JP2000322074A - Voice input section determination device, aural data extraction device, speech recognition device, vehicle navigation device and input microphone - Google Patents

Abstract

PROBLEM TO BE SOLVED: To remove effectively a noise component from an input signal even in a noisy environment, to execute excellently the determination of a voice input section, and to improve a recognition ratio of a speech of a speaker. SOLUTION: A voice microphone 19 and a noise microphone 20 are arranged in the same microphone assembly so as to have different directivities by 180 degrees respectively. A frame dividing part 29 generates aural signal data and noise signal data based on input signals of the voice microphone 19 and the noise microphone 20, and a determination part 39 of a voice section determination device 30 determines an input section of a voice relative to the aural signal data based on the difference of short-time powers between the aural signal data and the noise signal data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention improves the recognition rate of a speech signal by removing a noise signal component from an input signal in which a speech signal component to be recognized and a noise signal component are mixed. The present invention relates to a voice input section determination device, a voice data extraction device, a voice recognition device, a vehicle navigation device including the voice recognition device, or an input microphone used in any of these devices.

[0002]

2. Description of the Related Art A speech recognition apparatus is used, for example, in a car navigation apparatus, and recognizes a voice uttered by an occupant of a car to set a destination, for example, and performs setting. In this case, the speech recognition device compares the input speech with a plurality of comparison target patterns stored in advance, and determines the one having the highest degree of coincidence as the recognition result.

[0003] For example, as a conventional technique for determining a section where a voice is being input for a signal input to an input microphone, as shown in FIG. 16, the power of an input signal is set in advance. In some cases, a section exceeding a threshold is determined as a speech input section. However, in a car under the actual use environment of the car navigation device, it is difficult to avoid the noise component when the voice uttered by the speaker is input, so that it is difficult to determine the voice input section. there were.

Therefore, for example, as one of the methods for removing a noise component from an input signal after determining a voice input section, a spectrum subtraction method (for example, STEVEN F BOL) is used.
L, Suppression of Acoustic Noise in Speech Using Sp
ectral Subtraction, IEEE Tran.on Acoustics, Speech a
nd Signal processing, Vol.Assp-27, No.2 April 1979, p
p.113-120). The spectrum subtraction method divides input audio data into short-time frames, and subtracts the noise spectrum estimated in advance from the spectrum of the input audio on which noise is superimposed for each frame, to obtain the original audio data. The spectrum is estimated.

[0005]

However, since the noise estimated in advance is estimated by referring to the input state immediately before the input of the voice, the voice to be recognized is actually input. There is a problem that the noise does not coincide with the noise generated in the section.

Therefore, as disclosed in Japanese Patent Application Laid-Open No. 4-245300, a microphone for voice input and a microphone for noise input are separately provided, and a noise component is estimated based on data obtained from both. , There is one that subtracts in real time. However, even in this method, there is a problem that if the positions of both microphones are too close, sound is mixed into the noise input microphone. Further, if the positions of both microphones are too far apart, the correlation between voice and noise will be lost, and it will not be possible to accurately remove the noise component.

The present invention has been made in view of the above circumstances, and an object of the present invention is to enable a noise component to be effectively removed from an input signal even in a noisy environment and to make a good determination of a voice input section. Voice input section determination device,
It is an object of the present invention to provide a voice recognition device capable of improving the recognition rate of a speaker's voice, a vehicle navigation device including the voice recognition device, and an input microphone used in any of these devices.

[0008]

According to the first aspect of the present invention, the voice signal data generating means includes a voice signal based on an input signal of a voice microphone whose directivity is set to a direction of a speaker. The signal data is generated, and the noise data generating means generates noise signal data based on the input signal of the noise microphone set so that the directivity is different from that of the voice microphone. Then, the voice input section determining means determines a voice input section in the voice signal data based on a difference between the voice signal data and the noise signal data.

That is, when a speaker utters a voice, noise is input to the voice microphone together with the voice of the speaker, and noise is exclusively input to the noise microphone having different directivity. By taking the difference from the signal data, the noise component contained in the audio signal data is removed, and the audio component of the speaker remains. Therefore, based on the remaining voice component, it is possible to reliably determine a section in the voice signal data where the voice of the speaker is input.

According to the second or third aspect of the present invention, the directivity of the noise microphone is approximately 180 degrees (claim 2) or approximately 90 degrees (claim). 3) Since the settings are made differently, the interference between the voice microphone and the noise microphone can be minimized, and the correlation between the signals obtained from both microphones can be maintained to some extent. Determination accuracy can be improved.

[0011] According to the voice input section determining apparatus, the difference amount detection means is input to the voice microphone and the noise microphone in advance before the voice input section is actually determined. The directivity control means detects a difference amount based on the noise signal, and controls the relative directivity between the voice microphone and the noise microphone by a directivity changing mechanism so that the difference amount is maximized. I do.

Here, a simple addition / subtraction is performed, where S (f) is a voice signal component input to the voice microphone, N1 (f) is a noise signal component, and N2 (f) is a noise signal component input to the noise microphone. Considering the model, the difference D (f) when determining the voice input section is: D (f) = (S (f) + N1 (f)) − α (f) · N2
(F). Where f is a frequency parameter and a coefficient α
(F) is a coefficient for adjusting so that D (f) = 0 when the audio signal component S (f) is not input to the audio microphone.

Therefore, as for the signal to be input to the noise microphone, N1 (f) is input without inputting S (f) as much as possible.
It is preferable to obtain N2 (f) with high accuracy by obtaining N2 (f) similar to the above at a higher level than N1 (f). That is, the difference amount Dn of the noise signal
If (f) is defined as Dn (f) = N2 (f) -N1 (f), it is desirable that Dn (f) be the largest. As a result, when determining the voice input section,
N1 (f) input together with S (f) to the voice microphone side
Can be effectively suppressed, and the difference signal D (f) = S (f) makes it possible to satisfactorily extract the audio signal S (f), thereby determining the audio input section with high accuracy. Can be.

According to the fifth aspect of the present invention, the voice input section determining means determines that the difference amount detected by the difference amount detecting means among a plurality of pairs of voice microphones and noise microphones is maximum. The voice input section is determined based on the input signals respectively input to the input devices. Therefore, similarly to the fourth aspect, the voice input section is determined with high accuracy based on the input signal obtained from the pair of the voice microphone and the noise microphone in which the difference amount of the noise signal is maximized according to the actual use environment. Can be.

According to a sixth aspect of the present invention, the voice input section determining means determines a voice input section based on a difference between average powers of voice signal data and noise signal data. For example, when each of the audio signal data generating means and the noise signal data generating means generates each signal data based on the input signal of each microphone, it is generally common to handle the input signal by dividing it as a frame at predetermined intervals. Is being done.

