JP5797009B2 - Voice recognition apparatus, robot, and voice recognition method - Google Patents

Description

  The present invention relates to a voice recognition device, a robot, and a voice recognition method.

  In a robot with a voice recognition function that executes a specific motion command according to the user's utterance, while reducing the false recognition of voice due to a false reaction to noise other than human voice, increase the success rate of voice recognition It is demanded.

Here, in Patent Document 1, voice data emitted by the user is detected by a microphone, and whether or not the user's mouth is moving is determined from an image obtained by capturing the user's face, and it is determined that the mouth is moving. A technique for controlling a robot apparatus by issuing only voice commands included in voice data during operation as operation commands is disclosed.
Further, Patent Document 2 recognizes the voice of a specific word or sentence used at the beginning of a conversation made by a speaker and its direction based on the voice input from the microphone array, and detects the direction of the detected voice. A technology is disclosed in which a camera is pointed, a human face is detected from an image input from the camera, and interactive processing is performed. Further, Patent Document 2 restricts directivity to the detected speaker direction, recognizes the voice and direction of the speaker, performs face detection processing, moves to the detected face direction, and performs speech recognition accuracy. A technique for further improving the above is disclosed.

JP 2007-190620 A JP 2006-251266 A

However, since the technique disclosed in Patent Document 1 recognizes only a voice command when there is a mouth movement, the success rate of voice recognition at the start of speech greatly decreases due to a delay in image recognition. there is a possibility. Further, the technique disclosed in Patent Document 1 only selects whether to accept a voice command from an image, and thus does not contribute to improving the success rate of voice recognition.
Further, as in the technique disclosed in Patent Document 2, merely changing the directivity cannot prevent overresponse due to noise in the direction of the directivity. In addition, indoors, the component of the reverberation sound in the room is very large, and the volume ratio of the target sound and noise by changing the directivity is hardly improved, and there is a high possibility that the effect will not appear.

  Note that, generally speaking, utterance determination by image is less affected by noise but lacks accuracy in timing (for example, because the mouth movement at the start of utterance is small), and speech utterance determination is accurate by timing. However, it is said to be vulnerable to noise.

  The present invention has been made in view of such circumstances, and provides a speech recognition device, a robot, and a speech recognition method capable of increasing the recognition rate of speech recognition while reducing overresponse. Objective.

  In order to solve the above-described problems, the speech recognition apparatus, robot, and speech recognition method of the present invention employ the following means.

That is, the speech recognition apparatus according to the present invention captures a subject and obtains image information indicating the subject, and sound information indicating a sound generated when the imaging unit performs imaging. A person speaks based on sound acquisition means to be acquired, voice recognition means for recognizing a person's voice based on the sound information acquired by the sound acquisition means, and the image information acquired by the imaging means. The speech recognition means by the speech recognition means in comparison with the case in which the speech section detected by the speech section detection means for detecting the speech section indicating the period during which the speech is detected is not detected in the speech section detected by the speech section detection means. includes a sensitivity changing unit for raising the sensitivity, the said voice recognition means, the state quantity of more than a predetermined threshold value, recognizes that it is the voice of a human, said sensitivity changing means, In detected by serial voice activity detection means and the speech period, as compared with the case where the speech segment is not detected, by lowering the threshold value, increasing the sensitivity of the speech recognition by the speech recognition means.

  According to the present invention, the image information indicating the subject is acquired by the imaging unit and the image information indicating the subject is acquired, and the sound information indicating the sound generated when the imaging unit is capturing the image by the sound acquisition unit. Is acquired.

Further, the voice recognition means recognizes the human voice based on the sound information. However, in speech recognition by the speech recognition means, noise other than speech may be misrecognized as speech. Misrecognition results in overresponse, and the speech recognition rate decreases.
Such misrecognition can be prevented by lowering the sensitivity of voice recognition. However, if the sensitivity of voice recognition is lowered, there is a possibility that a sound that should be recognized as a human voice is not recognized.

Therefore, the utterance section indicating the period during which the person is speaking is detected by the utterance section detection means based on the image information acquired by the imaging means. That is, based on image information, a human face is recognized, and an utterance section that is not affected by noise is detected from the movement of the recognized human face organ.
Then, the sensitivity of the voice recognition means by the voice recognition means is increased by the sensitivity change means, compared to the case where the speech section is not detected in the speech section detected by the speech section detection means.

