for voice recognition. However, in this case, resources such as a CPU (Central Processing Unit) and a DSP (Digital Signal Processor) are unnecessarily occupied. The use efficiency of will become worse.

  Therefore, in a conventional electronic device, the voice recognition device adopts a technology that recognizes the user's voice from the collected sounds and recognizes the contents of the voice when the voice signal reaches a predetermined threshold. (See Patent Document 1).

  Specifically, the conventional speech recognition apparatus has the following level detection unit and speech recognition unit. For example, the level detection unit detects whether or not an input signal of sound collected by a microphone is equal to or higher than a predetermined level. In addition, when the level detection unit detects that the input signal is equal to or higher than the predetermined level and is a voice, the voice recognition unit recognizes the voice included in the sound and performs voice recognition.

  At this time, when the voice recognition unit starts voice recognition for the sound after being detected by the level detection unit, the first part is lost from the target of voice recognition (hereinafter referred to as “head cut”). It is not possible to correctly recognize what the voice part represents as a whole.

  Therefore, the conventional speech recognition apparatus further includes the following delay unit in addition to the above-described configuration. This delay unit is provided between the microphone that collects the voice and the voice recognition unit. For example, the input signal itself is delayed when the start signal indicating the start of the voice part is input, and the input signal is cut off. A technique to prevent this is adopted.

Japanese Patent Laid-Open No. 5-27795 (paragraph number 0012, FIG. 2 (b))

  According to the above prior art, although it is possible to prevent the head of the input signal from being cut off, the start of the speech recognition processing itself is naturally delayed as the input signal is delayed. If the voice recognition process is delayed in this way, an operator who tries to operate the electronic device will have some time to wait until the electronic device actually starts operating after making a voice. A sense of incongruity was produced, and as a result, the operability was not good.

  The problem to be solved by the present invention includes the above-described problem as an example.

  In order to solve the above problem, the invention according to claim 1 is characterized in that overwriting recording means for continuously overwriting and recording sound data based on the collected sound, and the volume level of the sound data exceeds a threshold volume level. Detection means for detecting this, and when the detection means detects that the volume level of the sound data exceeds the threshold volume level, detection by the detection means among the sound data recorded in the overwrite recording means A voice recognition unit that identifies a voice part from recorded sound data before a predetermined time before the time and performs voice recognition on the voice part, and a control content represented by the voice part identified by the voice recognition unit Control means for controlling the operation based on the control means.

  In order to solve the above-mentioned problem, the invention according to claim 7 is characterized in that the sound data based on the collected sound is continuously overwritten and recorded on the overwriting recording means, and the sound data volume level is a threshold volume level. And when detecting that the volume level of the sound data exceeds the threshold volume level in the detection step, the sound data recorded in the overwrite recording means A voice recognition step of identifying a voice part from recorded sound data retroactively before a detection time in the detection step and performing voice recognition on the voice part; and the voice identified in the voice recognition step And a control step for controlling the operation based on the control content represented by the portion.

  In order to solve the above-mentioned problem, the invention according to claim 8 is characterized in that overwriting recording means for continuously overwriting recording sound data based on the collected sound, and the volume level of the sound data exceeds a threshold volume level. Detection means for detecting this, and when the detection means detects that the volume level of the sound data exceeds the threshold volume level, detection by the detection means among the sound data recorded in the overwrite recording means Voice recognition means for identifying a voice part from recorded sound data by going back a predetermined time before the time and performing voice recognition on the voice part.

  In order to solve the above-mentioned problem, the invention according to claim 9 is characterized in that the sound data based on the collected sound is continuously overwritten and recorded on the overwriting recording means, and the sound data volume level is a threshold volume level. And when detecting that the volume level of the sound data exceeds the threshold volume level in the detection step, the sound data recorded in the overwrite recording means A voice recognition step of identifying a voice part from recorded sound data retroactive to a predetermined time before the detection time in the detection step and performing voice recognition on the voice part.

  In order to solve the above-mentioned problem, the invention according to claim 10 is characterized in that the sound data based on the collected sound is continuously overwritten and recorded in the overwrite recording means, and the sound data volume level is a threshold volume level. And when detecting that the volume level of the sound data exceeds the threshold volume level in the detection step, the sound data recorded in the overwrite recording means A computer recognizes a voice part from recorded sound data by a predetermined time before the detection time in the detection step, and performs voice recognition on the voice part.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
<First Embodiment>
FIG. 1 is a block diagram illustrating a configuration example of a robot 100 as an example to which the electronic device according to the first embodiment is applied. Note that the electronic apparatus can be applied to various devices that operate in response to a user's voice even if the electronic apparatus is other than the robot 100.