Therefore, if the average power is calculated for each frame for the voice signal data and the noise signal data and the difference between the two is calculated, the difference value for the frame containing the voice of the speaker becomes significantly large. Further, since the power is obtained by the square of the input signal amplitude, the dynamic range is larger than in the case where the amplitude levels are directly compared. Therefore, the voice input section can be easily determined.

According to the seventh aspect of the present invention, the voice input section determining means determines the voice input section based on a difference in power spectrum between the voice signal data and the noise signal data. That is, by obtaining the power spectrum for each signal data, a characteristic spectrum envelope appears in the audio data for a frame including the audio data. Therefore, the voice input section can be easily determined by detecting a portion including such a characteristic spectrum envelope.

According to the voice data extracting device of the present invention, the voice signal data generating means generates voice signal data based on an input signal of a voice microphone whose directivity is set to a direction of a speaker, and generates noise signal. The data generating means generates noise signal data based on the input signal of the noise microphone set so that the directivity is different from that of the voice microphone. Then, the audio data extracting means extracts audio data from the audio signal data based on a difference between the audio signal data and the noise signal data.

That is, when a speaker emits a voice, noise is input to the voice microphone together with the voice of the speaker, and noise is exclusively input to the noise microphone having different directivity. By taking the difference from the signal data, the noise component contained in the audio signal data is removed, and the audio component of the speaker remains. Therefore, it is possible to reliably extract the voice data of the speaker included in the voice signal data.

According to the audio data extracting apparatus of the ninth or tenth aspect, the directivity of the noise microphone is approximately 180 degrees (claim 9) or approximately 90 degrees (claim 10) with respect to the directivity of the audio microphone. ) Since the settings are made differently, it is possible to minimize the interference between the voice microphone and the noise microphone, and to maintain the correlation between the signals obtained from both microphones to some extent. The extraction accuracy can be improved.

According to the audio data extracting device of the eleventh aspect, the directivity control means uses the directivity changing mechanism and the voice microphone and the noise control means so that the difference amount detected by the difference amount detecting means is maximized. Control is performed so that the directivity with the microphone is relatively changed. Therefore, according to the actual use environment, the audio data can be extracted by adjusting the difference between the noise signals input to the audio microphone and the noise microphone to a state where the difference amount becomes maximum. Can be extracted with high accuracy.

According to the twelfth aspect of the present invention, the audio data extracting means has a maximum difference amount detected by the difference amount detecting means among a plurality of pairs of audio microphones and noise microphones. The audio data is extracted based on the input signals respectively input to the objects. Therefore, as in the eleventh aspect, audio data is extracted with high accuracy based on an input signal obtained from a pair of an audio microphone and a noise microphone in which the difference amount of the noise signal is maximized according to the actual use environment. be able to.

According to a thirteenth aspect of the present invention, the audio data extracting means extracts the audio data based on a difference in power spectrum within a predetermined range between the audio signal data and the noise signal data. That is, when a power spectrum is obtained for each signal data, a frame including audio data appears in a state where a characteristic spectrum envelope is superimposed on the audio data. Therefore, by subtracting the spectrum envelope of the noise data therefrom, audio data can be extracted with high accuracy.

According to a fourteenth aspect of the present invention, the voice data extracting device includes the voice input section determining device according to any one of the first to seventh aspects. By extracting the voice data included in the voice input section determined by the section determining means, the voice data can be easily extracted.

According to the voice data extracting device of the present invention, the voice microphone, the noise microphone, the noise data generating means and the input signal data generating means are configured in common with the respective elements of the voice input section determining device. , The whole can be made small.

According to the speech recognition apparatus of the sixteenth aspect,
By providing the voice input section determination device according to any one of claims 1 to 7, the voice recognition section analyzes voice data included in the voice input section determined by the voice input section determination section and performs voice recognition. Since the data pattern is recognized, the voice data pattern can be easily recognized based on the voice input section determined with high accuracy.

[0027] According to the speech recognition apparatus of the seventeenth aspect,
By providing the voice data extraction device according to any one of claims 8 to 15, the voice recognition unit recognizes a voice data pattern by analyzing the voice data extracted by the voice data extraction unit of the voice data extraction device. Therefore, it is possible to easily recognize the voice data pattern based on the voice data extracted with high accuracy.

[0028] According to the vehicle navigation apparatus of the eighteenth aspect, there is provided the voice recognition apparatus of the sixteenth or seventeenth aspect, and executes a predetermined process based on the voice data pattern recognized by the voice recognition means. In other words, even in an environment where the noise level is relatively high, such as when driving a vehicle, it is possible to eliminate the influence of the noise as much as possible and to reliably recognize the data pattern of the voice emitted by the occupant of the speaker vehicle. It becomes possible. Therefore, the occupant can reliably perform a desired operation by voice input without manually operating an input key or the like, thereby improving convenience and driving safety.

According to the input microphone of the nineteenth aspect,
The voice microphone and the noise microphone are arranged in the same housing and integrally formed, and the shape of the housing is formed such that the voice microphone and the noise microphone are apparently directed in the same direction.

That is, the speech input section determination device according to any one of claims 1 to 7, the speech data extraction device according to any one of claims 8 to 15, or the speech data extraction device according to claim 16 or 17.
In the described voice recognition device, the voice microphone and the noise microphone need to be arranged so as to have different directivities from each other.

In this case, in consideration of the installation work of each microphone, it is desirable that the voice microphone and the noise microphone are arranged in the same housing to be integrally formed. However, at this time, if the shape of the housing is formed in accordance with the different directivities of the respective microphones, the worker may install the microphone in any direction (directivity) during the microphone installation work. May not be easily determined.

Therefore, by forming the shape of the housing such that the voice microphone and the noise microphone are apparently directed in the same direction, the operator can direct the voice microphone in the direction of the speaker. If the microphone is mounted in such a manner, the directivity of the noise microphone is naturally set to be different from that of the voice microphone, so that the microphone can be easily installed and workability is improved.

[0033]

DESCRIPTION OF THE PREFERRED EMBODIMENTS (First Embodiment) A first embodiment in which the present invention is applied to a car navigation system will be described below with reference to FIGS. In FIG. 1, which shows the overall electrical configuration of a car navigation system (car navigation system) 1, the car navigation system 1 includes a position detector 2, a map data input device 3, an operation switch group 4, a control circuit 5 to which these are connected, and the like. I have. Further, the control circuit 5 includes an external memory 6, a display device 7 such as a display, a remote control sensor 8
And a voice recognition device 9 are connected. Note that the control circuit 5 is configured as a microcomputer, and includes a well-known CPU,
ROM, RAM, I / O, etc. are provided. These components are electrically connected to each other via a bus line.

