JP2013097273A - Sound source estimation device, method, and program and moving body - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a sound source estimation device, method, and program capable of estimating a sound source more accurately and to provide a moving body using the same.SOLUTION: The sound source estimation device uses observation signals acquired by microphones 11 and 12 to estimate a direction of a sound source and includes: a noise estimator 23 which estimates noise components included in the observation signals; a mask generation unit 24 which generates a mask M(ω,t) on the basis of the noise components estimated by the noise estimator 23; a reliability generation unit 25 which calculates reliability (t) of the mask generated by the mask generation unit 24; a CSP coefficient calculation unit 26 which calculates a CSP coefficient corrected by the mask and the reliability of the mask; and a direction estimation unit 28 which estimates the direction of the sound source on the basis of the CSP coefficient.

Description

  The present invention relates to a sound source estimation device, a sound source estimation method, a sound source estimation program, and a moving object, and particularly uses a sound source estimation device, a sound source estimation method, a sound source estimation program, and a sound source estimation device for estimating a sound source. Related to moving objects.

  Non-Patent Document 1 discloses a technique using a CSP method (Cross-power Spectrum Phase analysis). The CSP method is also called a GCC-PHAT (Generalized Cross-Correlation PHAse Transform) algorithm, and is used for estimating a sound source direction.

Frithof Hummes, Junge Qi, by Tim Fingcheid ROBUST ACOUSTIC SPEAKER LOCALIZATION WITCH DISTRIBUTED MICROPHONES 19th European Cep 20 US

Hereinafter, processing of the CSP method will be described. FIG. 4 is a block diagram showing a processing flow of the CSP method. The short-time DFT unit 121 performs short-time DFT (Discrete Fourier Transform) processing on the observation signals x ₁ (t) and x ₂ (t) observed by two microphones (hereinafter referred to as microphones). Thereby, the observation signals x ₁ (t) and x ₂ (t) in the time domain are converted into the observation signals X ₁ (ω, t) and X ₂ (ω, t) in the time-frequency domain, respectively.

The CPS coefficient calculation unit 126 calculates the CSP coefficient CSP (d, t) from the observation signals X ₁ (ω, t) and X ₂ (ω, t). The CSP coefficient is a cross-correlation function obtained by normalizing the observation signals X ₁ (ω, t) and X ₂ (ω, t) with their amplitudes. Then, the time difference estimation unit 127 estimates the arrival time difference τ based on the index d that maximizes the CSP coefficient. This arrival time difference τ corresponds to the arrival time difference of sound observed by the first microphone and the second microphone. Based on the estimated time difference, the direction estimation unit 128 estimates the direction θ.

Incidentally, the CSP algorithm has an assumption that the number of sound sources is one. In the CSP method, sound source azimuth estimation is performed using a full-band signal obtained by discrete Fourier transform based on this assumption. First, the observation signals X ₁ (ω, t) and X ₂ (ω, t) can be expressed by the following equations (1) and (2).

ω is frequency and t is time. X _1s is an observation signal when the sound from the target sound source is acquired by the first microphone, and X _2s is an observation signal when the sound from the target sound source is acquired by the second microphone. t _1s is a time corresponding to the distance between the target sound source and the first microphone, and t _2s is a time corresponding to the distance between the target sound source and the second microphone. τ (t) is the difference in arrival time of sound between the first microphone and the second microphone. Assuming that the number of sound sources is one, the CSP coefficient can be expressed by equation (3).

In the formula (3), * indicates a conjugate. However, in an actual environment, the number of sound sources is not necessarily one, and there is a mixture of environmental noise and interference sound. For this reason, a mixed signal of a plurality of sound sources is observed. This mixed signal can be expressed by equations (4) and (5).

In Expressions (4) and (5), n is the number of sound sources (noise sources) that become noise. X _1Nn is an observation signal when sound from the nth noise source is observed by the first microphone, and t _1Nn is a time corresponding to the distance between the nth noise source and the first microphone. is there. Similarly, X _2Nn is an observation signal when sound from the nth noise source is observed by the second microphone, and t _2Nn is a time corresponding to the distance between the nth noise source and the second microphone. It corresponds to. τ _s is the arrival time difference of the sound from the target sound source, and τ _Nn is the arrival time difference of the sound from the nth noise source.

Hereinafter, in order to simplify the description, the number of target sound sources is 1 and the number of noise sources is 1. In this case, the observation signal is expressed by the following equations (6) and (7).

The CSP coefficient is developed by the following equation (8).

In the case of a high SNR (Signal Noise Ratio), that is, in a low noise environment, the following approximate expression (9) is established.

Accordingly, the calculation formula for the CSP coefficient is expanded as shown in the following formula (10), and thus it is possible to estimate the direction of the target sound source.

