JP2003044092A - Voice recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice recognizing device with which the voices of a speaker can be collected with a satisfactory S/N (signal-noise ratio) at all times even when the position of a microphone changes. SOLUTION: The angle of a sun visor 19, to which microphones M1 and M2 are attached, is detected by an angle sensor 18. On the basis of the output of the angle sensor 18, the direction of the mouth of the speaker to the microphones M1 and M2 is found by a microphone directivity calculating part 16, and suitable delay time is calculated. Signal delayers 12a and 12b are controlled by a delay time control part 15 so that the delay time calculated by the microphone directivity calculating part 16 can be provided. Thus, even when the sun visor 19 is moved, voices from the direction of the speaker are extracted with the satisfactory S/N at all times and the recognition rate of voices by a voice recognizing processing part 14 is prevented from being lowered.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device used in an environment where the relative positional relationship between a microphone and a speaker changes.

[0002]

2. Description of the Related Art In recent years, voice recognition technology has been remarkably developed and has been used in various fields. For example, in many on-vehicle navigation devices, operations such as destination setting and screen switching can be performed by voice. In this case, a voice input microphone is installed in the vehicle interior, but it is necessary to devise a microphone that does not obstruct the driver's view, collects the driver's voice well, and minimizes the effect of noise in the vehicle interior. is there.

It is also conceivable to use a headset microphone as a voice input microphone of a vehicle-mounted navigation device having a voice recognition function. However, driving a vehicle with the headset microphone mounted on the head is cumbersome and not common. Therefore, in a general vehicle-mounted navigation device, a directional microphone is attached to the sun visor in front of the driver's seat. This method also has an advantage that the microphone can be easily attached to the vehicle interior.

[0004]

However, in the method of attaching the directional microphone to the sun visor, for example, when the driver lowers the sun visor to avoid the sunshine when the sunshine from the front of the vehicle is dazzling, the direction of the mic Will change. For this reason, it becomes difficult for the driver's voice to be picked up, the SN ratio (ratio of signal to noise) deteriorates, and the accuracy of voice recognition significantly decreases.

In view of the above, an object of the present invention is to provide a voice recognition device which can always pick up the voice of the speaker with a good SN ratio even if the position of the microphone changes.

[0006]

A voice recognition device of the present invention comprises a plurality of microphones arranged apart from each other, and a microphone position change amount detecting means for detecting a position change amount of the plurality of microphones. The microphone position change amount detected by the microphone position change amount detecting means is input, and a signal from the direction of the speaker is output from the outputs of the plurality of microphones by using a signal delay according to a difference in position of each microphone. It is characterized by comprising a voice signal extraction means for selectively extracting a signal and a voice recognition processing section for performing voice recognition processing on the voice signal extracted by the voice signal extraction means.

In the present invention, the output of a plurality of microphones arranged apart from each other is subjected to signal processing by the voice signal extraction means to emphasize the signal by the voice from the direction of the speaker, or a predetermined direction. The sound signal from the direction of the speaker is selectively extracted by attenuating the signal due to the noise from. When a plurality of microphones arranged apart from each other are used, a time difference occurs in signals from each microphone due to a difference in distance between each microphone and a speaker. How much time difference occurs depends on the direction of the speaker's mouth with respect to the microphone. For example, if a signal delay device is connected between each microphone and the signal synthesizer so that the output signals of each microphone from the speaker's voice are input to the signal synthesizer at the same time, the voice from the speaker's direction will have the same phase. Therefore, it is emphasized by addition. On the other hand, noises coming from different directions do not have the same phase due to the delay, so even if they are added, they are not emphasized as much as speech. As a result, the SN ratio is improved.

Further, for example, when there is a noise source in a predetermined direction, the noise component is attenuated by setting the delay time of the signal delay device so that the signal due to the noise from that direction has the opposite phase. On the other hand, the voice signal from the direction of the speaker is not completely out of phase and is not attenuated as much as noise, resulting in an improvement in the SN ratio. With either method, when the position of the microphone changes, the appropriate delay time changes. Therefore, a microphone position change amount detecting means for detecting the change amount of the microphone position, a delay time determining unit for determining the delay time of the signal delay unit according to the change amount of the microphone position,
A delay time control unit that controls the signal delay device so that the delay time determined by the delay time determination unit is required.