Accordingly, since the sensitivity of speech recognition is increased corresponding to the utterance period that is not affected by noise detected based on the image information, the present invention reduces the over-response while recognizing speech recognition. The rate can be increased.
Furthermore, since the state quantity equal to or greater than a predetermined threshold is recognized as human speech and the speech recognition sensitivity is increased by lowering the threshold in the utterance section detected based on the image information, The sensitivity of voice recognition can be changed.
In the speech recognition apparatus of the present invention, the magnitude of the threshold value in the utterance section and the magnitude of the threshold value in the section where the utterance section is not detected may be made different according to the surrounding environment.
Further, the speech recognition device of the present invention lowers the threshold value in the utterance section detected by the utterance section detection unit, and based on the sound indicated by the sound information acquired by the sound acquisition unit, the state quantity And a section in which the state quantity equal to or greater than the threshold value is generated may be detected as a speech section.
In the speech recognition device of the present invention, the detection of the utterance section based on the sound indicated by the sound information is performed after the change of the threshold, and the speech recognition based on the sound information is performed by the sound information. It may be performed after the detection of the utterance section based on the sound shown.

  Further, in the speech recognition apparatus according to the present invention, the sensitivity changing means may recognize the speech recognition at a predetermined time before and after the utterance interval together with the utterance interval detected by the utterance interval detection means. The sensitivity of voice recognition by means may be increased.

  At the start of a human utterance (beginning of a word) or at the end of an utterance (end of a word), for example, since the mouth cannot be opened widely, the beginning or ending of the utterance may not be detected as the utterance section.

  According to the present invention, the sensitivity of speech recognition by the speech recognition means is increased at the predetermined time before and after the utterance section together with the utterance section, that is, in the utterance section thickened before and after. . The predetermined time is a time corresponding to the beginning or ending of the utterance that may not be detected by the utterance section detecting means, and is a value obtained by experiments or the like, and is set in advance.

  Therefore, the sensitivity of voice recognition can be increased more reliably even at the beginning or end of an utterance.

  According to the present invention, a sound volume that is equal to or higher than a predetermined threshold is recognized as a human voice, and the sensitivity of voice recognition is increased by lowering the threshold in an utterance section detected based on image information. Therefore, the sensitivity of voice recognition can be easily changed.

  In the speech recognition apparatus of the present invention, the utterance section detecting means may detect a utterance section in which the person is speaking based on the movement of the mouth included in the person's face.

According to the present invention, since the utterance section is detected based on the movement of the person's mouth, the utterance section can be easily detected based on the image information.
The speech recognition apparatus of the present invention may determine that the mouth is moving when the utterance section detecting means detects a tooth.

  In the speech recognition apparatus of the present invention, the utterance section detecting means may detect the utterance section that the person is speaking based on the direction of eyes included in the face of the person.

According to the present invention, since an utterance section is detected based on the direction of a person's eyes, that is, a line of sight, the utterance section of a person who is speaking to a device equipped with a speech recognition device can be easily detected.
Further, in the speech recognition apparatus of the present invention, the utterance section detecting means may detect the utterance section in which the person is speaking based on the change in the head direction and position.

  On the other hand, a robot according to the present invention includes the above-described voice recognition device.

Furthermore, the speech recognition method according to the present invention captures a subject, acquires image information indicating the subject by the imaging unit, and obtains sound information indicating a sound generated when the imaging unit performs imaging. A first step that is acquired by a sound acquisition unit; a second step that detects a speech section in which a person is speaking based on the image information acquired by the imaging unit; and the second step that is detected by the second step. in speech periods, as compared with the case where the speech segment is not detected, seen including a third step, the increasing the sensitivity of the speech recognition of a person based on the sound information, the first step, the threshold than the predetermined Is recognized as a human voice, and the third step lowers the threshold in the utterance interval detected in the second step as compared to the case where the utterance interval is not detected. And by increasing the sensitivity of the speech recognition.

  According to the present invention, there is an excellent effect that the recognition rate of voice recognition can be increased while reducing overresponse.

It is a front view of the robot which concerns on embodiment of this invention. It is a figure required for description of the conventional voice recognition. It is a functional block diagram which shows the function of the speech recognition apparatus which concerns on embodiment of this invention. It is the schematic diagram which showed concretely the content of the process performed in the speech area detection part which concerns on embodiment of this invention. It is the schematic diagram which showed the result of the speech recognition by the speech recognition apparatus which concerns on embodiment of this invention.