  The robot 100 is mounted on, for example, a vehicle, and greets when a user who is a driver or a passenger gets in, or responds lovably in response to a user's call, thereby contributing to, for example, a driver's safe driving or driving It has a function to relieve fatigue and anxiety of passengers and passengers.

  The robot 100 recognizes an operation content instructed by a user's voice and performs an operation according to the operation content. For example, when a name is given to the robot 100, the user calls the name of the robot 100 to perform a predetermined operation, and a voice part as a command word for instructing a desired operation, such a voice. Spoken sentences, conversations, etc. including parts.

  The robot 100 includes a voice recognition device 1 as follows. The voice recognition device 1 is not limited to a mode built in the robot 100 but may be a mode that performs voice recognition as an independent device. The voice recognition device 1 includes a digitizing unit 5, an overwrite memory 7, a detection unit 9, and a voice recognition unit 11, and may further include a microphone 3.

  The microphone 3 corresponds to sound collection means and has a function of collecting sound generated in the surrounding environment and outputting sound data 4. In addition, this microphone 3 may be not only a built-in form but also a removable form. In this case, the sound recognition device 1 is configured to receive sound data 4 based on ambient sounds collected by the microphone 3. In the present embodiment, it is assumed that the sound data 4 output from the microphone 3 is analog data, for example.

  The digitizing unit 5 is connected to the microphone 3 and the overwriting memory 16, converts the sound data 4 that is analog data into sound data 4 of digital data, and outputs the sound data 4 to the overwriting memory 7. The overwrite memory 7 is an information recording medium on which the sound data 4 from the digitizing unit 5 is repeatedly overwritten and recorded in predetermined buffering units.

  The detection unit 9 corresponds to detection means, and has a function of detecting that the volume level of the specific frequency band of the sound data 4 described above exceeds a prescribed threshold volume level. Specifically, the detection unit 9 outputs a trigger TG to the voice recognition unit 11 when the volume level of the sound data 4 exceeds a prescribed threshold volume level.

  Here, the detection unit 9 detects whether or not the volume level of the specific frequency band of the sound data 4 exceeds a predetermined threshold volume level in consideration of the noise level included in the collected sound. Here, this specific frequency band represents the voice frequency band of the user. In the following description, the volume level of the specific frequency band of the sound data 4 is expressed as “the volume level of the sound data 4” unless otherwise required.

  The detection unit 9 is connected to a vehicle sensor (not shown), and constantly acquires vehicle information such as the engine speed of the vehicle from the vehicle sensor when the engine is operating.

  The voice recognition unit 11 corresponds to a voice recognition unit, and when the detection unit 9 detects that the volume level of the specific frequency band of the sound data 4 exceeds such a threshold value, that is, the trigger described above from the detection unit 9. For example, the time when the trigger TG is received is managed and the following operation is performed. In the present embodiment, the time when the voice recognition unit 11 receives the trigger TG from the detection unit 9 is called “detection time”.

  That is, the voice recognizing unit 11 identifies a voice part from the recorded sound data 4 retroactively before the detection time by the detecting unit 9 among the sound data 4 recorded in the overwrite memory 7, and the voice part is identified. Perform voice recognition. In addition, the voice recognition unit 11 stops the voice recognition process when the volume level of the sound data 4 that has risen falls to the noise level and a predetermined time elapses. Here, the noise level represents a volume level of a sound that does not include a sound that is constantly generated around the robot 100.

  The control part 10 mentioned above controls operation | movement based on the control content which the audio | voice part identified by the audio | voice recognition part 11 represents. This control part 10 controls operation | movement so that it may react with respect to the user who emitted the audio | voice part according to the control content. That is, the control unit 10 controls the operation of the robot 100 as will be described later according to the control content grasped by the voice recognition.

  The robot 100 incorporating the voice recognition device 1 is one example of the configuration as described above. Next, an example of a method for controlling the robot 100 according to the example of the configuration will be described with reference to FIG. The control method of the robot 100 includes each step of the voice recognition method executed in the voice recognition device 1. This speech recognition method is constituted by steps that a speech recognition program causes the speech recognition unit 11 and the detection unit 9 of the speech recognition apparatus 1 to execute.