[0009]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the accompanying drawings. (First Embodiment) FIG. 1 is a diagram showing the principle of a voice recognition apparatus according to a first embodiment of the present invention. In the voice recognition device of the present embodiment, two omnidirectional microphones M ₁ ,
M ₂ is used, and the voice of the speaker is transmitted to these microphones M ₁ and M _2.
The sound from the direction of the speaker (the mouth of the speaker) is selectively picked up by using the difference in the time until reaching each of the.

For example, as shown in FIG. 1, the distance between the first microphone M ₁ and the second microphone M ₂ is L (m), and the speed of sound is C.
(M / sec), the signal output from the first microphone M ₁ has a time difference of L / C seconds with respect to the signal output from the second microphone M ₂ . Therefore, the output of the second microphone M ₂ is delayed by L / C seconds and the first microphone M ₂ is delayed.
_If the output of ₁ is summed, the voice outputs of the microphones M ₁ and M ₂ are in-phase, so that the voice in the direction of the speaker is emphasized, the SN ratio is improved, and the direction of the speaker is increased. Only the voice of can be selectively picked up.

As shown in FIG. 2, when the distance between the first microphone M ₁ and the second microphone M ₂ is d and the direction of the speaker is θ, the distance from the speaker to the microphones M ₁ and M ₂ is The distance difference L is
As shown in the following equation (1), dsin θ.

[0012]

[Equation 1]

Therefore, the output of the second microphone M ₂ is (dsi
By delaying by n θ) / C seconds and summing with the output of the first microphone M ₁ , only the voice from the direction of the speaker can be selectively picked up. FIG. 3 is a block diagram showing the configuration of the voice recognition device according to the first embodiment. FIG. 3 shows an example in which the present invention is applied to a voice recognition device of a vehicle-mounted navigation device.

The speech recognition apparatus of this embodiment has two
Directional microphone M₁, M₂And the signal delay devices 12a and 12b
, Adder (signal synthesizer) 13, and voice recognition processing unit 14
, Delay time control unit 15, microphone directivity calculation unit (delay
(Time determination unit) 16, seat position sensor 17, angle sensor
It is composed of the service 18 and the service. Omnidirectional microphone M₁，
M₂Are separated from each other by the sun visor 19 in front of the driver's seat.
And is installed. Mike M₁Is the signal delay unit 12
a, connected to a microphone M ₂Is connected to the signal delay device 12b
Has been. These signal delay devices 12a and 12b are
The delay time is individually controlled by the time control unit 15. Na
Oh, Mike M₁, M₂Need not be omnidirectional
However, even if the position changes, the sensitivity to the direction of the speaker changes.
It is preferable that there is little change.

The adder 13 is a signal delay device 12a, 12b.
Add the signals output from. The signal output from the adder 13 is sent to the voice recognition processing unit 14. The voice recognition processing unit 14 performs voice processing on the signal input from the signal synthesis unit 13 and converts the signal into character data. Speech recognition processing unit 14
The character data output from the device is sent to the navigation device body (not shown), and the navigation device body executes processing according to the character data.

The seat position sensor 17 detects a change in the position of the speaker (or his / her mouth) caused by sliding the driver's seat in the longitudinal direction of the vehicle. Although the seat position sensor 17 is not essential, in order to improve the accuracy of voice recognition, the seat position sensor 17 is used to detect the seat position, or the height of the speaker is input from an input device such as a keyboard. , It is preferable to more accurately determine the position of the speaker's mouth from those data.

The angle sensor 18 is, for example, as shown in FIG. 4, a sun visor 1 to which microphones M ₁ and M ₂ are attached.
It is arranged in the bearing portion 19a of 9 and detects the rotation angle of the sun visor 19 with respect to the rotation shaft 19b. Angle sensor 1
As 8, for example, a potentiometer whose resistance value changes according to the rotation angle, or a liquid-filled type sensor that detects the rotation angle by utilizing the impedance change according to the position change of the liquid level in the sensor is used. can do.

The microphone directivity calculation unit 16 determines the microphone M from the output of the seat position sensor 17 and the output of the angle sensor 18.
₁ calculates a mouth direction of the speaker with respect to M _2, further calculates the delay time required for the speaker of the audio signal outputted from the microphone M _1, M ₂ are the same phase. Delay time control unit 15
Controls the signal delay units 12a and 12b so that the delay time calculated by the microphone directivity calculation unit 16 is reached.

