JP4410378B2 - Speech recognition method and apparatus - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an on-vehicle speech recognition device that recognizes a vocabulary of input instructions given to a control target device controlled by speech.
[0002]
[Prior art]
FIG. 8 is a block diagram showing a system configuration of a conventional in-vehicle speech recognition device. Hereinafter, it demonstrates according to a figure. When operating a navigation device or audio device while driving a vehicle, etc., to reduce the burden on the driver due to the switch operation, the voice spoken by the speaker (speaker) such as the driver is recognized and connected. There is a voice recognition device for instructing input to a device.
[0003]
Reference numeral 1 denotes a voice input unit that converts a voice uttered by a speaker into an electrical signal, and includes a plurality of omnidirectional microphones 1a to 1c. A directivity adjustment unit 3 adjusts signals detected by the microphones 1a to 1c to output a voice signal in which the directivity of the voice input unit 1 is adjusted in the direction of the speaker. Reference numeral 5 denotes a voice recognition unit connected to an input unit such as a navigation device or an audio device. Reference numeral 8 denotes direction detection means for detecting the direction of the speaker, which detects the direction of voice from the angle of the rearview mirror, the seat slide position, the reclining angle, and the like. A directivity control unit 41 controls the directivity adjustment unit 3 based on the detection result of the direction detection unit 8.
[0004]
Next, the operation will be described.
FIG. 9 is a flowchart showing the operation content of a conventional in-vehicle voice recognition device. First, in step S0, a voice recognition start operation is performed. Next, in step S1, the direction detection means 8 detects the direction of the speaker and acquires the position information of the speaker. Next, in step S2, the directivity control unit 41 controls the directivity adjusting unit 3 so that the directivity control unit 41 changes the directivity to the direction of the speaker based on the position information of the speaker acquired in step S1. Next, in step S3, the voice of the speaker is input from the microphones 1a to 1c. Subsequently, in step S4, the voice recognition unit 5 performs a recognition process on the voice whose directivity adjustment unit 3 has changed the directivity. In step S5, the speech recognition unit 5 outputs a recognition result.
[0005]
[Problems to be solved by the invention]
In the conventional speech recognition apparatus, when adjusting the directivity, a means for detecting the direction of the speaker is required in order to specify the position of the speaker. Conventionally, this direction detection means is detected from the angle of the rearview mirror, the seat slide position, the reclining angle, etc. in the case of in-vehicle use. For this reason, the speaker is limited to being a driver, and when the person other than the driver is a speaker, the direction detection unit cannot detect the speaker. When detecting the voice as a speaker other than the driver, not only the configuration of the direction detection means is complicated, but also a switch for determining the speaker is necessary as it is necessary to detect who speaks. Necessary. Even when these configurations are realized, the specified directivity is not necessarily a characteristic that can provide high performance in the speech recognition processing.
[0006]
The present invention improves the speech recognition performance by reducing noise other than the voice of the speaker, and is highly convenient to handle the voice of the speaker existing in a plurality of directions without a switch for speaker discrimination. An object is to provide a speech recognition method and apparatus.
[0007]
[Means for Solving the Problems]
The speech recognition apparatus according to claim 1 includes a speech input unit, a speech buffer unit that stores a signal from the speech input unit, and a directivity adjustment that generates a signal having a changed directivity from the signal stored in the speech buffer unit. A voice recognition unit that performs voice recognition processing on a signal whose directivity has been changed by the directivity adjustment unit, a voice buffer control unit that reproduces a signal stored in the voice buffer unit, and a directivity adjustment unit. A directivity information storage section that stores a plurality of directivity information to be changed, a directivity control section that selects one of the directivity information stored in the directivity information storage section and controls the directivity adjustment section; A speech recognition dictionary that stores recognition targets in the recognition unit;
Each directivity signal is used by changing the directivity multiple times in the directivity adjustment section under the control of the directivity control section. Speech recognition control unit that executes recognition processing multiple times, and recognition results for multiple modified directivities And the data stored in the voice recognition dictionary A determination result storage unit that stores the degree of coincidence, and a coincidence degree determination unit that determines the directivity having the highest degree of coincidence among the recognition results stored in the determination result storage unit. , Determine the directivity with the highest degree of coincidence, and continue to use the determined directivity in a series of subsequent voice conversations, and input the voice from the voice operator with the directivity with the highest recognition performance Is.