  FIG. 2 is a flowchart illustrating an example of a processing procedure of the robot 100. Note that the voice recognition unit 11 is not normally operated, and is operated as necessary as described later.

  First, in step S1, the control unit 10 detects the engine speed by a vehicle sensor (not shown) and determines whether or not the engine is started. Next, in step S2, the voice recognition apparatus 1 starts overwriting recording of the sound generated in the surrounding environment.

  Specifically, the digitizing unit 5 binarizes the analog sound data 4 based on the sound collected by the microphone 3 and outputs the digital sound data 4 to the overwrite memory 7. In the overwrite memory 7, the sound data 4 is overwritten and recorded in a predetermined unit.

  Next, in step S3, the detection unit 9 specifies the frequency at which voice recognition should be performed using the voice recognition unit 11 based on, for example, the frequency band of voice uttered by the user. In the following description, the frequency specified in this way is called “specific frequency”. Here, in the present embodiment, as a method for determining whether or not the collected sound is a voice, for example, a characteristic on the frequency characteristics of being a human voice such as a driver (corresponding to the voice described above) or the like Judgment based on. The voice user may be a vehicle occupant.

  Next, in step S4, the detection unit 9 specifies a noise level that is constantly generated in the surrounding environment. This noise level corresponds to the environmental sound among the collected sounds. In the following description, the noise level specified in this way is referred to as a “specific noise level”. Next, in step S5, the detection unit 9 starts monitoring the volume level of the specific frequency.

  Next, in step S6, the detection unit 9 determines whether or not the vehicle engine is stopped based on the engine speed included in the vehicle information from a vehicle sensor (not shown). When the vehicle engine is stopped, the drive record process ends. On the other hand, when the vehicle engine is not stopped, the detection unit 9 executes the next step S7.

<Detection step>
In step S7, the detection unit 9 determines whether sound data 4 having a volume level exceeding a specific noise level has been input. The detection unit 9 returns to the above-described step S6 when the sound of such a volume level is not input, and triggers the sound recognition unit 11 when the sound of such a volume level is input. Output.

  Next, in step S8, it is determined whether or not the voice recognition unit 11 has received a trigger from the detection unit 9. If not, the process returns to step S6 described above and executed.

<Voice recognition step>
On the other hand, when the voice recognition unit 11 receives a trigger from the detection unit 9, voice recognition starts here. Next, in step S9, the voice recognition unit 11 acquires the sound data 4 recorded in the overwrite memory 7. Next, in step S10, although the details will be described later, the speech recognition unit 11 executes speech recognition processing.

  Next, in step S11, the speech recognition unit 11 determines whether or not a series of speech recognition processing has been completed, and when completed, waits until a predetermined time has elapsed, and returns to step S6 when the predetermined time has elapsed (step S11). S12). Even when the voice recognition process is completed, if the vehicle continues to run or ride, the recording of the sound data 4 in the overwrite memory 7 and the detection by the detection unit 9 will continue, and the vehicle running or boarding will continue. When finished, the recording to the overwrite memory 7 and the detection by the detection unit 9 are also finished.

<Control step>
The control unit 10 controls the operation based on the control content indicated by the voice part that has been voice-recognized in this way. Therefore, the electronic device 100 operates so as to respond to the user's voice as the control unit 10 controls according to the control content. At this time, since the voice part described above is voice-recognized without a loss of the head part, the electronic device 100 can surely respond to the user's wishes.

FIG. 3 is a diagram illustrating an example of a state in which the voice recognition process illustrated in FIG. 2 is performed. In the illustrated example, the horizontal axis represents time t and the vertical axis represents volume level L.
In this example, sound data 4 having a waveform as shown is recorded in the overwrite memory 7. In the sound data 4, the volume level L changes due to the user's voice from time Ts to time Te.

<Start of voice recognition processing>
As described above, the detection unit 9 outputs the trigger TG when the volume level L of the sound data 4 rises from the noise level L0 and exceeds the threshold volume level L1. The detection unit 9 measures a steady running noise level L0 and outputs a trigger TG when the volume level L1 exceeds the steady running noise level L0 by a predetermined amount or more.

  In the illustrated example, the volume level L increases from the noise level L0 at time Ts, but exceeds the threshold volume level L1 at time Tp. Accordingly, the detection unit 9 has a time difference Tw from when the volume level L starts to rise until when the trigger TG is actually output. When the voice recognition unit 11 described above receives the trigger TG from the detection unit 9, it first manages the detection time Tp and operates as follows.