The position detector 2 is a well-known geomagnetic sensor 1
0, gyroscope 11, distance sensor 12, and GP
A GPS receiver 13 for detecting the position of the vehicle based on radio waves from an S (Global Positioning System) satellite is provided. Each of the sensors 10 to 13 generates an error having a different property.
By referring to the information from, each information is used while making corrections as necessary. Note that, depending on the required accuracy, a configuration may be made using a part of the sensors 10 to 13 or a steering sensor, a wheel sensor, or the like may be added.

The map data input device 3 is a device for inputting various data including map data and landmark data, so-called map matching data for improving position detection accuracy, and the like. CD-R as data storage medium
Although OM is generally used, DVD (Digital Ver.
satile Disk) or a memory card may be used.

On the screen of the display device 7, map data input from the map data input device 3 is displayed, and a marker indicating the current position of the vehicle input from the position detector 2 is displayed on the map data. In addition, additional data such as a guidance route and a mark of a set point are displayed in a superimposed manner.

When the position of the destination is input from the remote control sensor 8 via a remote control terminal (hereinafter, referred to as a remote controller) 14 or via the operation switch group 4, the car navigation system 1 starts from the current position. It also has a so-called route guidance function that automatically selects the optimal travel route to the destination and displays it as a guidance route. As a method for automatically setting an optimum route, for example, the Dijkstra method is known. The operation switch group 4 is
For example, a touch switch or a mechanical switch formed integrally with the display device 7 is used, and is used for various input operations.

The voice recognition device 9 is used when the operation switch group 4 and the remote controller 14 are manually operated to indicate a destination or the like, and similarly when the user performs voice input. It is arranged so that a destination or the like can be instructed.

The voice recognition device 9 includes a voice recognition unit (voice recognition means) 15, a dialogue control unit 16, a voice synthesis unit 17, a voice extraction device 18, a voice microphone 19, and a noise microphone 2.
0, PTT (Push To Talk) switch 21, speaker 22
And a control unit 23.

When the speech recognition unit 15 performs speech recognition processing on the speech data supplied from the speech extraction device 18 in accordance with the instruction of the dialog control unit 16, the speech recognition unit 15 outputs the recognition result.
6 is output. That is, the voice recognition unit 15
Compares and compares the voice data obtained from the voice extraction device 18 with a plurality of comparison target pattern candidates registered in the dictionary data stored in advance, and determines a higher comparison target pattern having a high degree of coincidence with the dialog control unit. 16 is output.

The recognition of a word sequence in the input voice is performed by sequentially analyzing the voice data by acoustic analysis, for example, by LPC (Linear Prediction).
Coefficient) Acoustic features such as cepstrum are extracted, and their time series data is obtained. Then, the obtained time series data is divided into several sections, and each section is stored as dictionary data using a known DP matching method, HMM (Hidden Markov Model), or a method such as a neural network. Ask if they are compatible with

The dialog control unit 16 instructs the voice synthesizing unit 17 to generate a response voice from the result of the voice recognition and the internal state managed by the control unit 5, and controls the overall control of the car navigation system 1.
For example, a destination required for the navigation process is notified and an instruction to execute the setting process is issued. That is, if this voice recognition device 9 is used,
The user can give a destination instruction or the like to the car navigation system 1 by voice input without operating the operation switch group 4 or the remote controller 14.

As shown in FIG. 2, the audio microphone 19 is
It is provided for inputting a voice uttered by the user, and its directivity is centered toward the speaker (user) 24. The noise microphone 20 is provided for inputting noise existing as a background of a voice uttered by the user, and has a directivity center opposite to the direction of the speaker (user) 24 ( (Directions differing by 180 degrees horizontally). The voice microphone 19 and the noise microphone 20 are arranged inside one microphone assembly (housing) 25 to constitute an integrated input microphone 60.

FIGS. 3A and 3B show the appearance of the microphone assembly 25. FIG. 3A is a front view, FIG.
(C) is a plan view showing a disassembled state. At the left and right ends of the microphone assembly 25, storage sections 25A in a rectangular box state for storing the audio microphone 19 and the noise microphone 20, respectively.
25B, and has a shape in which the storage portions 25A and 25B are supported by a prismatic support portion 25C.

As shown in FIGS. 3C and 2, inside the microphone assembly 25, the voice microphone 19 is disposed toward the speaker 24, and the noise microphone 20 is disposed.
Is arranged facing the speaker 24 side. However, the appearance of the microphone assembly 25 is
9 and 20 are also formed so as to be apparently directed toward the speaker 24 (the front side shown in FIG. 3A).

An opening 25D for inputting an acoustic signal is provided on the front side of the storage section 25A in which the audio microphone 19 is stored.
The dummy opening is provided on the front side of the storage section 25B in which the noise microphone 20 is stored. As for the storage section 25B, FIG.
A slit 25E for inputting an acoustic signal is provided on the rear side shown in FIG.
Is provided, so that noise outside the microphone assembly 25 can be input.

The input microphone 6 configured as described above
0, as shown in FIG. 4, is disposed on the steering column 26 disposed on the driver's seat side of the automobile such that the front side faces the driver as the user. That is, by forming the shape of the microphone assembly 25 as shown in FIG. 3, the worker can simply mount the microphone assembly 25 without considering the directivity of the internal microphones 19 and 20 when mounting the microphone assembly 25. The input microphone 60 may be arranged such that the front side faces the driver. On the left side of the steering column 26, a PTT switch 21 constituted by a momentary push switch is arranged.

Referring again to FIG. 1, voice extracting device 18
When the control unit 23 detects that the PTT switch 21 has been turned on by the user, as described later, the voice uttered by the driver based on the signals input to the voice microphone 19 and the noise microphone 20 respectively. Are extracted as digital data and output to the voice recognition unit 15. The control unit 23 gives an instruction to the amplifier 27 of the in-vehicle audio device to temporarily reduce the volume of the audio signal output to the audio speaker 28 to almost zero (mute), or It is also configured to be able to give an instruction to return the volume of the audio signal to the original state (unmute).

Here, a detailed configuration of the voice extracting device 18 will be described with reference to FIG. Voice extraction device 18
Is a frame division unit (sound signal data generation means, noise signal data generation means) 29, a speech section determination device (speech input section determination device) 30, a speech buffer 31, a noise buffer 32, a speech Fourier transform unit 33 , A noise Fourier transform unit 34 and a subtraction unit (sound data extracting means) 35.

The frame dividing section 29 includes microphones 19 and 20
The input audio signal is sampled at a rate of, for example, about several tens of KHz and A / D converted, and the obtained discrete data sequence is cut out for each section (frame) set to a time width of about several tens of ms. The voice section determination device 30 is divided and multiplied by a window function (for example, a Hanning window) for suppressing a harmonic component generated by an end of the cutout.
Output.