[0008]
The speech recognition apparatus according to claim 2 includes a speech input unit, a speech buffer unit that stores a signal from the speech input unit, and a directivity that generates a signal in which directivity and gain are changed from the signal stored in the speech buffer unit. A directivity gain adjustment unit, a speech recognition unit that performs speech recognition processing on a signal whose directivity and gain have been changed by the directivity gain adjustment unit, a speech buffer control unit that reproduces a signal stored in the speech buffer unit, A directivity gain information storage unit storing a plurality of directivity information and gain information to be changed by the directivity gain adjustment unit, and one each selected from the directivity information and gain information stored in the directivity gain information storage unit A directivity gain control unit that controls the directivity gain adjustment unit, a speech recognition dictionary unit that stores recognition targets in the speech recognition unit, and
Each directivity signal is used by changing the directivity and gain multiple times in the directivity adjustment section under the control of the directivity control section. Speech recognition control unit that executes recognition processing multiple times, and recognition results for multiple modified directivities and gains And the data stored in the voice recognition dictionary A determination result storage unit that stores the degree of coincidence, and a coincidence degree determination unit that determines the directivity and gain having the highest degree of coincidence among the recognition results stored in the determination result storage unit. Determine the directivity and gain with the highest degree of coincidence, and continue to use the determined directivity and gain in a series of subsequent voice conversations, and input the voice from the voice operator with the directivity with the highest recognition performance. Is.
[0009]
According to a third aspect of the present invention, there is provided a speech recognition method for performing recognition processing on an input speech signal and determining whether or not a keyword is recognized. When a keyword is recognized, the speech signal of the keyword is recognized. Changed directivity Multiple times using each directional audio signal Perform recognition processing, The recognition results of multiple modified directivities and the data stored in the speech recognition dictionary A step of obtaining a degree of coincidence, a step of determining the directivity having the highest degree of coincidence among these recognition results, and the determination Set the directivity to the directivity with the highest degree of coincidence, and then recognize the operation commands for audio equipment etc. The step which performs is performed.
[0010]
According to a fourth aspect of the present invention, there is provided a speech recognition method for performing recognition processing on an input speech signal and determining whether or not a keyword is recognized. When a keyword is recognized, the speech signal of the keyword is recognized. Changed directivity and gain Multiple times using audio signals of each directivity and gain Perform recognition processing, The recognition result of multiple modified directivities and gains and the data stored in the speech recognition dictionary A step of obtaining a degree of coincidence, a step of determining directivity and gain having the highest degree of coincidence among the recognition results, and the determination Set the directivity and gain to the directivity and gain with the highest degree of coincidence, and then recognize the operation commands for audio equipment etc. The step which performs is performed.
[0011]
DETAILED DESCRIPTION OF THE INVENTION
Embodiment 1 FIG.
Hereinafter, embodiments of the present invention will be described with reference to an on-vehicle speech recognition apparatus.
FIG. 1 is a block diagram showing a system configuration of a speech recognition apparatus according to Embodiment 1 of the present invention. In the figure, reference numeral 1 denotes a voice input unit that converts a voice uttered by a speaker into an electrical signal (hereinafter referred to as a voice signal), and includes a plurality (three in this case) of microphones 1a to 1c.
FIG. 2 is a plan view showing an example of a position where the microphone is attached to the vehicle. Within the vehicle 11, three microphones 1 a to 1 c are installed on the dashboard 12 at a central portion so as to form a triangle at regular intervals. For example, by adjusting the gain balance of the output signals of the microphones 1a to 1c, the directivity can be changed to the front, the driver's 13 direction, or the passenger's 14 direction.