  Further, the voice recognition unit 11 is a part of the sound data 4 recorded in the overwrite memory 7, the sound data of a specific portion recorded after a predetermined time Tw 0 including Tw 1 before the detection time Tp by the detection unit 9, that is, after time Ts 4, the voice part is determined, and voice recognition is performed on the voice part. This time Tw0 is set in advance to a time sufficient to include, for example, Tw1. Note that Tw1 is a time that varies each time, but can be accurately determined when an ideal mechanism is employed.

<End of voice recognition processing>
On the other hand, the voice recognition unit 11 stops the voice recognition process when the volume level L of the sound data 4 that has increased in this way falls to the noise level L0 and a predetermined time Tw2 has elapsed.

  4-10 is an image figure which shows the flow of a process of the sound data 4 in the speech recognition process mentioned above. 4 to 10, it is assumed that the sound data 4 in the overwrite memory 7 becomes data that has been saved as it progresses from the left to the right along the time axis t. That is, the sound data 4 in the overwrite memory 7 shown in the figure indicates that the left part is newly recorded data. In FIG. 4 and the like, the digitizing unit 5 is not shown.

  First, as shown in FIG. 4, the sound data 4 is gradually accumulated in the overwrite memory 7 as time passes, and as shown in FIG. 5, the voice recognition unit 11 receives this trigger TG from the detection unit 9 as a trigger. The sound data 4 stored in the overwrite memory 7 until the time is acquired. This time point represents the detection time Tp described above.

  The voice recognition unit 11 executes a voice recognition process based on the acquired sound data 4, but the sound data 4 is gradually overwritten and recorded in the overwrite memory 7 as shown in FIG.

  When the voice recognition unit 11 finishes the voice recognition process on the acquired sound data 4 as shown in FIG. 7, more sound data 4 is overwritten and recorded in the overwrite memory 7 according to the elapsed time.

  As shown in FIG. 8, the detection unit 9 outputs a trigger TG to the voice recognition unit 11 at a time when the volume level L of the sound data 4 exceeds the threshold volume level L1. This time corresponds to the detection time Tp described above. Then, the overwrite memory 7 delivers the stored sound data 4 up to the detection time Tp to the voice recognition unit 11 as shown in FIG.

  Here, the sound data 4 delivered to the voice recognition unit 11 includes only the sound data 4a representing a part of the sound data 4 from the voice partial start time Ts to the detection time Tp described above and a margin before the sound data 4a. Thereafter, recording of new sound data 4 continues, and the new sound data 4 is overwritten and recorded in the overwrite memory 7 as shown in FIG. The sound data 4 is delivered to the voice recognition unit 11 in small increments, for example, in short time units. Therefore, in addition to the sound data 4a and the margin described above, the sound recognition unit 11 further stores sound data 4b and sound data 4. The sound data 4b represents a part of the sound data 4 from the detection time Tp to the sound partial end time Te, and the sound data 4c is one of the sound data 4 from the sound partial end time Te to the end determination time Tx. Part. In the present embodiment, such delivery processing of sound data 4 is continued until such sound data 4c is delivered. The delivery timing may be after all the sound data 4 is accumulated in the overwrite memory 7, or may be delivered continuously in order to execute the speech recognition processing early. This voice part represents a command by a user's voice, for example. The voice recognition unit 11 performs the voice recognition process as described above, using the sound data 4a and 4b of the voice part corresponding to the time Ts to the time Te among the sound data 4a to 4c.

  The electronic device 100 in the above embodiment includes an overwrite recording means 7 (corresponding to an overwrite memory) in which the sound data 4 based on the collected sound is continuously overwritten, and the volume level L of the sound data 4 is a threshold volume level. Detection means 9 (corresponding to a detection unit) that detects that the sound volume exceeds L1, and when the detection means 9 detects that the volume level L of the sound data 4 exceeds the threshold volume level L1, the overwriting Of the sound data 4 recorded in the recording means 7, a sound part is identified from the sound data 4 recorded before a predetermined time Tw 1 before the detection time Ts by the detection means 9, and sound recognition is performed on the sound part. A voice recognition unit 11 (corresponding to a voice recognition unit) and a control unit 10 (control) that controls the operation based on the control content represented by the voice part identified by the voice recognition unit 11. And equivalent) in part, characterized by having a.