The voice section determination device 30 includes the voice microphone 1
9 for each frame of the audio signal input to speaker 2.
4. It is determined whether or not the signal of the uttered voice is included.
The data of the detected frame including the audio signal is output to the audio buffer 31 and accumulated. In addition, the voice section determination device 30 converts the data of the frame corresponding to the frame including the voice signal (that is, the same time) among the frame data of the audio signal input to the noise microphone 20 into the noise. Buffer 32
To be stored.

In the following, digital data of the 19 audio microphones after the frame division unit 29 will be referred to as audio signal data, and digital data of the 20 noise microphones will be referred to as miscellaneous audio signal data.

The Fourier transform units 33, 34
The short-time spectrum is generated by performing Fast Fourier Transform (FFT) on the data stored in each of the first and second data, and is output to the subtracter 35. The subtractor 35 is generated by the noise Fourier transformer 34 from the short-time spectrum (spectrum including voice and noise, input spectrum) generated by the voice Fourier transformer 33 according to the algorithm of the spectrum subtraction method. The noise is removed by subtracting the short-time spectrum (spectrum including only noise, noise spectrum). At that time, for example, the subtraction may be performed by multiplying the noise spectrum by a subtraction coefficient set to one or more.
Further, the spectrum may be any one of power and amplitude. Then, the result of the subtraction in the subtracting unit 35 is output to the speech recognition unit 15.

FIG. 6 is a functional block diagram showing a detailed configuration of the voice section judging device 30. The voice section determination device 30 includes a power calculation unit (voice signal data generation unit) 36 corresponding to the voice microphone 19 and the noise microphone 20.
Power calculation unit (noise signal data generation means) corresponding to the side
37. These power calculators 36 and 37
Calculates the power (short-time power), that is, the power of the acoustic signal, for each frame divided by the frame dividing unit 29 and outputs the calculated power to the subtracter 38.

Then, in the subtracter 38, the power calculator 3
The power calculated by the power calculation unit 37 is subtracted from the power calculated in 6, and the result of the subtraction is provided to a determination unit (speech input section determination unit) 39. Then, the judgment unit 39
If the result of the subtraction is equal to or greater than a predetermined value, it is determined that the frame input to the power calculation unit 36 at that time is a frame (voice input section) including the voice data of the speaker 24. When the gates (output buffers) 40 and 41 are opened, the frame data supplied to the power calculation units 36 and 37 from the frame division unit 29 is output to the audio buffer 31 and the noise buffer 32, respectively. ing.

Next, the operation of the present embodiment will be described with reference to FIGS. First, an outline of the operation of the car navigation 1 will be described. When the power of the car navigation 1 is turned on, an operation menu is displayed on the screen of the display device 7.
Then, the driver operates the operation switch group 4 or the remote controller 1
A similar selection instruction is given from the dialog control unit 16 to the control circuit 5 by selecting a process of operating the display unit 4 to display the guidance route on the display device 7 or by inputting a voice through the voice recognition device 9. Then, the processing is executed as follows.

That is, when the driver inputs a destination by voice or the operation of the remote controller 14 with reference to the map displayed on the display device 7, the current position of the vehicle is obtained based on the position data obtained from the GPS receiver 13. Can be And
The control circuit 5 calculates the cost using the Dijkstra method, and obtains the route that is the shortest distance from the current position to the destination as the guidance route. Then, by superimposing and displaying the obtained guidance route on the map of the display device 7, an appropriate route is guided to the driver. The process of calculating the guidance route and the process of providing guidance in this manner are generally well-known, and thus detailed description thereof will be omitted.

Next, the operation of the voice recognition device 9 will be described by taking as an example a case where the destination is input by voice for the above-mentioned route guidance. FIG. 7 is a flowchart showing the control contents of the control unit 23 in the speech recognition device 9. First, the control unit 23 determines whether or not the PTT switch 21 has been pressed (ON) (step S1).
S ", and instructs the amplifier 27 to mute the audio signal output to the audio speaker 28 (step S2).

At that time, if the amplifier 27 is outputting an audio signal to the audio speaker 28, the audio signal is muted as shown in FIGS. 19 and noise microphone 2
The level of the sound signal input to 0 decreases.

Then, the control unit 23 controls the sound extraction device 1
8 is instructed to start the processing for capturing and storing the acoustic signals input to the audio microphone 19 and the noise microphone 20 (step S3). For example, it is assumed that the speaker (driver) 24 utters and inputs the destination as "Showa-cho, Kariya city, Aichi prefecture". Then, the frame dividing unit 29
Samples audio signals input from the microphones 19 and 20 and divides them into frames, and outputs audio signal data and noise signal data to the audio section determination device 30.

At this time, noise is input to the voice microphone 19 pointing in the direction of the speaker 24 together with the voice of the speaker 24, and the noise is directed in the opposite direction (180 degrees) with respect to the speaker 24. The voice of the speaker 24 is hardly input to the noise microphone 20 and the noise is input exclusively.

When the power calculating sections 36 and 37 calculate the short-time power for each of the divided frames of the voice signal data and the noise signal data, the voice section determination device 30 Performs value subtraction. Determination unit 39 of voice section determination device 30
Outputs the data of the frame whose subtraction result is equal to or greater than a predetermined value to the audio buffer 31 and the noise buffer 32, respectively.

That is, as shown in FIG. 8, during the period from the time when the PTT switch 21 is turned on to the time when the speaker (driver) 24 makes a sound, the voice microphone 1
The noise 9 and the noise microphone 20 are input with substantially the same level of noise (for example, an engine sound or a running sound of an automobile). Therefore, when the short-time power is calculated for each of the audio signal data and the noise signal data of the frame belonging to the period-, there is almost no difference between the two, and the result of subtracting both is a low value.

During the period (voice section) in which the speaker 24 is uttering “Showa-cho, Kariya-shi, Aichi”, the voice is input to the voice microphone 19.
As shown in (b), the amplitude level of the input signal rises, but the sound is directed in the opposite direction.
Is hardly input, and the amplitude level of the input signal hardly changes.

Accordingly, the acoustic signal input to the noise microphone 20 during the period-is substantially equal to the noise input to the audio microphone 19 simultaneously with the voice. Therefore, the period-
When the short-time power is calculated for the audio signal data and the noise signal data of the frame belonging to the same manner as described above, the output value of the power calculation unit 36 becomes significantly larger than the output value of the power calculation unit 37, Is considerably high, the determination unit 39 determines that the frame is a voice input section, and outputs the frame data via the gates 40 and 41 to the voice buffer 31 and the noise buffer 32, respectively, and accumulates them.

Then, the control unit 23 waits until the PTT switch 21 is turned off (step S4), and
If the answer is F, it is determined to be "YES" and when the voice extraction device 18 is instructed to stop taking in the input signal (Step S5)
Subsequently, an instruction to start the audio extraction processing is given to the audio extraction device 18 (step S6). Then, when the amplifier 27 is instructed to release the mute of the audio signal (step S7), the process proceeds to step S1.