[0012]
Returning to FIG. 1, reference numeral 2 denotes an audio buffer unit for storing an audio signal from the audio input unit 1, and includes a plurality of audio buffers 2a to 2c provided corresponding to the microphones 1a to 1c, respectively. The directivity adjusting unit 3 adjusts the audio signal stored in the audio buffer unit 2 and outputs the audio signal whose directivity is changed. The reference numeral 5 denotes the audio signal whose directivity is changed by the directivity adjusting unit 3. The speech recognition unit executes speech recognition processing, and outputs the degree of coincidence between the recognition result and data stored in the speech recognition dictionary unit 6 described below. Reference numeral 6 denotes a voice recognition dictionary unit for storing recognition targets, in which data serving as a reference in the voice recognition processing of the voice recognition unit 5 is stored.
[0013]
4 is a control unit that controls the audio buffer unit 2, the directivity adjustment unit 3, and the voice recognition unit 5, 43 is an audio buffer control unit that controls the storage and reproduction of audio signals in the audio buffer unit 2, and 41 is a directivity adjustment. A directivity control unit that controls the directivity change in the unit 3, and 42 is a directivity information storage unit that stores a plurality of directivity information for directivity control by the directivity control unit 41. Then, up to ± 90 ° is stored at intervals of 5 °, and the directivity control unit 41 selects and controls one by one. 44 is a speech recognition control unit that starts and stops the recognition process of the speech recognition unit 5 and acquires the degree of coincidence with the recognition result, and 45 is the directivity information from the directivity control unit 41 and the recognition from the speech recognition control unit 44. A coincidence determination unit that determines which directivity is optimal from the result and the coincidence, 46 is a determination result for storing the recognition result and the coincidence in the speech recognition unit, and the determination result in the coincidence determination unit 45 It is a memory | storage part and the control part 4 is comprised by said 41-46.
[0014]
Next, the operation will be described.
FIG. 3 is a flowchart showing the operation contents of the speech recognition apparatus shown in FIG. First, in step A0, initialization of each part of the control part 4 and start operation of the process are executed.
Next, in step A1, the directivity setting of the directivity adjustment unit 3 is set to non-directivity by the control from the directivity control unit 41.
Next, in step A2, the signals input to the microphones 1a to 1c and converted into audio signals are stored in the audio buffers 2a to 2c under the control of the audio buffer control unit 43, and the stored audio signals are stored as audio signals. Reproduction is performed under the control of the buffer control unit 43, and this audio signal is input to the directivity adjustment unit 3 set to non-directionality, and the output of the directivity adjustment unit 3 is input to the audio recognition unit 5. The voice recognition unit 5 executes a process of recognizing a keyword that is a voice operation start command, for example, “recognition start”, by the control from the voice recognition control unit 44 on the input voice signal.
[0015]
Next, in step A3, based on the recognition result of the speech recognition unit 5, the speech recognition control unit 44 determines whether or not the keyword “recognition start” has been recognized. Input processing and keyword recognition processing are executed. If recognized, the process proceeds to step A4.
Next, when the process proceeds to step A4, the voice input to the voice buffers 2a to 2c is stopped by the control from the voice buffer control unit 43, and the voice signal when the keyword “recognition start” is recognized is stored. To do.
[0016]
Next, in step A5, the directivity setting of the directivity adjustment unit 3 is set in the directivity information storage unit 42, for example, in the 0 ° front direction, under the control of the directivity control unit 41.
Next, in step A6, the voice signal of the keyword “recognition start” stored in the voice buffers 2a to 2c in step A4 is reproduced under the control of the voice buffer control unit 43, and the directivity control unit 3 controls the directivity. The voice signal having directivity set by the unit 41 is generated, the keyword recognition process is executed again by the voice recognition unit 5, and the voice recognition control unit 44 acquires the result of the recognition process and the degree of coincidence from the voice recognition unit 5. And transmitted to the coincidence degree determination unit 45. The coincidence determination unit 45 transmits the currently set directivity information, the recognition result, and the coincidence to the determination result storage unit 46 for storage.