  In this way, the voice recognition unit 11 records the sound that has been recorded retroactively to the predetermined time Tw1 from the detection time Ts by the detection unit 9 so that the beginning of the voice part to be voice recognition target is not lost. Since the data 4 is used, it is not necessary to delay the input timing of the sound data 4 itself.

  For this reason, the voice recognition unit 11 can perform voice recognition on the voice part almost in real time at the timing at which voice recognition should be performed without missing the head of the voice part to be processed. Therefore, the control means 10 can control the operation of the electronic device 100 in real time based on the control content grasped by the voice recognition, and the electronic device 100 reacts quickly to the control content of the voice part. And can work.

  Moreover, in this electronic device 100, since the voice recognition unit 11 does not need to be constantly operating and may be started or stopped as necessary, when the voice recognition process is constantly executed. Compared to the resource burden.

  In the control method of the electronic device 100 in the above embodiment, the sound data 4 based on the collected sound is continuously overwritten and recorded in the overwrite recording means 7, and the sound volume level L of the sound data 4 is the threshold sound volume level. A detection step for detecting that the value exceeds L0, and when the detection step detects that the volume level L of the sound data 4 exceeds the threshold volume level L0, A voice recognition step of identifying a voice part from recorded sound data retroactive to a predetermined time Tw1 before the detection time Tp in the detection step, and performing voice recognition on the voice part; and the voice recognition step And a control step for controlling the operation based on the control content represented by the voice part identified in (1).

  In this way, in the voice recognition step, the recorded sound data is traced back to the predetermined time Tw1 from the detection time Ts in the detection step so that the head of the voice part to be subjected to voice recognition is not lost. 4 is used, it is not necessary to delay the input timing of the sound data 4 itself.

  For this reason, in the speech recognition step, speech recognition can be performed for this speech portion almost in real time at the timing at which speech recognition should be performed without missing the beginning of the speech portion to be processed. Therefore, in the control step, the operation of the electronic device 100 can be controlled in real time based on the control content grasped by the speech recognition, and the electronic device 100 reacts quickly to the control content of the voice part. Can work.

  Moreover, in this electronic device 100, since it is not necessary to constantly operate the voice recognition, it is only necessary to start or stop the voice recognition as necessary. Therefore, compared with the case where the voice recognition processing is constantly executed. Resource burden can be reduced.

  In the electronic device 100 according to the embodiment, in addition to the above-described configuration, the volume level of the sound data is a volume level of a specific frequency band of the sound data.

  In addition to the above-described configuration, the electronic device 100 according to the embodiment further controls the operation of the control unit 11 so as to react to a user who has emitted the voice part according to the control content.

  In this way, the electronic device 100 can operate in real time based on the control content grasped by the voice recognition, and can operate in agile response to the control content of the voice part. For this reason, the electronic device can react and operate quickly without giving the user a sense of incongruity.

  In addition to the above-described configuration, the electronic device 100 in the embodiment further includes the detection unit 9 (corresponding to a detection unit) that takes into account the noise level L0 included in the collected sound, and It is characterized by detecting whether or not the volume level L has reached the threshold volume level L1.

  If it does in this way, the detection means 9 can detect correctly that the audio | voice part is contained in the extract | collected sound.

  In addition to the above-described configuration, the electronic device 100 according to the embodiment further includes sound collection means 3 (corresponding to a microphone) that collects sound from the surrounding environment and outputs sound data based on the sound. To do.

  In this way, if the sound sampling means 3 is directed in a suitable direction, the voice recognition unit 11 can perform voice recognition by distinguishing the voice part from various sounds including the voice.

  The voice recognition device 1 in the above embodiment includes the overwrite recording means 7 in which the sound data 4 based on the collected sound is continuously overwritten and the volume level L of the sound data 4 exceeds the threshold volume level L1. When the detection means 9 detects that the volume level L of the sound data 4 has exceeded the threshold volume level L1, the detection means 9 detects the sound data 4 recorded in the overwrite recording means 7. Of these, a voice recognition unit 11 that identifies a voice part from the recorded sound data 4 before the predetermined time Tw1 before the detection time Tp by the detection unit 9 and performs voice recognition on the voice part is provided. And

  In this way, the voice recognition unit 11 records the sound that has been recorded retroactively to the predetermined time Tw1 from the detection time Ts by the detection unit 9 so that the beginning of the voice part to be voice recognition target is not lost. Since the data 4 is used, it is not necessary to delay the input timing of the sound data 4 itself.