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

FIG. 1 is a front view of the robot 10 according to the present embodiment.
As shown in FIG. 1, the robot 10 includes a head 12, a chest 14 that supports the head 12 from below, a right arm 16 a provided on the right side of the chest 14, and a left arm provided on the left side of the chest 14. A portion 16b, a waist portion 18 connected below the chest portion 14, a skirt portion 20 connected below the waist portion 18, and a leg portion 22 connected below the skirt portion 20 are provided.

In the vicinity of the center of the front surface of the head 12, as shown in FIG. 1, a camera 30 and a microphone 32 (hereinafter simply referred to as “microphone 32”) for imaging the front are provided.
The camera 30 captures an image of a subject and acquires image information indicating the subject, and the microphone 32 acquires sound information indicating a sound generated when the camera 30 is capturing an image.

The robot 10 according to the present embodiment recognizes a person's face based on the image information acquired by the camera 30 and recognizes a person's voice based on the sound information acquired by the microphone 32. Perform recognition processing.
That is, the robot 10 recognizes the face of a person who is trying to communicate with the robot 10, recognizes the voice of the person, and performs an operation according to the recognition result.

Here, conventional speech recognition will be described with reference to FIG.
The robot 10 recognizes sound information having a volume (power) exceeding a predetermined threshold (hereinafter referred to as “voice threshold”) as a voice uttered by a person.

  However, as shown in FIG. 2A, when noise (sound different from human voice) exceeding the voice threshold occurs, the robot 10 erroneously recognizes the noise as human voice. In addition, as shown in FIG. 2B, when the noise exceeding the voice threshold is overlapped with the human voice, the robot 10 not only erroneously recognizes the noise as the human voice but also correctly recognizes the human voice. It will not be recognized. Such misrecognition results in over-response and causes a reduction in the recognition rate of voice recognition.

  It is conceivable to prevent erroneous recognition of noise as shown in FIGS. 2A and 2B by lowering the sensitivity of speech recognition, that is, by increasing the value of the speech threshold. However, if the sensitivity of voice recognition is lowered, there is a possibility that a sound that should be recognized as a human voice is not recognized.

  Therefore, the robot 10 according to the present embodiment detects an utterance section indicating a period during which the person is speaking from the movement of the human face organ recognized based on the image information acquired by the camera 30, and detects the detected utterance. In the section, processing for increasing the sensitivity of speech recognition (decreasing the speech threshold) is performed as compared to the case where the speech section is not detected. Since this utterance section is obtained from the image information, it is not affected by noise.

FIG. 3 is a functional block diagram illustrating functions of the voice recognition device 40 that performs voice recognition processing.
Note that the robot 10 according to the present embodiment realizes processing by each component included in the speech recognition apparatus 40 by executing a program by a CPU (Central Processing Unit). In this case, the program is installed in a ROM (Read Only Memory) or other storage medium in advance, or is provided in a state stored in a portable storage medium such as a computer-readable CD-ROM. A form distributed via a wired or wireless communication means can be applied.

  The speech recognition device 40 includes an utterance section detection unit 42A that detects a human utterance section based on image information, an utterance section detection unit 42B that detects a human utterance section based on sound information, and human speech in the utterance section. Is provided.

  The utterance section detection unit 42 </ b> A includes a face organ detection unit 50, a motion amount calculation unit 52, a threshold processing unit 54, and a voice threshold change unit 56.

The facial organ detection unit 50 recognizes a human face based on the image information acquired by the camera 30 and detects a predetermined facial organ. A conventionally known method for detecting a facial organ may be used.
Note that the facial organ detection unit 50 according to the present embodiment detects a human mouth as a facial organ.

The movement amount calculation unit 52 calculates the movement of the facial organ detected by the facial organ detection unit 50.
In this embodiment, in order to detect a human mouth as a facial organ, the opening degree of the mouth is calculated as the movement of the mouth, more specifically, the opening amount of the upper lip and the lower lip is calculated.

  The threshold processing unit 54 determines whether or not the value calculated by the motion amount calculation unit 52 is equal to or greater than a predetermined threshold (hereinafter referred to as “image threshold”), and a period (time) when the value is equal to or greater than the image threshold. Is detected as an utterance section indicating a period during which a person is speaking.