When a start instruction of the sound extraction processing is given from the control unit 23, the sound extraction device 18
The frame data corresponding to the audio input section stored in the buffers 31 and 32 are read by the buffers 3 and 34, a short-time spectrum is generated by FFT, and the spectrum is output to the subtracter 35.

The subtracter 35 converts the short-time spectrum SN (f, t) generated by the Fourier transformer 34 by the spectrum subtraction method.
The noise component is removed by subtracting the short-time spectrum N (f, t) of the same frame number as the former generated by the Fourier transformer 33. Then, as a result of the subtraction, a spectrum component S (f, t) corresponding to the voice data generated by the speaker 24 remains. That is, S (f, t) = SN (f, t) -N (f, t) (1). Note that f is a frequency, and t is a frame number, that is, time.

The spectrum S (f, t) of the audio data output from the subtracter 35 is output to the audio recognizer 15. Then, as described above, the speech recognition unit 15 obtains time-series data from which the acoustic features have been extracted for the spectrum S (f, t), and the time-series data is divided into sections, and the appropriate words are divided into appropriate words. The collation with the data pattern registered in the dictionary data is performed.

The recognition result is provided from the voice recognition unit 15 to the dialog control unit 16, and the dialog control unit 16 gives an instruction to generate a response voice to the voice synthesis unit 17 based on the provided recognition result. Through the speaker 22, a synthesized voice is uttered with "Showa-cho, Kariya city, Aichi prefecture". Also, the voice recognition unit 15
Is also given to the control circuit 5, and the control circuit 5
Starts the calculation of the guide route with "Showa-cho, Kariya-shi, Aichi" as the destination.

As described above, according to the present embodiment, the voice microphone 19 and the noise microphone 20 are arranged in the same microphone assembly 25 so that the directivities differ from each other by 180 degrees. Then, the directivity of the audio microphone 19 is set in the direction of the speaker 24, and the frame dividing unit 29 sets the audio microphone 19
And voice signal data and noise signal data based on the input signal of the noise microphone 20 and the voice section determination device 30.
Is determined based on the short-time power difference between the audio signal data and the noise signal data.

That is, when the speaker 24 emits voice, noise is input to the voice microphone 19 together with the voice of the speaker, and noise is exclusively input to the noise microphone 20 whose directivity is different by 180 degrees. By taking the difference between the short-term powers of the respective input signal data, the difference between the two input signal data can be remarkably detected.
The section in which the voice of No. 4 is input can be reliably determined. Then, the directivity of both microphones 19 and 20 is set to 180
By setting the microphones to be different from each other, the interference between the audio microphone 19 and the noise microphone 20 can be minimized, and the correlation between the signals input to the microphones 19 and 20 can be maintained to some extent. The accuracy of section determination can be improved.

Further, according to the present embodiment, the voice extracting device 1
8 extracts the audio data based on the difference between the power spectrum of each frame corresponding to the audio signal data and the noise signal data. Therefore, the audio envelope is obtained by subtracting the spectrum envelope of the noise data from the frame including the audio data. Data can be accurately extracted.
Then, synergy with the operation and effect of the voice section determination device 30 can further enhance the extraction accuracy. As a result, it is possible to improve the recognition accuracy of the voice in the voice recognition device 9.

The microphone assembly 25 constituting the input microphone 60 is formed such that both the microphone 19 for noise and the microphone 20 for noise seem to be directed to the speaker 24 in appearance. , The operator uses the audio microphone 19
If the input microphone 60 is attached so as to direct the direction of the speaker 24, the directivity of the noise microphone 20 is naturally set to be different from that of the voice microphone 19. In addition to the fact that the two microphones 19 and 20 are integrally arranged in the microphone assembly 25, the work of installing the microphones 19 and 20 can be easily performed, and the workability is improved. In addition, wiring can be easily arranged.

Further, according to the present embodiment, the audio microphone 1
9, the noise microphone 20, and the frame division unit 29 are shared by the voice extraction device 18 and the voice segment determination device 30, so that the whole can be made small.

In addition, since the car navigation system 1 performs a predetermined process based on the voice data pattern of the speaker 24 recognized by the voice recognition device 9, the car navigation system 1 may be used in an environment where the noise level is relatively high, such as when driving a car. However, since the influence of the noise can be eliminated as much as possible and the data pattern of the voice uttered by the speaker 24 can be reliably recognized, the processing desired by the speaker 24 can be reliably executed. it can.
Then, the driver who is the speaker 24 is the steering column 2
Since the desired operation can be performed only by operating the PTT switch 21 installed in the speaker 6 and speaking, the driving of the automobile can be performed safely.

(Second Embodiment) FIG. 9 shows a second embodiment of the present invention. In the second embodiment, the microphone assembly (housing) 25a is the same as the microphone assembly 2 in the first embodiment.
Slightly different from 5. That is, in the microphone assembly 25a, the center of the directivity of the noise microphone 20 is different from the direction of the speaker 24 by about 90 degrees horizontally, that is, in the direction of the door (not shown) on the passenger seat side. Is turned. The appearance of the microphone assembly 25a is formed in the same manner as the microphone assembly 25 shown in FIG. The above constitutes the input microphone 60a.

That is, when the microphone assembly 25a is installed on the steering column 26, the traveling sound of a passing vehicle coming from the door side on the driver's seat side is transmitted to the audio microphone 19a.
It is also conceivable to mix in In addition, when traveling on a road provided with a soundproof wall, the traveling sound of a driving automobile may be reflected from the direction of the door on the passenger seat side and arrive. Therefore, when it is assumed that the frequency of the traveling state such as the latter is relatively high, as shown in FIG. 9, the noise microphone 20 is arranged so as to be directed to the door direction on the passenger seat side.
Further, when it is assumed that the frequency of the traveling state as the former is relatively high, the noise microphone 20 may be arranged so as to be directed to the door direction on the driver's seat side.

Here, the voice signal component input to the voice microphone 19 is S (f), the noise signal component is N1 (f), and the noise signal component input to the noise microphone 20 is N2.
Considering a simple addition / subtraction model as (f), the difference amount D (f) when determining a voice input section or extracting voice data is D (f) = (S (f) + N1 (f) )) − Α (f) · N2 (f) (2) However, the coefficient α (f) is the difference D (f) = 0 (that is, N1 (f) = α when the audio signal component S (f) is not input to the audio microphone 19).
(F) · N2 (f)).

When the difference amount of the short-time power is actually calculated, the coefficient α (f) is set in advance and the calculation is performed according to the equation (2). To obtain a high level of the noise signal component N2 (f) by setting the audio signal component S (f) to zero as much as possible,
It is preferable that the dynamic range between (f) and N2 (f) be larger (S (f) << N2 (f)).