[0017]
Next, in step A7, if re-recognition processing for all directivities stored in the directivity information storage unit 42 and acquisition of recognition results and coincidences are not completed, the process returns to step A5, and the directivity information storage unit It repeats until it complete | finishes about all the directivity memorize | stored in 42. FIG. If the re-recognition process and the recognition result and the degree of coincidence have been acquired for all directivities, the process proceeds to step A8.
[0018]
Next, when proceeding to Step A8, the coincidence degree determination unit 45 determines that the recognition result is correct, that is, the keyword “recognition start” from the recognition results and the coincidence degrees for all directivities stored in the determination result storage unit 46. The directivity control unit 41 controls the directivity adjustment unit 3 so that the directivity determined by the coincidence determination unit 45 is obtained. .
[0019]
Next, in step A9, the audio input to the audio buffers 2a to 2c stopped in step A4 is resumed. That is, the respective signals input to the microphones 1 a to 1 c and converted into audio signals are stored in the audio buffers 2 a to 2 c under the control of the audio buffer control unit 43, and the stored audio signals are transmitted from the audio buffer control unit 43. The voice signal is input to the directivity adjustment unit 3 that is set to the directivity with the correct recognition result and the highest coincidence in step A8, and the output of the directivity adjustment unit 3 is recognized as a voice. Input to part 5. The voice recognition unit 5 executes processing for recognizing the recognition vocabulary stored in the voice recognition dictionary unit 6 under the control of the voice recognition control unit 44 with respect to the input voice signal.
Next, in step A10, the voice recognition unit 5 outputs the result of the recognition process, and operates an audio device or the like not shown.
In step A8, if the recognition completion of the keyword “recognition start” and the directionality setting direction are displayed on a display or lamp (not shown), the speaker confirms the display, and in step A9. , You can enter the following command.
[0020]
As described above, in the speech recognition method and apparatus of the first embodiment, the directivity is changed using the speech signal of the speech buffer at the time when the keyword “recognition start” is recognized, and the speaker's degree of coincidence in speech recognition is determined. Since the speaker's voice is extracted by determining the direction, it is possible to effectively extract and recognize the speaker's voice even when the speaker's direction is not fixed. Further, no speaker direction detection means is required, and costs such as a speaker discrimination switch or a direction detection sensor can be reduced.
[0021]
Embodiment 2. FIG.
FIG. 4 is a block diagram showing a system configuration of the speech recognition apparatus according to Embodiment 2 of the present invention.
In the present embodiment, in addition to the method of changing the directivity performed in the first embodiment and selecting the one with the highest degree of matching, a method of viewing the degree of matching by changing the gain, that is, the signal level, is used.
In FIG. 4, instead of the directivity adjustment unit 3, the directivity control unit 41, and the directivity storage unit 42 of FIG. 1, a directivity gain adjustment unit 31, a directivity gain control unit 47, and a directivity gain storage unit 48 are provided. ing.
[0022]
In FIG. 4, 31 is a directivity gain adjustment unit that adjusts an audio signal held in the audio buffer unit 2 and outputs an audio signal whose directivity and gain are changed, and 47 is a directivity in the directivity gain adjustment unit 31. A directivity gain control unit 48 that controls the change in gain, and a directivity gain information storage unit 48 that stores a plurality of directivity information and a plurality of gain information in the directivity and gain control of the directivity gain control unit 47. For example, the front is 0 °, and ± 90 is stored at 5 ° intervals, and the initial gain is 0 dB, and ± 15 dB is stored at 3 dB intervals.
[0023]
The voice recognition unit 5 performs voice recognition processing on the voice signal whose directivity and gain are changed by the directivity gain adjustment unit 31. The coincidence determination unit 45 determines which directivity and gain are optimal from the directivity information and gain information from the directivity gain control unit 47 and the recognition result and coincidence from the speech recognition control unit 44.
The control unit 4 includes 43 to 48. The other parts are the same as in FIG.
[0024]
Next, the operation will be described.
FIG. 5 is a flowchart showing the operation contents of the speech recognition apparatus shown in FIG. First, in step B0, initialization of each part of the control part 4 and start operation of the process are executed.
Next, in step B1, the directivity and gain setting of the directivity gain adjustment unit 31 are set to non-directivity and initial gain by the control from the directivity gain control unit 47.