In the case of low SNR, that is, in a high noise environment, the approximate expression of Expression (9) is not established. Therefore, as shown in Expression (11), the arrival phase difference of the noise component or the like (region term in Expression (11)) affects the CSP coefficient sequence.

As shown in Expression (12) and Expression (13), the index d that maximizes the CSP coefficient is searched, and the direction of the sound source is calculated by converting the index d.

  As described above, in the CSP method, amplitude information is normalized, and calculation is performed using only phase difference information. Furthermore, in the method described in Non-Patent Document 1, a Wiener filter is used. In such estimation of the sound source direction, higher accuracy is desired. For example, when a noise source and a target sound source are moving relative to a microphone, it is desired to estimate the direction more accurately.

  The present invention has been made in view of the above problems, and an object thereof is to provide a sound source estimation apparatus, a sound source estimation method, and a sound source estimation program that can accurately estimate a sound source.

  A sound source estimation apparatus according to an aspect of the present invention is a sound source estimation apparatus that estimates a direction of a sound source using observation signals acquired by at least two microphones, and estimates a noise component included in the observation signal. A noise estimation unit; a mask generation unit that generates a mask based on a noise component estimated by the noise estimation unit; a reliability calculation unit that calculates a reliability of the mask generated by the mask generation unit; A mask, a CSP coefficient calculation unit that calculates a CSP coefficient corrected by the reliability of the mask, and an estimation unit that estimates a direction of a sound source based on the corrected CSP coefficient. Thus, by introducing the mask and the reliability of the mask, the sound source can be accurately estimated.

  In the sound source estimation device, based on the index that maximizes the corrected CSP coefficient, an arrival time difference between the two microphones is calculated, and based on the arrival time difference, the estimation unit determines the sound source A direction may be estimated, and whether or not a target sound source exists in the estimated direction may be determined according to the corrected maximum value of the CSP coefficient. By doing so, the sound source can be estimated more accurately.

  In the sound source estimation apparatus, the mask is a discrete value corresponding to a frequency, and is corrected by performing an inverse Fourier transform on a product of the cross-correlation of observation signals from the two microphones and the mask. The CSP coefficient may be calculated.

  In the sound source estimation apparatus, the corrected CSP coefficient may be calculated by multiplying the inverse Fourier transform value by the reliability.

  A moving object according to an aspect of the present invention is a moving object equipped with the above-described sound source estimation device, and the noise estimation unit performs noise estimation based on a vehicle signal according to an operation state of the moving object. It is characterized by. By doing in this way, a sound source can be estimated using a noise component estimated at an appropriate timing.

  In the above moving body, a mask storage unit in which the mobile body stores a mask in advance, a mask stored in the mask storage unit based on a vehicle signal corresponding to an operation state of the moving body, and the mask generation unit And a mask selection unit that selects any of the masks generated in (1). By doing in this way, a sound source can be estimated using an appropriate mask.

  A sound source estimation method according to an aspect of the present invention is a sound source estimation method that estimates a direction of a sound source using observation signals acquired by at least two microphones, and estimates a noise component included in the observation signal. Generating a mask based on the noise component; calculating a reliability of the mask; calculating a CSP coefficient corrected by the mask and the reliability of the mask; And estimating the direction of the sound source based on the corrected CSP coefficient. Thus, by introducing the mask and the reliability of the mask, the sound source can be accurately estimated.

  In the sound source estimation method, based on the index that maximizes the corrected CSP coefficient, a difference in arrival time when sound arrives at the two microphones is calculated. Based on the difference in arrival time, the estimation unit determines the sound source A direction may be estimated, and whether or not a target sound source exists in the estimated direction may be determined according to the corrected maximum value of the CSP coefficient. By doing so, the sound source can be estimated more accurately.

  In the sound source estimation method, the mask is a discrete value corresponding to a frequency, and is corrected by performing an inverse Fourier transform on a product of a cross-correlation of observation signals from the two microphones and the mask. The CSP coefficient may be calculated.

  In the above sound source estimation method, the corrected CSP coefficient may be calculated by multiplying the inverse Fourier transform value by the reliability.

  The sound source estimation program is a sound source estimation program that estimates the direction of a sound source using observation signals acquired by at least two microphones, and estimates a noise component included in the observation signal to a computer Generating a mask based on the noise component; calculating a reliability of the mask; calculating a CSP coefficient corrected by the mask and the reliability of the mask; And a step of estimating a direction of a sound source based on the corrected CSP coefficient. Thus, by introducing the mask and the reliability of the mask, the sound source can be accurately estimated.

  The sound source estimation program causes a computer to calculate an arrival time difference between two microphones based on an index that maximizes the corrected CSP coefficient, and based on the arrival time difference, The estimation unit may cause the direction of the sound source to be estimated and determine whether or not the target sound source exists in the estimated direction according to the corrected maximum value of the CSP coefficient. By doing so, the sound source can be estimated more accurately.