Further, in other words, to maximize the dynamic range on the noise microphone 20 side, the difference Dn (f) of the noise signal is calculated by: Dn (f) = N2 (f) -N1 (f) (3) is equivalent to maximizing Dn (f).
In this case, the ratio between N2 (f) and N1 (f) is Rn.
Rn (f) = N2 (f) / N1 (f) (4) Maximizing Rn (f) is synonymous with maximizing Dn (f). .

As a result, on the noise microphone 20 side, the noise signal component N2 (similar to N1 (f))
Since information about (f) can be obtained with high accuracy,
When the expression (2) is calculated when determining the voice input section,
The noise signal component N (f) input together with the audio signal component S (f) to the audio microphone 19 can be effectively suppressed to N1 (f), and the audio signal component S (f) is excellent as the difference D (f). Can be extracted.

Therefore, according to the second embodiment configured as described above, the noise microphone 20 is disposed so as to be directed to the door direction on the passenger seat side, so that the vehicle arriving from the door direction can travel. A noise signal such as a sound can be input to the noise microphone 20 at a high level. Then, the subtracter 38 of the voice section determination device 30 or the subtracter 35 of the voice extraction device 18 can determine the voice input section and extract the voice data with high accuracy.

(Third Embodiment) FIG. 10 shows a third embodiment of the present invention. In the third embodiment, a microphone assembly (housing) 25b is similar to the second embodiment in the first embodiment. It is slightly different from Mike Assy 25 in the example. That is, the noise microphone 20 is set inside the microphone assembly 25b.
Of the direction of the speaker 24 is approximately 9 perpendicular to the direction of the speaker 24.
They are oriented in directions different by 0 degrees, that is, in the direction of the ceiling of the cab. The appearance of the microphone assembly 25b is formed in the same manner as the microphone assembly 25 shown in FIG. The above constitutes the input microphone 60b.

That is, depending on the noise environment, it is assumed that it is desirable to set the directivity of the noise microphone 20 in this way. Therefore, according to the second embodiment configured as described above, the same effect as that of the first or second embodiment can be obtained.

(Fourth Embodiment) FIGS. 11 and 12 show a fourth embodiment of the present invention. The same parts as those of the first embodiment are denoted by the same reference numerals, and the description thereof will be omitted. Only the parts will be described. An actuator (directivity changing mechanism) 42 is disposed inside the microphone assembly (housing) 25c in the fourth embodiment, and the audio microphone 19 and the noise microphone 20 are driven by the actuator 42 to rotate in the horizontal direction. It is configured to be movable. The above constitutes the input microphone 60c. Actuator 42
Is controlled by a drive control unit (directivity control unit) 43, and the drive control unit 43 performs the control based on a subtraction result from a subtractor (difference amount detection unit) 38 in the voice section determination device 30. The drive of the actuator 42 is controlled.

Next, the operation of the fourth embodiment will be described. The control unit 23 of the voice recognition device 9 includes the PTT switch 2
1 is turned on and before the voice of the speaker 24 is input to the voice microphone 19, an instruction is given to the voice section determination device 30, and the noise signal components N1 (f) and N2 ( The short-time power f) is calculated by the power calculation units 36 and 37, respectively, to obtain the difference Dn (f) between the two, and the difference Dn (f) is output to the drive control unit 43.

Then, by repeating the process a plurality of times, the drive control unit 43 sends the audio microphone 19 and the noise microphone 19 via the actuator 42 such that the value of the difference Dn (f) becomes maximum. 20 are changed relatively. In this case, for example, the directivity of the audio microphone 19 is fixed in the direction of the speaker 24 and the noise microphone 20 is fixed.
To stop at the position where the value of Dn (f) becomes maximum. Next, the position of the noise microphone 20 is fixed, and the directivity of the voice microphone 19 is rotated within a certain range centered on the speaker 24 direction, and the position of the microphone Dn (f) is maximized. Stop it. Note that the latter process may be omitted.

As described above, according to the fourth embodiment, the drive control unit 43 controls the actuator 4 so that the difference Dn (f) of the noise signal obtained by the subtractor 38 is maximized.
2, the control is performed so that the directivity of the audio microphone 19 and the noise microphone 20 is relatively changed.

Accordingly, the audio signal S (f) is extracted with high precision as described above by adjusting the difference amount Dn (f) to a maximum value according to the actual various different use environments. It is possible for the voice segment determination device 30 to:
The voice input section can be determined with high accuracy. Also,
The accuracy of voice data extraction by the voice extraction device 18 can be improved. In addition, the voice recognition accuracy of the voice recognition device 9 can be improved.

(Fifth Embodiment) FIG. 13 shows a fifth embodiment of the present invention. The same parts as those in the first embodiment are denoted by the same reference numerals, and description thereof will be omitted. explain. In the fifth embodiment, a plurality of input microphones 60 (1), 60 (2),..., 60 (n) are used, and the plurality of input microphones 60 include the steering column 26. It is arranged at several places in the driver's seat. However, in each case, the directivity of the voice microphone 19 is arranged so as to face the speaker 24.

The audio signal input to each input microphone 60 is provided to the frame division unit 29 via the microphone selector 44, and the selection control unit (voice input section determination means) 45 Microphone selector 44 based on the result of subtraction from subtractor 38 in determination device 30
Output a select signal. And
The microphone selector 44 outputs an audio signal input to the input microphone 60 corresponding to the given select signal to the frame dividing unit 29.

Next, the operation of the fifth embodiment will be described. As in the fourth embodiment, the control unit 23 of the voice recognition device 9 sets the PTT switch 21
Is input to the selection control unit 45 to input a plurality of input microphones 60 (1), 60 (2),.
The audio signals input from each of 60 (n) are sequentially switched and output to the frame division unit 29.

Then, the voice section determination device 30 outputs the noise signal component N1 (f) obtained by each input microphone 60.
And the short-time power of N2 (f) are calculated by the power calculation unit 36, respectively.
, 37 and the difference Dn (f) is obtained and output to the selection control unit 45. Then, the selection control unit 45
Determines the input microphone 60 in which the value of the difference amount Dn (f) is maximum based on the result, and uses the input microphone 60 for subsequent voice input section determination and voice data extraction.

As described above, according to the fifth embodiment, the selection control unit 45 includes the subtractor 3 among the plurality of input microphones 60.
8, the voice section determination device 30 selects the input that maximizes the noise signal difference Dn (f) according to the actual use environment. The voice input section can be determined with high accuracy based on the input signal obtained from the microphone 60 (that is, the pair of the voice microphone 19 and the noise microphone 20). Further, the accuracy of voice data extraction and voice recognition in the voice extraction device 18 and the voice recognition device 9 can be improved.