Next, in step B2, the signals input to the microphones 1a to 1c and converted into audio signals are stored in the audio buffers 2a to 2c under the control of the audio buffer control unit 43, and the stored audio signals are stored as audio signals. The audio signal is reproduced under the control of the buffer control unit 43, and this audio signal is input to the directivity gain adjustment unit 31 set to omnidirectionality and initial gain, and the output of the directivity gain adjustment unit 31 is input to the audio recognition unit 5. input. The voice recognition unit 5 executes a process of recognizing a keyword that is a voice operation start command, for example, “recognition start”, by the control from the voice recognition control unit 44 on the input voice signal.
[0025]
Next, in step B3, based on the recognition result of the voice recognition unit 5, the voice recognition control unit 44 determines whether or not the keyword “recognition start” has been recognized. Processing and keyword recognition processing are executed. If recognized, the process proceeds to step B4.
Next, when the process proceeds to step B4, the voice input to the voice buffers 2a to 2c is stopped by the control from the voice buffer control unit 43, and the voice signal when the keyword “recognition start” is recognized is stored. To do.
[0026]
Next, in step B 5, the directivity setting of the directivity gain adjustment unit 31 is set to, for example, the front 0 ° direction stored in the directivity gain information storage unit 48 by the control from the directivity gain control unit 47.
Next, in step B 6, the voice signal of the keyword “recognition start” stored in the voice buffers 2 a to 2 c in step B 4 is reproduced under the control of the voice buffer control unit 43, and the directivity gain adjustment unit 31 directs the directivity gain. A voice signal having directivity set by the control unit 47 is generated, the keyword recognition process is executed again by the voice recognition unit 5, and the voice recognition control unit 44 acquires the result of the recognition process and the degree of coincidence from the voice recognition unit 5. Then, it transmits to the coincidence degree determination unit 45. The coincidence determination unit 45 transmits the currently set directivity information, the recognition result, and the coincidence to the determination result storage unit 46 for storage.
[0027]
Next, in step B7, if re-recognition processing for all directivities stored in the directivity gain information storage unit 48 and acquisition of recognition results and coincidence have not been completed, the process returns to step B5, and directivity gain information is acquired. The process is repeated until the re-recognition process and the recognition result and the degree of coincidence are acquired for all directivities stored in the storage unit 48. If the re-recognition process and the recognition result and the degree of coincidence have been acquired for all directivities, the process proceeds to step B8.
[0028]
Next, when the process proceeds to step B8, the coincidence determination unit 45 determines that the recognition result is the correct answer, that is, the keyword “recognition start” from the recognition results and the coincidence of all directivities stored in the determination result storage unit 46. And the directivity gain control unit 47 controls the directivity gain adjustment unit 31 so that the directivity determined by the coincidence determination unit 45 is obtained. Control.
[0029]
Next, in step B9, the gain setting of the directivity gain adjustment unit 31 is set to a gain that is 3 dB higher than the initial gain stored in the directivity gain information storage unit 48 under the control of the directivity gain control unit 47, for example. In this case, the gain is adjusted only in the direction of directivity determined in step B8.
[0030]
Next, in step B10, the speech signal of the keyword “recognition start” stored in the speech buffers 2a to 2c in step B4 is reproduced under the control of the speech buffer control unit 43, and the directivity gain adjustment unit 31 directs the directivity gain. A voice signal having the directivity and gain set by the control unit 47 is generated, and the keyword recognition process is executed again by the voice recognition unit 5, and the voice recognition control unit 44 determines the degree of coincidence with the result of the recognition process. And sent to the coincidence determination unit 45. The coincidence determination unit 45 transmits the currently set directivity information, gain information, recognition result, and coincidence to the determination result storage unit 46 for storage.
[0031]
Next, in step B11, when the re-recognition processing for all gains stored in the directivity gain information storage unit 48 and the recognition result and the degree of coincidence have not been completed, the process returns to step B9 to return to the directivity gain information storage unit. The process is repeated until the re-recognition processing for all the gains stored in 48 and the acquisition of the recognition result and the matching degree are completed. If the re-recognition process and the recognition result and the degree of coincidence have been acquired for all gains, the process proceeds to step B12.