In the sound source estimation program, the mask is a discrete value corresponding to a frequency,
The corrected CSP coefficient may be calculated by performing inverse Fourier transform on the product of the cross-correlation of the observation signals from the two microphones and the mask.

  The sound source estimation program may calculate the corrected CSP coefficient by multiplying the inverse Fourier transform value by the reliability.

  ADVANTAGE OF THE INVENTION According to this invention, the sound source estimation apparatus which can estimate a sound source correctly, a sound source estimation method, a sound source estimation program, and a mobile body using the same can be provided.

It is a block diagram which shows the structure of the sound source estimation apparatus concerning embodiment. It is a block diagram which shows the flow in a sound source estimation apparatus. It is a block diagram which shows the application example of the sound source estimation apparatus concerning embodiment. It is a figure explaining the sound source estimation by CSP method.

  Hereinafter, embodiments of a sound source estimation apparatus according to the present invention will be described in detail with reference to the drawings. However, the present invention is not limited to the following embodiments. In addition, for clarity of explanation, the following description and drawings are simplified as appropriate.

  First, a sound source estimation apparatus according to an embodiment of the present invention will be described with reference to FIG. FIG. 1 is a block diagram showing a system configuration of the sound source estimation apparatus. The sound source estimation apparatus according to the present embodiment estimates the direction of the sound source. Further, it is determined whether or not a target sound source exists in the estimated direction of the sound source. For example, the sound source estimation apparatus according to the present embodiment is mounted on a vehicle. And the direction of the other vehicle which is a sound source, and whether other vehicles exist near are detected. By doing in this way, the presence or absence of an approaching vehicle and its direction can be detected. Thereby, it can notify effectively that the vehicle is approaching, and it can contribute to prevention of a traffic accident.

  As shown in FIG. 1, the sound source estimation apparatus includes a microphone 11, a microphone 12, a microphone amplifier 13, a microphone amplifier 14, an AD converter 15, and a CPU 16. In FIG. 1, only two microphones 11 and 12 are shown, but the number of microphones is not particularly limited. The number of microphones may be plural, for example, three or more. For example, a microphone array in which a plurality of microphones are arranged in an array can be used. Then, the following processing is performed on two of the many microphones. By doing so, the direction of the sound source can be estimated. Furthermore, the position of the sound source can be specified by preparing a plurality of pairs of microphones and performing the following processing on each of them.

  The microphone 11 and the microphone 12 are spaced apart by a distance D. Assume that the microphones 11 and 12 detect sound from the direction of θ (t). That is, in FIG. 1, it is assumed that the target sound source is in the direction of θ (t). The microphones 11 and 12 output an observation signal corresponding to the detected sound.

  The microphone amplifiers 13 and 14 amplify the observation signals from the microphones 11 and 12, respectively, and output the amplified signals to the A / D converter 15. The AD converter 15 performs A / D conversion on the input observation signal. The digital observation signal output from the A / D converter 15 is input to a CPU (Central Processing Unit). The CPU 16 performs arithmetic processing for estimating the sound source direction on the observation signal from the A / D converter 15. The CPU 16 performs processing with reference to programs and parameters stored in a ROM (Read Only Memory) and a RAM (Random Access Memory) (not shown).

  Next, the configuration of processing blocks in the CPU 16 will be described with reference to FIG. FIG. 2 is a block diagram showing the configuration of the CPU 12. The CPU 16 performs processing according to the block on the observation signal from the A / D converter 15. The CPU 16 includes a short-time DFT unit 21, a short-time DFT unit 22, a noise estimator 23, a mask generation unit 24, a reliability generation unit 25, a time-frequency correction type CSP coefficient calculation unit 26, and a time difference estimation unit. 27 and an azimuth estimation unit 28.

An observation signal observed by the microphone 11 is an observation signal x ₁ (t), and a signal observed by the microphone 12 is an observation signal x ₂ (t). The short-time DFT units 21 and 22 perform short-time discrete Fourier transform on the observation signals x ₁ and x ₂ (t). For example, an observation signal for a predetermined time is stored in a buffer or a memory, and the observation signal is divided into a plurality of frames. For example, in the time domain, the frame is divided by half shift so that adjacent frames are overlapped by half. Further, the frame may be divided using a window function. Further, the observation signal divided into frames is subjected to discrete Fourier transform. In this way, the time domain observation signals x ₁ (t) and x ₂ (t) are converted into time-frequency domain observation signals X ₁ (ω, t) and X ₂ (ω, t), respectively. The The short-time DFT units 21 and 22 output the observation signals X ₁ (ω, t) and X ₂ (ω, t) to the noise estimator 23, the mask generation unit 24, and the CSP coefficient calculation unit 26.