  For this reason, the voice recognition unit 11 can perform voice recognition on the voice part almost in real time at the timing at which voice recognition should be performed without missing the head of the voice part to be processed. Further, since the voice recognition unit 11 does not need to be constantly operating and may be started or stopped as necessary, the burden of resources is larger than when the voice recognition process is constantly executed. Can be reduced.

  In the speech recognition method in the above embodiment, the sound volume level L of the sound data 4 exceeds the threshold volume level L1 in a state where the sound data 4 based on the collected sound is continuously overwritten and recorded in the overwrite recording means 7. And detecting that the volume level L of the sound data 4 exceeds the threshold volume level L1 in the detection step, the sound data 4 already recorded in the overwrite recording means 7 A voice recognition step of identifying a voice part from the recorded sound data 4 before a predetermined time Tw1 before the detection time Tp in the detection step and performing voice recognition on the voice part. And

  In the voice recognition program in the above embodiment, the sound volume level L of the sound data 4 exceeds the threshold volume level L0 in a state where the sound data 4 based on the collected sound is continuously overwritten and recorded in the overwrite recording means 7. When the detection step detects that the volume level L of the sound data 4 has exceeded the threshold volume level L0, the sound data 4 recorded in the overwrite recording means 7 is detected. A voice recognition step of identifying a voice part from the recorded sound data 4 before a predetermined time Tw1 before the detection time Tp in the detection step, and performing voice recognition on the voice part; and the voice recognition step And a control step for controlling the operation based on the control content represented by the voice part identified in the electronic device 100. .

  In this manner, in the voice recognition step, the recorded sound is traced back to the predetermined time Tw1 from the detection time Ts in the detection step so that the head of the voice part to be subjected to voice recognition is not lost. Since the data 4 is used, it is not necessary to delay the input timing of the sound data 4 itself.

  For this reason, in the speech recognition step, speech recognition can be performed for this speech portion almost in real time at the timing at which speech recognition should be performed without missing the beginning of the speech portion to be processed. In addition, since the voice recognition step does not need to be constantly operating and can be started or stopped as necessary, it is less resource intensive than when the voice recognition process is constantly executed. Can be reduced.

Second Embodiment
FIG. 11 is a block diagram illustrating a configuration example of the electronic device 100a according to the second embodiment.
The electronic device 100a in the second embodiment has substantially the same configuration as the electronic device 100 in the first embodiment, and performs substantially the same operation. For this reason, in the second embodiment, the same configurations and operations are denoted by the same reference numerals as those in FIGS. 1 to 10 in the first embodiment, and the description thereof is omitted. .

  The second embodiment is different in that the digitizing unit 5 and the voice recognition unit 11 are connected. Specifically, the voice recognition unit 11 acquires the sound data 4 directly from the digitizing unit 5 without going through the overwrite memory 7 in accordance with the progress of the voice recognition related to the voice part described above.

  In the first embodiment, the voice recognizing unit 11 acquires a voice part from the sound data 4 that goes back a predetermined time Tw1 before the detection time Tp by the detecting unit 9, and is only for the predetermined time Tw1 that goes back. Slightly not real-time processing.

  Since the voice recognition process by the voice recognition unit 11 is faster than the recording of the new sound data 4 in the overwrite memory 7, the voice recognition unit 11 performs voice recognition on the voice part of the sound data 4 that goes back in this way. Yes, this very slight delay can be recovered.

  Therefore, in the second embodiment, the voice recognition unit 11 directly acquires, instead of acquiring the sound data 4 from the overwrite memory 7, for example, after recovering such a delay according to the progress of the voice recognition regarding the voice part. Sound data 4 is acquired from the digitizing unit 5. Then, the voice recognition unit 11 can save the writing time for recording the sound data 4 in the overwrite memory 7 and can acquire the sound data 4 at an earlier stage to execute the voice recognition process.

  In addition to the above-described configuration, the electronic device 100 in the embodiment further includes the voice recognition unit 11 (corresponding to a voice recognition unit) via the overwrite recording unit 7 according to the progress of voice recognition related to the voice part. The sound data 4 is directly acquired without performing the above process.

  In this way, since the voice recognition unit 11 acquires the sound data 4 without going through the overwrite recording unit 7, for example, when there is a margin in the voice recognition process, the voice recognition unit 11 acquires the sound data 4 earlier. The speech recognition process can be completed earlier than in the embodiment.