[0032]
Next, when the process proceeds to step B12, the coincidence degree determination unit 45 determines that the recognition result is correct and has the highest degree of coincidence based on the recognition results and coincidence values for all gains stored in the determination result storage unit 46. The directivity gain control unit 47 controls the directivity gain adjustment unit 31 so that the directivity and gain determined by the coincidence determination unit 45 are obtained.
[0033]
Next, in step B13, the audio input to the audio buffers 2a to 2c stopped in step B4 is resumed. That is, the respective signals input to the microphones 1 a to 1 c and converted into audio signals are stored in the audio buffers 2 a to 2 c under the control of the audio buffer control unit 43, and the stored audio signals are transmitted from the audio buffer control unit 43. The voice signal is input to the directivity gain adjustment unit 31 that is set to the directivity and gain with the highest recognition and the recognition result is correct in step B12. The output is input to the voice recognition unit 5. The voice recognition unit 5 executes processing for recognizing the recognition vocabulary stored in the voice recognition dictionary unit 6 under the control of the voice recognition control unit 44 with respect to the input voice signal.
Next, in step B14, the voice recognition unit 5 outputs the result of the recognition process, and controls an audio device (not shown).
[0034]
As described above, in the speech recognition method and apparatus according to the second embodiment, the direction of the speaker is determined using the degree of coincidence in speech recognition, and the optimum gain, ie, speech recognition, is further obtained using the degree of coincidence in speech recognition. Because the optimal input signal level is determined and the speaker's voice is extracted, the speaker's voice is effectively extracted even when the speaker's direction is not fixed, and the distance from the speaker to the microphone, etc. Even if the voice input level to the microphone is different, the recognition process can be executed with the optimum input signal level for voice recognition. In addition, no speaker direction detection means is required, and the cost of a direction detection sensor or the like can be reduced.
[0035]
Embodiment 3 FIG.
FIG. 6 is a block diagram showing a system configuration of the speech recognition apparatus according to Embodiment 3 of the present invention.
This embodiment shows an example of controlling a navigation device by voice. In FIG. 6, in addition to what was shown in FIG. 4, the navigation apparatus 7 is shown.
In FIG. 6, reference numeral 7 denotes a navigation device in which various operations are executed by the control unit 4 based on the result of voice recognition. Others are the same as in FIG.
[0036]
Next, the operation will be described.
FIG. 7 is a flowchart showing the operation content of the speech recognition apparatus shown in FIG. Steps C0 to C12 are the same as steps B0 to B12 in FIG. However, although an example of a keyword is described as “recognition start” in FIG. 5, another keyword such as “navigation” is used in FIG. 7.
[0037]
In step C13 following step C12, the audio input to the audio buffers 2a to 2c stopped in step C4 is resumed. That is, the respective signals input to the microphones 1 a to 1 c and converted into audio signals are stored in the audio buffers 2 a to 2 c under the control of the audio buffer control unit 43, and the stored audio signals are transmitted from the audio buffer control unit 43. The speech signal is input to the directivity gain adjustment unit 31 that is set to the directivity and gain with the highest degree of coincidence and the recognition result is correct in step C12. The output is input to the voice recognition unit 5. In the voice recognition unit 5, the control command vocabulary of the navigation device 7 stored in the voice recognition dictionary unit 6, for example, “detail display”, “wide area display”, is controlled by the voice recognition control unit 44 with respect to the input voice signal. ”And“ Destination setting ”are recognized.
[0038]
Next, in step C14, the voice recognition control unit 44 acquires the recognition result of the voice recognition unit 5, for example, “detailed display”, and transmits a control signal corresponding to the recognition result “detailed display” to the navigation device 7. Next, in step C15, a process of displaying the display screen in detail according to the control signal corresponding to the recognition result “detailed display” received by the navigation device 7 is executed.