The operation of the speech recognition apparatus of this embodiment will be described below. In the following description, it is assumed that the position of the speaker's mouth is constant. FIG. 5 is a schematic diagram for explaining the method of calculating the delay time. The position of the rotation axis of the sun visor is the origin O (0,0), and the microphones M ₁ and M
The mounting position of ₂ (microphone array) is m ₀ (a ₀ ,
b ₀ ), the speaker position is V (a ₂ , b ₂ ), and the direction of the speaker's mouth with respect to the microphone array is θ ₀ . Here, it is assumed that the sun visor is arranged horizontally. Therefore, a ₀ = 0, and the distance from the rotation axis of the sun visor to the microphone array is b ₀ .

In this case, the following equation (2) is established from the definition of the inner product of the vectors.

[0021]

[Equation 2]

Since the position V (a ₂ , b ₂ ) of the speaker's mouth and the position m ₀ (a ₀ , b ₀ ) of the microphone array are known, the direction of the speaker with respect to the microphone array is expressed by the equation (2). θ
₀ can be calculated. After that, the delay time required to calculate the difference L in the distance from the speaker to the microphones M ₁ and M ₂ from the relationship shown in equation (1) and selectively pick up the voice from the direction of the speaker. Calculate (L / C). By setting this delay time in the signal delay unit 12b, the voice from the direction of the speaker can be selectively picked up.

FIG. 6 shows the sun visor rotated from the position shown in FIG. 5 by an angle θ. Here, assuming that the position of the microphone array is m ₁ (a ₁ , b ₁ ) and the direction of the speaker's mouth with respect to the microphone array is θ ₁ , the following expression (3) is established from the definition of the inner product of the vector.

[0024]

[Equation 3]

The X coordinate a ₁ of the microphone array is a ₁ = b ₀ co
Since s θ and the Y coordinate b ₁ are b ₁ = b ₀ sin θ, and the position V (a ₂ , b ₂ ) of the speaker's mouth is known, the direction of the speaker with respect to the microphone array is calculated from the equation (3). θ ₁ can be calculated. After that, the delay time required to calculate the difference L in the distance from the speaker to the microphones M ₁ and M ₂ from the relationship shown in equation (1) and selectively pick up the voice from the direction of the speaker. To calculate. By setting this delay time in the signal delay unit 12b, the voice from the direction of the speaker can be selectively picked up.

In this way, the microphone array (microphone M ₁ ,
If the delay times of the signal delay units 12a and 12b are adjusted according to the angle of the sun visor 19 to which M ₂ ) is attached,
The voice from the direction of the speaker can always be selectively picked up. In the voice recognition processing unit 14, the voice signal output from the adder 13 is subjected to voice recognition processing and converted into character data,
Output to the navigation device body (not shown). In the main body of the navigation device, the character data output from the voice recognition processing unit 14 is input and a process according to the character data is executed.

As described above, the voice recognition device of the present embodiment can selectively pick up the voice signal from the mouth direction of the speaker even if the sun visor 19 is moved. A signal having a relatively small noise component is input. This prevents the voice recognition rate from decreasing. In the above embodiment, the microphone directivity calculation unit 16 calculates the direction of the speaker's mouth with respect to the microphone from the rotation angle θ of the sun visor 19. However, instead of the microphone directivity calculation unit 16, a data storage unit that stores a table showing the relationship between the rotation angle θ of the sun visor and the direction of the speaker's mouth with respect to the microphone may be provided. In this case, when the angle θ is input from the angle sensor 18, the data storage unit outputs the direction of the speaker's mouth to the microphone corresponding to the angle θ using the table. The delay time control unit 15 controls the delay time of the signal delay units 12a and 12b based on the direction information of the speaker mouth position with respect to the microphone.

In the above embodiment, the case where the delay sum array for taking the delay sum of the outputs of the two microphones is used has been described, but by using the delay sum array of three or more, the direction of the speaker can be improved. Can be emphasized with even higher accuracy. (Second Embodiment) FIG. 7 is a block diagram showing the configuration of a speech recognition apparatus according to the second embodiment of the present invention. In the present embodiment, by using a subtraction type array that forms a blind spot (direction with low sensitivity) in the direction of the noise source, when the direction of the noise source can be specified, the voice from the direction of the speaker is selected. The sound can be picked up. This embodiment also shows an example in which the present invention is applied to a voice recognition device of a vehicle-mounted navigation device.