  The voice threshold change unit 56 raises the sensitivity of voice recognition by lowering the voice threshold in the utterance section detected by the threshold processing unit 54 as compared to the case where the utterance section is not detected. The voice threshold change information indicating the magnitude and interval of the lowered voice threshold is output from the voice threshold change unit 56 to the utterance section detection unit 42B.

  The utterance section detection unit 42B includes a volume calculation unit 60 and a threshold processing unit 62.

  The volume calculation unit 60 calculates the volume from the amplitude of the waveform indicated by the sound information acquired by the microphone 32.

  The threshold processing unit 62 determines whether or not the volume calculated by the volume calculation unit 60 is equal to or higher than the voice threshold, and detects a volume higher than the voice threshold as an utterance section. Note that the threshold processing unit 62 according to the present embodiment detects an utterance section using the section indicated by the voice threshold change information input from the voice threshold change section 56 and the lowered voice threshold value, and the utterance The section is output to the speech recognition unit 44 as utterance section information.

The sound information input to the utterance section detection unit 42B is input with a predetermined time delay via the delay processing unit 70A.
As described above, since the threshold value processing unit 62 detects the utterance period using the threshold value change information output from the voice threshold value change unit 56, the threshold value change information of the voice threshold value change unit 56 included in the utterance period detection unit 42A is used. This is because the processing by the threshold processing unit 62 is started after the output is completed.

  The voice recognition unit 44 includes a feature amount extraction unit 80 and a matching processing unit 82.

  The feature quantity extraction unit 80 extracts sound features (feature quantities) by, for example, Fourier transforming the sound information acquired by the microphone 32.

Based on the recognition dictionary information indicating the utterance content corresponding to the sound feature amount, the matching processing unit 82 calculates the human feature from the sound feature amount in the utterance section indicated by the utterance section information output from the threshold processing unit 62. The content of the utterance is specified (matching process) and output as a speech recognition result. Note that the recognition dictionary information is stored in advance in storage means (not shown).
Then, for example, when the voice recognition result indicates an utterance “good morning”, the robot 10 performs an operation based on the voice recognition result, such as outputting a voice “good morning”.

The sound information input to the voice recognition unit 44 is input with a predetermined time delay via the delay processing unit 70B.
As described above, since the matching processing unit 82 performs speech recognition using the utterance section information output from the threshold processing unit 62, the matching processing unit 82 after the output of the utterance section information by the threshold processing unit 62 is completed. This is for starting the process. For this reason, the time delay by the delay processing unit 70B must be further delayed than the time delay by the delay processing unit 70A.

  FIGS. 4A and 4B are schematic diagrams specifically showing the contents of the processing performed by the utterance section detection units 42A and 42B.

  As shown in FIG. 4A, the movement amount calculation unit 52 calculates the movement amount of the mouth every 1 / 60th (1/60) second or 1 / 30th (1/30) second, for example. . The threshold processing unit 54 determines whether or not the amount of motion is equal to or greater than the image threshold, and detects a period in which the amount of motion is equal to or greater than the image threshold as an utterance section. By this threshold determination, a small mouth movement is not detected as an utterance section.

Note that the threshold processing unit 54 according to the present embodiment performs a thickening process that extends the detected utterance section to a predetermined time before and after the utterance section.
Since the mouth cannot be opened widely at the start of the utterance (start of word) or at the end of the utterance (end of word) by a person, there is a possibility that the beginning or ending of the utterance is not detected as the utterance section.
Therefore, by expanding the utterance interval before and after, the beginning and ending of the utterance are included in the utterance interval. The predetermined time for expanding the utterance section is a time corresponding to the beginning or ending of the word that may not be detected by the threshold processing unit 54, and is a value obtained by experiments or the like, and is set in advance.

  Then, the voice threshold value is reduced by the voice threshold change unit 56 in the utterance section where the thickening process is performed, and is output to the threshold processing unit 62 as voice threshold change information.

On the other hand, as shown in FIG. 4 (B), the volume calculation unit 60, for example, based on the sound indicated by the sound information input with a time delay, for example, the average value of the maximum value of the amplitude at every predetermined time interval Is calculated as the volume.
The volume calculated by the volume calculation unit 60 is output to the threshold processing unit 62, and the threshold processing unit 62 performs threshold determination using the audio threshold indicated by the volume change information, and a volume equal to or higher than the audio threshold is generated. A section is detected as an utterance section.