[0039]
As described above, in the speech recognition method and apparatus of the third embodiment, the directivity is changed by using the speech signal of the speech buffer at the time when the keyword “navigation” is recognized, and the directivity having the highest degree of coincidence in speech recognition is obtained. Since the determination is made and this is set as the direction of the speaker, and the gain with the highest degree of matching is determined and the subsequent speech recognition process is executed, a voice recognition start switch or the like is unnecessary, and the operation becomes simple. Further, since the directivity and gain with the highest degree of coincidence in the keyword recognition process are determined, the direction of the speaker is not limited to the driver, and voice operation from the passenger seat is possible, which is optimal in the subsequent command recognition process. Recognition processing can be executed with directivity and gain, and recognition performance can be improved. In addition, no speaker direction detection means is required, and the cost of a direction detection sensor or the like can be reduced.
[0040]
【The invention's effect】
According to the speech recognition apparatus of the first aspect, the speech recognition unit that performs speech recognition processing on the signal with the changed directivity, and the coincidence determination unit that determines the directivity having the highest coincidence among the recognition results. Therefore, there is no switch or sensor for speaker identification even when there are multiple speakers and the direction is not fixed by setting the directivity that the matching score is determined to be the highest by the matching score determination unit. Thus, noise other than the speaker can be reduced to effectively extract the voice of the speaker, and a voice recognition device with high voice recognition performance and high convenience can be obtained.
[0041]
According to the speech recognition apparatus of the second aspect, the speech recognition unit that performs speech recognition processing on the signal with the changed directivity and gain, and the directivity and gain having the highest degree of coincidence are determined from the recognition results. Since it has a coincidence determination unit, the direction of the speaker is not fixed by setting to the directivity and gain determined to be the highest in the coincidence determination unit, and to the voice input unit (microphone) Even if the input level is high or low, it is possible to extract the speaker's voice effectively and perform recognition processing at an appropriate signal level without using a switch or sensor for speaker identification. A highly recognizable speech recognition apparatus can be obtained.
[0042]
According to the speech recognition method according to claim 3, a plurality of recognition processes with different directivities are performed on the recognized keyword speech signal, and the directivity having the highest degree of matching is determined from the recognition results. Because the voice recognition is performed after setting this directivity, the voice of the speaker can be extracted effectively without the switch or sensor for speaker discrimination even when the direction of the speaker is not fixed. And a highly convenient speech recognition method can be obtained.
[0043]
According to the speech recognition method according to claim 4, a plurality of recognition processes with different directivities and gains are performed on the recognized keyword speech signal, and the directivity having the highest degree of matching among the recognition results and Since the gain is determined, and the subsequent speech recognition is performed with this directivity and gain set, the speaker's direction is not fixed, and the speaker input is used even when the input level to the voice input unit is high or low. Without the switches and sensors, the voice of the speaker can be extracted effectively and the recognition process can be performed at an appropriate signal level, so that a voice recognition method with high voice recognition performance and high convenience can be obtained.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a system configuration of a speech recognition apparatus according to Embodiment 1 of the present invention.
2 is a plan view showing a microphone attachment position of the speech recognition apparatus of FIG. 1. FIG.
FIG. 3 is a flowchart showing the operation content of the speech recognition apparatus of FIG. 1;
FIG. 4 is a block diagram showing a system configuration of a speech recognition apparatus according to Embodiment 2 of the present invention.
FIG. 5 is a flowchart showing the operation content of the speech recognition apparatus of FIG. 4;
FIG. 6 is a block diagram showing a system configuration of a speech recognition apparatus according to Embodiment 3 of the present invention.
7 is a flowchart showing the operation content of the speech recognition apparatus of FIG. 6;
FIG. 8 is a block diagram showing a system configuration of a conventional speech recognition apparatus.