The speech recognition apparatus of this embodiment has two omnidirectional microphones M ₁ and M ₂ , a signal delay unit 22, a subtractor (signal synthesizer) 23, a speech recognition processing unit 24, The delay time control unit 25, the microphone directivity calculation unit 26, the seat position sensor 27, and the angle sensor 28 are included. The microphones M ₁ and M ₂ are similar to those in the first embodiment.
The sun visors (not shown) in front of the driver's seat are mounted separately from each other. In this example, microphone M ₁ is microphone M _1.
It is placed closer to the speaker compared to ₂ . The output of the microphone M ₁ is directly input to the subtractor 23, and the output of the microphone M ₂ is input to the subtractor 23 via the signal delay unit 22. The subtractor 23 outputs a signal obtained by subtracting the output of the signal delay unit 22 from the output of the microphone M ₁ . The voice recognition processing unit 24 performs voice recognition processing on the signal output from the subtractor 23 and outputs character data. This character data is sent to the navigation device body (not shown).

The seat position sensor 27 detects the amount of change in the position of the speaker (or his / her mouth) when the driver's seat is slid in the longitudinal direction of the vehicle. Further, the angle sensor 28 outputs a signal according to the angle of the sun visor to which the microphones M ₁ and M ₂ are attached. Similar to the first embodiment, the microphone directivity calculation unit 26 obtains the direction in which the noise source is generated with respect to the microphones M ₁ and M ₂ from the output of the angle sensor 28, and delays necessary to attenuate the noise component. Calculate time.
The delay time control unit 25 controls the signal delay unit 22 so that the delay time calculated by the microphone directivity calculation unit 26 is reached.

The operation of the speech recognition apparatus having the above-mentioned structure will be described below. In the present embodiment, the microphone M ₁ is used as the reference microphone, and the delay time of the signal delay device 22 connected to the microphone M ₂ is set according to the direction of the noise source. For example,
When removing noise from the direction of the engine room, noise from the direction of the engine room detected by the microphone M ₁ and noise from the direction of the engine room detected by the microphone M ₂ are input to the subtractor 23 at the same timing. Thus, the delay time of the signal delay unit 22 is set.

The subtractor 23 subtracts the output of the signal delay unit 22 from the output of the microphone M ₁ to output an audio signal in which the noise component from the direction of the noise source is suppressed (reduced). This output signal is input to the voice recognition processing unit 24 and subjected to voice recognition processing. The character data output from the voice recognition processing unit 24 is input to the navigation device body, and the navigation device body executes processing according to the character data input from the voice recognition processing unit 24.

On the other hand, the angle sensor 28 outputs a signal corresponding to the angle of the sun visor, as in the first embodiment. The microphone directivity calculation unit 26 obtains the displacements of the microphones M ₁ and M ₂ based on the output of the angle sensor 28, and determines the microphones M ₁ and M.
Calculate the direction of the noise source with respect to ₂ . The delay time control unit 25 controls the signal delay unit 22 so that the delay time calculated by the microphone directivity calculation unit 26 is reached.

In this embodiment, the angle sensor 28 detects the angle of the sun visor to which the microphones M ₁ and M ₂ are attached, and the delay time of the signal delay device 22 is controlled according to the detection result. Even if the visor is moved, a blind spot is always formed in the direction of the noise vocal source. As a result, in the present embodiment as well, a decrease in the voice recognition rate in the voice recognition processing unit 24 is avoided, as in the first embodiment.

(Third Embodiment) The third embodiment of the present invention will be described below.
The voice recognition device according to the embodiment will be described. In the present embodiment, a Griffith-Jim type array is used to selectively extract voice from the direction of the speaker. This embodiment also shows an example in which the present invention is applied to a voice recognition device of a vehicle-mounted navigation device.

FIG. 8 is a block diagram showing the structure of a speech recognition apparatus according to the third embodiment of the present invention. The voice recognition device according to the present embodiment includes four omnidirectional microphones M _{1 to} M ₄ ,
Four signal delay units D1 to D4, four subtractors (signal combiners) S0 to S3, and three adaptive filters W1 to W3
, Two adders (signal synthesizers) A1 and A2, a voice recognition processing unit 34, a delay time control unit 35, a microphone directivity calculation unit 36, a seat position sensor 37, and an angle sensor 38.
It is composed of and.

Mike M₁~ M_FourBoth are in front of the driver's seat
Mounted separately on a sun visor (not shown)
To be The signal delay devices D1 to D4 respectively correspond to
Ku M ₁~ M_FourIs connected to the microphone M₁~ M_FourFrom
The output signal is delayed and output. These signal delays
The delay time of the delay time of the units D1 to D4 is controlled by the delay time control unit 35.
Individually controlled. Mike M₁Is the reference microphone,
Mike M₁The output of the signal delay device D1 connected to
It is transmitted to the calculator S1 and also to the subtractor S0.
Be done.