In addition, this embodiment is not restricted above, A various deformation | transformation is possible. Hereinafter, such modifications will be described in order.
In the embodiment described above, the functions of the voice recognition device 1 may be configured by software using the voice recognition program described above, or may be implemented by hardware using a circuit or the like. You may comprise by. The digitizing unit 5 may also be configured by software as part of the voice recognition program.

  In the embodiment described above, the detection unit 9 outputs the trigger TG based on the volume level L of the sound data 4, but is not limited to this, and instead, for example, a gesture of a user who speaks is detected by image recognition. Then, the trigger TG may be output.

  In the above-described embodiment, the microphone 3 faces the seat where the driver or the passenger is present, and the voice recognition unit 11 is limited to, for example, a voice frequency band based on the sound data 4 acquired by the microphone 3. Then, the presence or absence of sound may be determined.

  In the above embodiment, the specific frequency to be detected by the detection unit 9 may be acquired from, for example, general experimental data or statistical data. The acquisition location may be, for example, a memory in a predetermined terminal (not shown) or a server on a network (not shown). Further, the specific frequency may be acquired from measurement data of a voice of a device such as the robot 100 or the owner of the vehicle.

  The measurement data may be registered in advance by the owner. For example, if the speech recognition apparatus 1 has a function of identifying the owner's voice, data on which frequency band the owner's past utterance was distributed to The frequency band may be determined by acquiring and using the accumulated data.

  This determination may include all frequencies that have been distributed even once, or may be determined by extracting only statistically meaningful data from the accumulated data. In addition, this determination may be performed by acquiring accumulated data relating to which frequency band the uttered sound uttered in the vehicle has been distributed in the past and using the accumulated data.

  Further, this determination may be made based on the frequency band data specific to the gender and age by acquiring the gender and age of the driver, the front seat passenger, or all the passengers. In this determination, for example, when both men and women are on board, the frequency band is determined by combining both frequency bands, or when there are passengers of various ages, the frequency bands of both are similarly set. You may make it determine to the combined frequency band.

  Further, in the above determination, data regarding to which frequency band the utterance sound uttered by the currently appearing passenger is distributed may be acquired, and the frequency band may be determined using the accumulated data. For example, when recognizing the first utterance appearing on the vehicle of the day, the frequency band may be determined from past data such as the previous boarding time, and the utterance data of the current boarding may be accumulated to be reviewed sequentially.

  Moreover, in the said embodiment, although the detection part 9 acquires steady noise level L0, you may use the following methods. That is, the detection unit 9 may acquire the noise level L0 from, for example, general experimental data or statistical data, or may acquire data obtained by measuring and accumulating the noise level in the vehicle in the past. . Moreover, the detection part 9 may determine the noise level L0 by measuring the noise level in the vehicle boarded this time, for example.

  Furthermore, the detection unit 9 may continuously measure the noise level L0 during boarding and review the noise level L0 in real time. That is, for example, the detection unit 9 may always determine whether or not the noise level L0 is appropriate based on the immediately preceding noise level. Further, the detection unit 9 may use the noise level L0 by removing the voice part from the collected sound. For example, the detection unit 9 may separately acquire the noise level during audio playback and when it is not, and determine whether the current audio playback is in use and determine which noise level to use. The detection unit 9 may monitor the volume level L of the specific frequency and determine whether or not a sound having a volume level L exceeding the specified noise level L0 is input.

  Moreover, in the said embodiment, the speech recognition part 11 complete | finishes a speech recognition process, when the input of a sound does not go over time Tw2 among the sound data 4, regarding whether the speech recognition process should be complete | finished, for example. However, the present invention is not limited to this, and the following determination may be made.

  That is, the voice recognition unit 11 executes the voice recognition process when there is a voice part corresponding to a voice or a sentence or a word, for example, as a result of starting the voice recognition process, and the voice or the voice corresponding to the sentence or the word. When the end of the part is detected, the voice recognition process may be terminated. Also. The voice recognition unit 11 executes the voice recognition process when, for example, there is a voice or a voice part corresponding to a sentence or a word as a result of starting the voice recognition process, and detects the end of the sentence or word for a predetermined time. If the next sound or the sound determined to be the next sound part corresponding to a sentence or a word is not detected, the process may be terminated.