5A and 5B are schematic views showing the results of speech recognition by the speech recognition apparatus 40 according to the present embodiment. 5A and 5B show the results of conventional speech recognition (see FIG. 2), and the right diagrams in FIGS. 5A and 5B show the speech according to the present embodiment. It is the result of recognition.
As shown in the right diagram of FIG. 5A, it is possible to prevent erroneous recognition of noise by lowering the voice threshold. Then, since the voice threshold is lowered in the utterance section detected based on the image information, the voice recognition sensitivity is increased. Therefore, the voice recognition device 40 does not over-respond to noise, and the voice of the person is correct. Will be recognized.
Further, as shown in the right diagram of FIG. 5B, since noise is not erroneously recognized even if noise and human voice overlap, human voice is recognized correctly.

Further, depending on the environment around the robot 10, the size of the voice threshold in the utterance section detected based on the image information (hereinafter referred to as “intra-section voice threshold”), the voice threshold in the section where the utterance section is not detected ( Hereinafter, the magnitude of “out-of-interval voice threshold” may be varied. For example, in an environment where the volume of noise is high (for example, in an amusement facility), the out-of-interval voice threshold is set higher. Further, in an environment in which a voice spoken by the person to the robot 10 is likely to be low (for example, in a document hall), the intra-section voice threshold is set smaller.
As described above, by changing the ratio between the intra-speech voice threshold and the non-speech voice threshold according to the environment around the robot 10, the ratio of over-response to noise is lowered and the voice recognition rate is improved. Is desirable.

  Moreover, you may change the magnitude | size and ratio of the voice threshold within a zone, and the voice threshold outside a zone according to a language.

As described above, the speech recognition apparatus 40 according to the present embodiment is generated when the subject 30 is imaged and image information indicating the subject is acquired by the camera 30 and the camera 30 is imaging. Sound information indicating sound is acquired by the microphone 32. Then, the voice recognition device 40 detects the utterance section in which a person is speaking based on the image information acquired by the camera 30, and the sound information is compared with the case where the utterance section is not detected in the detected utterance section. Raise the sensitivity of human voice recognition based on.
Therefore, the speech recognition apparatus 40 according to the present embodiment reduces the overresponse because the sensitivity of speech recognition is increased in correspondence with the speech section that is not affected by the noise detected based on the image information. However, the recognition rate of voice recognition can be increased.

  In addition, the speech recognition device 40 according to the present embodiment increases the sensitivity of speech recognition before and after the utterance interval, together with the utterance interval detected based on the image information, so even at the beginning or end of the utterance, The sensitivity of voice recognition can be increased more reliably.

  In addition, the speech recognition apparatus 40 according to the present embodiment recognizes that the sound volume is equal to or higher than a predetermined speech threshold as human speech, and lowers the speech threshold in the utterance section detected based on the image information. Since the voice recognition sensitivity is increased, the voice recognition sensitivity can be easily changed.

  Moreover, since the speech recognition apparatus 40 according to the present embodiment detects the utterance section based on the movement of the person's mouth, the utterance section can be easily detected based on the image information.

  As mentioned above, although this invention was demonstrated using the said embodiment, the technical scope of this invention is not limited to the range as described in the said embodiment. Various changes or improvements can be added to the above-described embodiment without departing from the gist of the invention, and embodiments to which the changes or improvements are added are also included in the technical scope of the present invention.

For example, in the above-described embodiment, the voice recognition device 40 has been described based on the movement of the mouth included in the face of the person to detect the utterance section in which the person is speaking. However, the present invention is not limited to this, and the speech section in which the person is speaking may be detected based on the direction of eyes included in the face of the person, that is, the line of sight.
In the case of this form, the voice recognition device 40 identifies a person whose line of sight is facing the robot 10, and lowers the voice threshold when the line of sight of the identified person is facing the direction of the robot 10.

  The voice recognition device 40 may combine the movement of the mouth and the direction of the line of sight, and may lower the voice threshold when the line of sight faces the direction of the robot 10 and the movement of the mouth is equal to or greater than the image threshold. . Thereby, even when there are a plurality of persons in the imaging range of the camera 30, only the voice of the person who is talking to the robot 10 can be recognized.

  Further, in addition to the movement of the person's mouth and line of sight, for example, a change in the direction or position of the person's face or head is detected, and the person speaks when the person's face is facing the robot 10. It is good also as a form detected as a certain speech section.