The subtractor S1 subtracts the output of the signal delay device D2 from the output of the signal delay device D1 to obtain an adaptive filter W.
Output to 1. The subtracter S2 subtracts the output of the signal delay device D3 from the output of the signal delay device D2, and outputs the signal obtained by the adaptive filter W.
Output to 2. The subtractor S3 outputs a signal obtained by subtracting the output of the signal delay device D4 from the output of the signal delay device D3 to the adaptive filter W2.

The adder A2 adds a signal obtained by adding the output of the adaptive filter W2 and the output of the adaptive filter W3 to the adder A1.
Communicate to. Further, the adder A1 transmits the signal obtained by adding the output of the adaptive filter W1 and the output of the adder A2 to the subtractor S0. The subtractor S0 transfers the signal obtained by subtracting the output of the adder A1 from the output of the signal delay unit D1 to the voice recognition processing unit 24.

The adaptive filters W1 to W3 use the output of the subtractor S0 as an error and sequentially change the filter coefficient by an adaptive algorithm so that the error power is minimized. On the other hand, the angle sensor 38 detects the angle of the sun visor to which the microphones M _{1 to} M ₄ are attached, as in the _first embodiment. The microphone directivity calculation unit 36 obtains the displacements of the microphones M _{1 to} M ₄ based on the output of the angle sensor 38, and calculates the direction of the speaker mouth position with respect to the microphones M _{1 to} M ₄ . The delay time control unit 35 sets the signal delay units D1 to D4 so that the delay time calculated by the microphone directivity calculation unit 36 is obtained.
To control.

The operation of the speech recognition apparatus of this embodiment will be described below. The microphone M ₁ picks up the speaker's voice and ambient noise, and outputs a signal in which a noise component is superimposed on the voice signal. Further, the microphone M ₂ picks up the voice of the speaker and ambient noise and outputs a signal in which a noise component is superimposed on the voice signal. Therefore, when the delay times of the signal delay units D1 and D2 are appropriately adjusted, the subtractor S1 outputs a signal of only the noise component from which the audio signal has been removed. The signal of only this noise component is input to the adaptive filter W1.

Similarly, when the delay times of the signal delay units D3 and D4 are properly adjusted, only the noise component from which the audio signal has been removed is output from the subtracters S2 and S3.
Signals containing only these noise components are adaptive filters W respectively.
2, W3 is input. The subtractor S0 is a signal delay device D1.
The signal obtained by adding the outputs of the adaptive filters W1 to W3 is subtracted from the output signal of. At first, the noise component is superimposed on the audio signal at the output of the subtractor S0, but the adaptive filter W1
Since ~ W3 works so as to minimize the output of the subtractor S0, the audio signal from which the noise component is removed is output from the subtractor S0 after a sufficient time has elapsed.

In the voice recognition processing unit 34, since the voice signal from which the noise component has been removed is input from the subtractor S0, the recognition performance of voice recognition does not deteriorate as compared with that in a noise-free environment. By the way, in order to prevent the outputs of the subtracters S1 to S3 from including the audio signal, it is necessary to appropriately control the delay times of the signal delay units D1 to D4 as described above. When the angle of the sun visor of the microphone M ₁ ~M ₄ is attached changes, changes also suitable delay. Therefore, the microphone directivity calculation unit 36 calculates an appropriate delay time for each of the signal delay units D1 to D4 from the angle of the sun visor detected by the angle sensor 38, and the delay time control unit 35 calculates the appropriate delay time according to the calculation result. The signal delay devices D1 to D4 are controlled.

As a result, even if the sun visor is moved, the voice from the direction of the speaker can always be selectively picked up, and the reduction in the voice recognition rate in the voice recognition processing section 34 can be avoided. In addition, in the third embodiment, the case where the _four microphones M _{1 to} M ₄ are used has been described.
By increasing the number of microphones, the voice from the direction of the speaker can be extracted with higher accuracy.

In each of the first to third embodiments, the case where the voice signal extraction means is composed of the signal delay unit and the delay time control unit has been described. The audio signal extracting means is not limited to the above configuration. For example, as an audio signal extraction means, the output signal from each microphone is divided for each frequency,
The sound from the direction of the speaker may be extracted by calculating the time difference and the level difference for each frequency.