It is a block diagram which shows the structural example of the robot as an example to which the electronic device in 1st Embodiment is applied. It is a flowchart which shows the example of a procedure of the process of a robot. It is a figure which shows an example of a mode that the speech recognition process shown in FIG. 2 is performed. It is an image figure which shows the flow of a process of the sound data in a speech recognition process. It is an image figure which shows the flow of a process of the sound data in a speech recognition process. It is an image figure which shows the flow of a process of the sound data in a speech recognition process. It is an image figure which shows the flow of a process of the sound data in a speech recognition process. It is an image figure which shows the flow of a process of the sound data in a speech recognition process. It is an image figure which shows the flow of a process of the sound data in a speech recognition process. It is an image figure which shows the flow of a process of the sound data in a speech recognition process. It is a block diagram which shows the structural example of the electronic device in 2nd Embodiment.

Explanation of symbols

3 Microphone (equivalent to sound collection means)
4 Sound data 7 Overwrite memory (equivalent to overwrite recording means)
11 Voice recognition unit (equivalent to voice recognition means)
13 Control unit (equivalent to control means)
100 Electronic device L0 Noise level Ts Detection time Tw1 Predetermined time

Claims (10)

Overwrite recording means for continuously overwriting and recording sound data based on the collected sound;
Detecting means for detecting that the volume level of the sound data exceeds a threshold volume level;
When the detection means detects that the volume level of the sound data exceeds the threshold volume level, the sound data recorded in the overwrite recording means goes back a predetermined time before the detection time by the detection means. A voice recognition means for identifying a voice part from the recorded sound data and performing voice recognition on the voice part;
Control means for controlling the operation based on the control content represented by the voice portion identified by the voice recognition means;
An electronic device comprising: The electronic device according to claim 1,
The volume level of the sound data is a volume level of a specific frequency band of the sound data. The electronic device according to claim 1 or 2,
The control means includes
An electronic apparatus that controls an operation so as to react to a user who has emitted the voice portion according to the control content. The electronic device according to any one of claims 1 to 3,
The electronic device is characterized by detecting whether or not the volume level of the sound data has reached a threshold volume level in consideration of a noise level included in the collected sound. The electronic device according to any one of claims 1 to 4,
An electronic apparatus comprising sound collection means for collecting sound from a surrounding environment and outputting sound data based on the sound. The electronic device according to any one of claims 1 to 5,
The voice recognition means
An electronic apparatus characterized in that the sound data is directly acquired without going through the overwrite recording means in accordance with the progress of voice recognition relating to the voice portion. A detection step of detecting that the volume level of the sound data exceeds a threshold volume level in a state where the sound data based on the collected sound is continuously overwritten and recorded in the overwrite recording means;
When it is detected in the detection step that the volume level of the sound data exceeds the threshold volume level, a predetermined time before the detection time in the detection step of the sound data recorded in the overwrite recording means A voice recognition step of identifying a voice part from recorded sound data retroactively and performing voice recognition on the voice part;
A control step for controlling the operation based on the control content represented by the voice portion identified in the voice recognition step;
A method for controlling an electronic device, comprising: Overwrite recording means for continuously overwriting and recording sound data based on the collected sound;
Detecting means for detecting that the volume level of the sound data exceeds a threshold volume level;
When the detection means detects that the volume level of the sound data exceeds the threshold volume level, the sound data recorded in the overwrite recording means goes back a predetermined time before the detection time by the detection means. A voice recognition means for identifying a voice part from the recorded sound data and performing voice recognition on the voice part;
A speech recognition apparatus comprising: A detection step of detecting that the volume level of the sound data exceeds a threshold volume level in a state where the sound data based on the collected sound is continuously overwritten and recorded in the overwrite recording means;
When it is detected in the detection step that the volume level of the sound data exceeds the threshold volume level, a predetermined time before the detection time in the detection step of the sound data recorded in the overwrite recording means A voice recognition step of identifying a voice part from recorded sound data retroactively and performing voice recognition on the voice part;
A speech recognition method comprising: A detection step of detecting that the volume level of the sound data exceeds a threshold volume level in a state where the sound data based on the collected sound is continuously overwritten and recorded in the overwrite recording means;
When it is detected in the detection step that the volume level of the sound data exceeds the threshold volume level, a predetermined time before the detection time in the detection step of the sound data recorded in the overwrite recording means A voice recognition step of identifying a voice part from recorded sound data retroactively and performing voice recognition on the voice part;
Is executed by a computer.

Images