(Sixth Embodiment) FIGS. 14 and 15 show a sixth embodiment of the present invention. The same parts as those in the first embodiment are denoted by the same reference numerals, and description thereof is omitted. Only the parts will be described. As shown in FIG. 14, in the speech extracting apparatus 46 according to the sixth embodiment, the frame dividing unit 29 according to the first embodiment is replaced by an acoustic analyzing unit (sound signal data generating means, noise signal data generating means) 47. The voice section determination device 30 is replaced with a voice section determination device 48.

The acoustic analysis unit 47 has a function of performing FFT on each frame data to which a window function has been applied, in addition to the function of the frame division unit 29. Then, the buffers 31, 32 and the Fourier transform units 33, 34 are deleted, and the voice section determination device (voice input section determination device)
The data output from 48 is directly input to the subtract unit 35.

Further, as shown in FIG. 15, the voice section determination device 48 includes a power calculation section 36 of the voice section determination device 30.
And 37 are used as power spectrum calculators (audio signal data generator and noise signal data generator) 49 and 50.
Is replaced by

Next, the operation of the sixth embodiment will be described. As in the first embodiment, when the speaker 24 speaks and inputs a destination, the sound analysis unit 47 samples the sound signal input from the microphones 19 and 20 and divides the sound signal into frames, and sets a window function. The FFT is performed by multiplication, and the audio signal data and the noise signal data obtained as the amplitude spectrum are output to the audio section determination device 48.

When the power spectrum calculators 49 and 50 calculate the power spectrum for the given voice signal data and noise signal data for each frame, the subtractor 38 outputs the short-time power spectrum. And take the difference. Judgment section (speech input section judgment means) 39a of speech section judgment device 48
Outputs the speech signal data and the noise signal data of the frame in which the spectrum envelope specific to the speech remains as a result of the difference to the subtracter 35.

Then, the subtractor 35 uses the spectrum subtraction method for the audio signal data and the noise signal data given as the amplitude spectrum through the audio section determination device 48 in the same manner as in the first embodiment. The noise component is removed from the data.

As described above, according to the sixth embodiment, the voice section determination device 48 determines the voice input section based on the difference in power spectrum between the voice signal data and the noise signal data within a predetermined frequency range. I did it. That is, since a characteristic spectrum envelope appears in the audio data for a frame including the audio data, it is easy to determine a voice input section by detecting a portion including such a characteristic spectrum envelope. Can be.
In addition, when taking the difference, the difference may be obtained in a frequency range (a predetermined frequency range, for example, about 0 to 6 kHz) where a characteristic spectrum envelope appears in the audio data.

The present invention is not limited to the embodiment described above and shown in the drawings, and the following modifications or extensions are possible. In the first embodiment, the input microphone 60 is not limited to be mounted on the steering column 26, but may be mounted on a driver's seat side door or on a dashboard. The relationship between the directivity of the audio microphone 19 and the noise microphone 20 is not limited to the one shown in the first to third embodiments. What is necessary is just to arrange it differently.
For the difference amount D (f) expressed by the expression (2), when the coefficient α is expressed by the expression (5) as a scalar amount, D (f) = (S (f) + N1 (f)) − α · N2 ( f) (5) The coefficient α is determined so that Σ | D (f) | is minimized when S (f) = 0, or 係数 D (f) ² is minimized. You may decide.

In the third embodiment, the actuator 42
May be arranged outside the microphone assembly 25c. In the fourth embodiment, when the directivities of the microphones 19 and 20 are relatively changed, the microphones 19 and 20 may be vertically displaced, or may be combined with horizontal and vertical displacements. In the fourth embodiment, the directivity adjustment is automatically performed by the drive control unit 43. However, the adjustment may be performed manually. Further, in the fourth embodiment, instead of the subtracter 38 as the difference amount detecting means, the subtraction section (difference amount detecting means) 35 obtains the difference amount Dn (f) of the noise signal, and the drive control section 43
Is the difference Dn (f) obtained from the subtracter 35
A similar control may be performed based on

By combining the fourth embodiment and the fifth embodiment, a plurality of input microphones 60c are arranged and the difference Dn
The one that maximizes (f) may be used. In the sixth embodiment, the voice section determination device 48 may determine the voice section by calculating the difference of the amplitude spectrum instead of the power spectrum. In that case, the voice extraction unit 46
May directly use the result of the difference for audio data extraction. The voice section determination devices 30 and 48 are connected to the voice recognition device 9 and
You may comprise independently of the audio | voice extraction apparatus 18. Further, the voice extraction device 18 may be configured independently of the voice recognition device 9. Further, the voice section determination devices 30 and 48 are not limited to those used for the voice recognition device 9 and can be applied to devices that determine which part of the input signal sequence contains voice data. The dialogue control unit 1 as the voice recognition device 9
6, the voice synthesizer 17 and the speaker 22 are not always necessary. Without being limited to the car navigation device 1,
The present invention is applicable as long as it performs voice input.

[Brief description of the drawings]

FIG. 1 illustrates a first embodiment in which the present invention is applied to a car navigation device, and is a functional block diagram illustrating an electrical configuration of the entire car navigation device.

FIG. 2 is a plan view showing a setting state of directivity of a voice microphone and a noise microphone with respect to a speaker;

3A and 3B are views showing the appearance of a microphone assembly constituting an input microphone, wherein FIG. 3A is a front view, FIG.
(C) is a plan view showing a disassembled state.

FIG. 4 is a perspective view showing a state in which an input microphone is arranged on a steering column in a driver's seat.

FIG. 5 is a functional block diagram showing a detailed configuration of a voice extraction device.

FIG. 6 is a functional block diagram showing a detailed configuration of a voice section determination device.

FIG. 7 is a flowchart showing control contents of a control unit of the voice recognition device.

8A and 8B are waveform diagrams, respectively, wherein FIG. 8A shows ON / OFF states of a PTT switch, FIG. 8B shows an input signal of an audio microphone,
(C) shows an input signal of the noise microphone.

FIG. 9 is a view corresponding to FIG. 2, showing a second embodiment of the present invention;

FIG. 10 is a view corresponding to FIG. 2, showing a third embodiment of the present invention;

FIG. 11 is a view corresponding to FIG. 2, showing a fourth embodiment of the present invention;

FIG. 12 is a diagram corresponding to FIG. 6;

FIG. 13 is a view corresponding to FIG. 6, showing a fifth embodiment of the present invention.

FIG. 14 is a view corresponding to FIG. 5, showing a sixth embodiment of the present invention.

FIG. 15 is a diagram corresponding to FIG. 6;

FIG. 16 is a diagram for explaining a principle of determining a speech input section in the related art.