FIG. 9 is a flowchart showing the operation content of the speech recognition apparatus of FIG. 8;
[Explanation of symbols]
1 speech input unit, 2 speech buffer unit, 3 directivity adjustment unit, 5 speech recognition unit,
6 speech recognition dictionary unit, 31 directivity gain adjustment unit, 41 directivity control unit,
42 directivity information storage unit, 43 voice buffer control unit, 44 voice recognition control unit,
45 coincidence determination unit, 46 determination result storage unit, 47 directivity gain control unit,
48 Directivity gain information storage unit.

Claims (4)

An audio input unit, an audio buffer unit that stores a signal from the audio input unit, a directivity adjustment unit that generates a signal whose directivity is changed from the signal stored in the audio buffer unit, and the directivity adjustment unit A speech recognition unit that performs speech recognition processing on a signal whose directivity has been changed by the above, a speech buffer control unit that reproduces a signal stored in the speech buffer unit, and a plurality of directivities that are changed by the directivity adjustment unit A directivity information storage unit storing the directivity information; a directivity control unit that selects one of the directivity information stored in the directivity information storage unit to control the directivity adjustment unit; and the speech recognition unit A speech recognition dictionary part that stores the recognition target in
A voice recognition control unit that performs directivity processing a plurality of times using a signal of each directivity by changing the directivity a plurality of times in the directivity adjustment unit under the control of the directivity control unit;
A determination result storage unit that stores the degree of coincidence between the changed recognition results for the plurality of directivities and the data stored in the speech recognition dictionary unit, and the degree of coincidence among the recognition results stored in the determination result storage unit A matching degree determination unit that determines the highest directivity, determines the directivity with the highest degree of matching, and continuously uses the determined directivity in a series of subsequent voice conversations to receive voice from the voice operator Voice recognition device that inputs with the highest recognition performance directivity . An audio input unit, an audio buffer unit for storing a signal from the audio input unit, a directivity gain adjusting unit for generating a signal in which directivity and gain are changed from the signal stored in the audio buffer unit, and the directivity A speech recognition unit that performs speech recognition processing on a signal whose directivity and gain have been changed by a directivity gain adjustment unit, a speech buffer control unit that reproduces a signal stored in the speech buffer unit, and the directivity gain adjustment unit A directivity gain information storage unit storing a plurality of directivity information and gain information to be changed in step 1, and selecting each one from the directivity information and gain information stored in the directivity gain information storage unit A directivity gain control unit that controls the directional gain adjustment unit, a speech recognition dictionary unit that stores recognition targets in the speech recognition unit,
A voice recognition control unit that executes a recognition process a plurality of times using a signal of each directivity by changing the directivity and gain a plurality of times in the directivity adjustment unit under the control of the directivity control unit;
A determination result storage unit that stores the degree of coincidence between the recognition results for the changed plurality of directivities and gains and the data stored in the speech recognition dictionary unit, and a match among the recognition results stored in the determination result storage unit And a coincidence determination unit that determines the directivity and gain with the highest degree of determination, determine the directivity and gain with the highest degree of coincidence, and continue to use the determined directivity and gain in the subsequent series of voice conversations A voice recognition device for inputting voice from a voice operator with directivity having the highest recognition performance . A step of performing recognition processing on the input voice signal to determine whether or not a keyword is recognized. When a keyword is recognized, each directional voice in which directivity is changed with respect to the voice signal of the keyword. A signal is used to perform recognition processing multiple times to obtain a degree of coincidence between the changed recognition results for the plurality of directivities and the data stored in the speech recognition dictionary, and the degree of coincidence is the highest among these recognition results. A speech recognition method comprising: determining a high directivity; and setting a directivity to the determined directivity having the highest degree of coincidence to perform subsequent operation command speech recognition of an audio device or the like . A step of performing recognition processing on the input voice signal to determine whether or not a keyword is recognized. When a keyword is recognized, each directivity in which directivity and gain are changed with respect to the voice signal of the keyword . A step of performing recognition processing a plurality of times using the sound signal of each gain, and obtaining a degree of coincidence between the recognition results in the changed plurality of directivities and gains and the data stored in the speech recognition dictionary unit, and these recognition results Determining the directivity and gain with the highest degree of coincidence from the above, and setting the directivity and gain to the determined directivity and gain with the highest degree of coincidence for subsequent operation command speech recognition of audio equipment, etc. A speech recognition method comprising the steps of:

Images