  Further, in the above-described embodiment, the speech section detection unit 42A has described the form of calculating the opening / closing of the mouth as the movement of the mouth, but the present invention is not limited to this, and for example, the movement of the mouth ( Analyzing the frequency of opening and closing of the mouth), it may be judged that the mouth is moving when the frequency is above a predetermined frequency, or when white is detected between the upper lip and the lower lip In case) it may be determined that the mouth is moving.

  In the above embodiment, the utterance section detection unit 42A has been described as performing thickening processing to thicken the utterance section detected based on the image information back and forth, but the present invention is not limited to this. Further, only one of the front and rear of the utterance section may be thickened.

  Moreover, although the said embodiment demonstrated the form which applies the speech recognition apparatus 40 to the robot 10, this invention is not limited to this, As a form applied to other apparatuses, such as a personal computer and an IC recorder. Also good.

DESCRIPTION OF SYMBOLS 10 Robot 30 Camera 32 Microphone 40 Voice recognition apparatus 42A Speech area detection part 44 Voice recognition part 56 Voice threshold value change part

Claims (11)

Imaging means for imaging the subject and acquiring image information indicating the subject;
Sound acquisition means for acquiring sound information indicating sound that is occurring when imaging by the imaging means is being performed;
Voice recognition means for recognizing a person's voice based on the sound information acquired by the sound acquisition means;
Based on the image information acquired by the imaging means, an utterance section detecting means for detecting an utterance section indicating a period during which a person is speaking;
In the utterance section detected by the utterance section detection means, compared with a case where the utterance section is not detected, a sensitivity changing means for increasing the sensitivity of speech recognition by the speech recognition means;
Equipped with a,
The voice recognition means recognizes a state quantity equal to or greater than a predetermined threshold as human voice,
The sensitivity changing means lowers the threshold value in the utterance section detected by the utterance section detection means, compared with the case where the utterance section is not detected, thereby increasing the sensitivity of voice recognition by the voice recognition means. Recognition device. The speech recognition apparatus according to claim 1, wherein a magnitude of the threshold in the utterance section and a magnitude of the threshold in a section in which the utterance section is not detected differ depending on a surrounding environment. Lowering the threshold in the utterance interval detected by the utterance interval detection means,
  Based on the sound indicated by the sound information acquired by the sound acquisition means, the state quantity is calculated,
  The voice recognition device according to claim 1, wherein a section in which the state quantity equal to or greater than the threshold is generated is detected as an utterance section. Detection of the utterance section based on the sound indicated by the sound information is performed after the change of the threshold value,
The speech recognition apparatus according to any one of claims 1 to 3, wherein speech recognition based on the sound information is performed after detection of an utterance section based on a sound indicated by the sound information. The sensitivity changing means increases the sensitivity of voice recognition by the voice recognition means at a predetermined time before and after the utterance section together with the utterance section detected by the utterance section detection means. The speech recognition device according to any one of claims 1 to 4 . The voice activity detection means based on the movement of the mouth that is included in the human face, the speech recognition apparatus according to any one of claims 1 to 5 for detecting a speech interval the person is speaking. The speech recognition apparatus according to any one of claims 1 to 6, wherein the speech section detection means determines that the mouth is moving when a tooth is detected. The voice activity detection means based on the eye direction is included in the human face, the speech recognition apparatus according to any one of claims 1 to 7 for detecting a speech interval the person is speaking.   The speech recognition apparatus according to any one of claims 1 to 8, wherein the utterance section detecting unit detects an utterance section in which the person is speaking based on a change in a head direction or a position. A robot comprising the voice recognition device according to any one of claims 1 to 9 . A first step of capturing an image of a subject, acquiring image information indicating the subject by the imaging unit, and acquiring sound information indicating a sound generated when the imaging unit is capturing by the sound acquisition unit;
A second step of detecting an utterance section in which a person is speaking based on the image information acquired by the imaging means;
A third step of increasing the sensitivity of human speech recognition based on the sound information in the utterance interval detected by the second step, compared to a case where the utterance interval is not detected;
Only including,
The first step recognizes a state quantity equal to or greater than a predetermined threshold as human voice,
In the speech recognition method , the third step increases the sensitivity of speech recognition by lowering the threshold value in the utterance interval detected in the second step as compared to a case where the utterance interval is not detected .

Images