[Explanation of symbols]

1 is a car navigation device, 15 is a voice recognition unit (voice recognition means), 18 is a voice extraction device, 19 is a voice microphone, 20 is a noise microphone, 25, 25a, 25b and 2
5c is a microphone assembly (casing), 29 is a frame division unit (audio signal data generation means, noise signal data generation means), 30 is an audio section determination device, and 35 is a subtraction unit (audio data extraction means, difference amount detection means) , 36 and 3
7 is a power calculation unit (sound signal data generation means and noise signal data generation means), 38 is a subtractor (difference amount detection means), 39 and 39a are judgment units (speech input section judgment means), and 42 is an actuator (directivity). Change mechanism), 43
Is a drive control unit (directivity control unit), 45 is a selection control unit (voice input section determination unit), 46 is a voice extraction device, 47
Is an acoustic analysis unit (sound signal data generation means, noise signal data generation means), 48 is a voice section determination device, 49 and 50
Are power spectrum calculation units (audio signal data generation means and noise signal data generation means), 60, 60a, 60b
Reference numerals 60c and 60c denote input microphones.

Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat II (Reference) G10L 9/00 D (72) Inventor Hiroshi Ohno 1-1-1 Showa-cho, Kariya-shi, Aichi Pref. Inventor Norihide Kitaoka 1-1-1, Showa-cho, Kariya-shi, Aichi Prefecture, DENSO Corporation (72) Inventor Kazuhiro Higuchi 1-1-1, Showa-cho, Kariya-shi, Aichi Prefecture F-term in DENSO Corporation 2F029 AA02 AB01 AB07 AB09 AC01 AC02 AC04 AC18 5D015 CC01 DD02 DD03 EE05 KK01 9A001 HH16 HH17 JJ78

Claims (19)

[Claims] The directivity is set to the direction of a speaker,
A voice microphone for inputting the voice of the speaker; a directivity set to be different from the voice microphone; a noise microphone for inputting noise; and an input signal of the voice microphone. Audio signal data generating means for generating audio signal data based on the noise signal, noise signal data generating means for generating noise signal data based on the input signal of the noise microphone, and audio signal data generated by the input signal data generating means And a voice input section determining means for determining a voice input section in the voice signal data based on a difference between the voice signal section and the noise signal data generated by the noise data generating section. Judgment device. 2. The voice input section determination device according to claim 1, wherein the directivity of the noise microphone is set to be substantially 180 degrees different from the directivity of the voice microphone. 3. The voice input section determination device according to claim 1, wherein the directivity of the noise microphone is set to be different from the directivity of the voice microphone by approximately 90 degrees. 4. A directivity changing mechanism for relatively changing the directivity of the voice microphone and the noise microphone, and a difference amount based on noise signals input to the voice microphone and the noise microphone, respectively. Means for detecting a difference amount to be detected, and directivity control for controlling the directivity of both microphones to be relatively changed by the directivity changing mechanism so that the difference amount detected by the difference amount detection means is maximized. The voice input section determination device according to any one of claims 1 to 3, further comprising means. 5. A plurality of pairs of the voice microphone and the noise microphone, and a difference amount detecting means for determining a difference amount based on a noise signal input to each of the voice microphone and the noise microphone for each pair. The voice input section determination unit determines a voice input section based on an input signal input to a pair having a maximum difference amount detected by the difference amount detection unit. The voice input section determination device according to any one of claims 1 to 4. 6. The voice input section determination unit according to claim 1, wherein the voice input section determination unit determines the voice input section based on a difference in average power between the voice signal data and the noise signal data. 3. The voice input section determination device according to claim 1. 7. The voice input section determining unit determines a voice input section based on a difference in power spectrum between the voice signal data and the noise signal data. The voice input section determination device according to any one of the claims. 8. The directivity is set in the direction of the speaker,
A voice microphone for inputting the voice of the speaker; a directivity set to be different from the voice microphone; a noise microphone for inputting noise; and an input signal of the voice microphone. Audio signal data generation means for generating audio signal data based on the noise signal, noise signal data generation means for generating noise signal data based on the input signal of the noise microphone, and audio signal data generated by the audio signal data generation means And an audio data extraction unit for extracting audio data from the audio signal data based on a difference between the audio signal data and the noise signal data generated by the noise signal data generation unit. 9. The audio data extraction device according to claim 8, wherein the directivity of the noise microphone is set to be different from the directivity of the audio microphone by approximately 180 degrees. 10. The audio data extraction device according to claim 8, wherein the directivity of the noise microphone is set to be different from the directivity of the audio microphone by approximately 90 degrees. 11. A directivity changing mechanism for relatively changing the directivity of the audio microphone and the noise microphone, and a difference amount based on a noise signal input to each of the audio microphone and the noise microphone. Means for detecting a difference amount to be detected; and directivity control for controlling the directivity of both microphones to be relatively changed by the directivity changing mechanism so that the difference amount detected by the difference amount detection means is maximized. The audio data extraction device according to claim 8, further comprising: 12. A difference amount detecting means for detecting a difference amount based on a noise signal inputted to each of the sound microphone and the noise microphone for each pair, wherein a plurality of pairs of the sound microphone and the noise microphone are provided. The audio data extracting means extracts audio data based on an input signal input to a pair having a maximum difference amount detected by the difference amount detecting means. The audio data extraction device according to any one of the above. 13. The audio data extracting means according to claim 8, wherein the audio data extracting means extracts the audio data based on a difference in power spectrum between the audio signal data and the noise signal data. Audio data extraction device. 14. A voice input section determining apparatus according to claim 1, wherein said voice data extracting means includes a voice included in a voice input section determined by said voice input section determining means. 14. The audio data extraction device according to claim 8, wherein data is extracted. 15. The voice microphone, the noise microphone, the noise data generating means, and the input signal data generating means are commonly configured with each element of the voice input section determination device. The audio data extraction device according to the above. 16. The voice input section determination device according to claim 1, further comprising: voice data included in the voice input section determined by the voice input section determination unit of the voice input section determination device. A speech recognition device, comprising: speech recognition means for analyzing and recognizing a speech data pattern. 17. A voice recognition apparatus for recognizing a voice data pattern by analyzing the voice data extracted by the voice data extraction device according to claim 8 and voice data extraction means of the voice data extraction device. And a voice recognition device. 18. A vehicular navigation system comprising the voice recognition device according to claim 16 or 17, wherein a predetermined process is executed based on a voice data pattern recognized by voice recognition means of the voice recognition device. apparatus. 19. The sound microphone and the noise microphone are arranged in the same housing to be integrally formed, and the shape of the housing is such that the sound microphone and the noise microphone are apparently in the same direction. The voice input section determination device according to any one of claims 1 to 7, wherein the voice data extraction device according to any one of claims 8 to 15, or the voice data extraction device according to any one of claims 8 to 15. Item 16. An input microphone used in any of the speech recognition devices according to Item 16 or 17.