Further, in each of the first to third embodiments, the case where the present invention is applied to the voice recognition device of the vehicle-mounted navigation device has been described, but the present invention is applied to the voice recognition device for the vehicle. It is not limited. INDUSTRIAL APPLICABILITY The present invention can be applied to various voice recognition devices used in an environment where the relative position between a microphone and a speaker changes.

[0047]

As described above, according to the present invention,
The output of a plurality of microphones arranged apart from each other is subjected to signal processing by the voice signal extraction means to enhance the signal by the voice from the direction of the speaker or to attenuate the noise component, Since the voice signal from the direction is selectively extracted, the voice recognition performance of the voice recognition processing unit in a noise environment does not deteriorate as compared with that in a noise-free environment. Further, since the voice signal extraction means executes signal processing according to the change in the position of the microphone detected by the microphone position change amount detection means, even if the position of the microphone changes, the voice recognition rate in the voice recognition processing section is increased. Is reduced.

[Brief description of drawings]

FIG. 1 is a diagram (No. 1) showing the principle of a speech recognition apparatus according to a first embodiment of the present invention.

FIG. 2 is a diagram (part 2) showing the principle of the voice recognition device in the first embodiment of the present invention.

FIG. 3 is a block diagram showing a configuration of a voice recognition device according to the first embodiment.

FIG. 4 is a schematic diagram showing an example of a mounting portion of an angle sensor.

FIG. 5 is a schematic diagram (No. 1) for explaining a delay time calculation method.

FIG. 6 is a schematic view (No. 2) for explaining a delay time calculation method.

FIG. 7 is a block diagram showing a configuration of a voice recognition device according to a second embodiment of the present invention.

FIG. 8 is a block diagram showing a configuration of a voice recognition device according to a third embodiment of the present invention.

[Explanation of symbols]

12a, 12b, 22, D1 to D4 ... Signal delay device, 13, A1, A2 ... Adder (signal synthesizer), 14, 24, 34 ... Speech recognition processing unit, 15, 25.35 ... Delay time control unit, 16, 26, 36 ... microphone directivity calculation unit, 17, 27, 37 ... seat position sensor, 18,28,38 ... angle sensor, 23, S0 to S3 ... subtractor (signal combining unit), M ₁ ~M ₄ …Microphone.

Front page continuation (51) Int.Cl. ⁷ Identification code FI theme code (reference) G10L 15/20 G10L 3/02 301D 21/02 301E

Claims (8)

[Claims] 1. A plurality of microphones arranged apart from each other, a microphone position change amount detecting means for detecting a position change amount of the plurality of microphones, and a microphone detected by the microphone position change amount detecting means. A voice signal extraction means for inputting a position change amount and selectively extracting a voice signal from the direction of the speaker from the outputs of the plurality of microphones by utilizing the delay of the signal according to the difference in the position of each microphone. A voice recognition processing unit that performs voice recognition processing on the voice signal extracted by the voice signal extraction means. 2. The voice recognition device according to claim 1, wherein each of the plurality of microphones is an omnidirectional microphone. 3. The voice signal extracting means selectively extracts a voice signal from the direction of the speaker by emphasizing a voice from the direction of the speaker by using outputs of the plurality of microphones. The voice recognition device according to claim 1, wherein 4. The voice signal extracting means selectively extracts voice signals from the direction of the speaker by using outputs from the plurality of microphones to attenuate noise components from a predetermined direction. The voice recognition device according to claim 1, wherein the voice recognition device is a voice recognition device. 5. The audio signal extraction means is connected between a signal synthesizer for synthesizing outputs of the plurality of microphones, and between the signal synthesizer and at least one of the plurality of microphones. A signal delay unit, a delay time determination unit that determines a delay time of the signal delay unit based on a detection result of the microphone position change amount by the microphone position change amount detection unit, and a delay time determined by the delay time determination unit. The speech recognition apparatus according to claim 1, further comprising a delay time control unit that controls the signal delay device. 6. The delay time determination unit calculates the direction in which the speaker is located with respect to the plurality of microphones by calculation based on the detection result of the microphone position change amount detection means, and determines the delay time. The voice recognition device according to claim 5, wherein 7. The voice recognition device according to claim 1, wherein at least one of the plurality of microphones is attached to a sun visor in front of a driver's seat of an automobile. 8. The voice recognition device according to claim 7, wherein the microphone position change amount detection means is a sensor that detects an angle of the sun visor.