The noise estimator 23 estimates noise using the observation signals X ₁ (ω, t) and X ₂ (ω, t). For example, a time average in the past time or an estimation method using a microphone array such as Nullbeamformer can be used. Specifically, noise estimation can be performed using the following equation (14). In Equation (14), S is the number of frame divisions.

The noise estimator 23 outputs the estimated noise N (ω, t) to the mask generation unit 24. The mask generation unit 24 generates a mask M (ω, t) that masks the CSP coefficient according to the frequency. The mask generator 24 calculates the mask M (ω, t) using the noise N (ω, t) and the observation signals X ₁ (ω, t) and X ₂ (ω, t). For example, the establishment / non-establishment of the approximate expression shown in Expression (9) is determined by the SN ratio (SNR) at each frequency of the incoming sound. For this reason, the SN ratio is estimated and it is determined whether or not the approximate expression holds. For a frequency band in which the approximate expression is not satisfied, that is, a band with high noise, a process for performing a masking process is introduced. By doing so, it is possible to calculate the CSP coefficient only in the band where the influence of the noise component is small. This makes it possible to estimate the direction of a sound source that operates robustly even in a low SNR environment (under a high noise environment).

  For example, the noise N (ω, t) may be compared with a threshold value and M (ω, t) may be set according to the comparison result. Specifically, when the value of the noise N (ω, t) is larger than the threshold value, M (ω, t) = 0, and when smaller than the threshold value, M (ω, t) = 0. To do. Thus, the mask M (ω, t) has a discrete value corresponding to the frequency. The mask M (ω, t) generated by the mask generation unit 24 is input to the reliability generation unit 25 and the time-frequency correction type CSP coefficient calculation unit 26.

  The reliability generation unit 25 calculates Reliability (t) indicating the reliability of the mask M (ω, t). As described above, the value of M (ω, t) changes according to the noise N (ω, t). Therefore, it is considered that the more the frequency at which M (ω, t) = 1, the less the noise and the higher the reliability. On the other hand, it is considered that the greater the frequency at which M (ω, t) = 0, the greater the noise and the lower the reliability. In such a case, since there are few signal components from the target sound source included in the observation signal, the reliability of the estimated direction of the target sound source is lowered. Therefore, the direction of the sound source can be estimated more accurately by introducing Reliability (t) indicating the reliability of the mask M (ω, t). That is, based on the noise component and the signal component, the CSP coefficient subjected to time-frequency correction can be calculated by using the mask M (ω, t) and Reliability (t).

For example, Reliability (t) can be obtained using the following equation (15).

  Note that Ω is the count number of ω. That is, it is assumed that M (ω, t) for Ω ω has been calculated. For example, when Ω = 100, that is, when 100 M (ω, t) are calculated at a certain time, 10 M (ω, t) out of 100 are 1, and 90 M ( Let ω, t) be zero. At this time, Reliability (t) is 0.1 (= 10/100). In this case, the reliability is low. On the other hand, in the case of Ω = 100, 100 out of 100 M (ω, t) are 100, and 0 M (ω, t) is 0. At this time, Reliability (t) is 1 (= 100/100). In this case, the reliability is high.

The reliability generation unit 25 outputs Reliability (t) to the time-frequency correction type CSP coefficient calculation unit 26. Further, the observation signal X ₁ (ω, t) and the observation signal X ₂ (ω, t) from the short-time DFT units 21 and 22 are input to the CSP coefficient calculation unit 26.

The CSP coefficient calculation unit 26 calculates the CSP coefficient CSP (ω, t) based on Reliability (t), the observation signal X ₁ (ω, t), and the observation signal X ₂ (ω, t). CSP (ω, t) can be obtained using, for example, Expression (16).

As shown in Expression (16), the CSP coefficient calculation unit 26 normalizes the observed signals X ₁ (ω, t) and X ₂ (ω, t) with their amplitudes and a mask M (ω, An inverse discrete Fourier transform (IDFT) is performed on the product with t). Then, the CSP coefficient calculating unit 26 obtains the CSP coefficient by multiplying the value obtained by the inverse discrete Fourier transform by Reliability (t). In other words, Reliability (t) is the weighting value of the CSP coefficient. In this way, the CSP coefficient CSP that has been corrected for time and frequency can be obtained.

The CSP coefficient calculation unit 26 outputs the calculated CSP coefficient to the time difference estimation unit 27. The time difference estimation unit 27 estimates the arrival time difference τ (t) from the CSP coefficient. Thereby, the time difference between the sounds arriving at the two microphones 11 and 12 can be obtained. For example, the arrival time difference τ (t) can be calculated using Expression (17).

The sampling frequency is a sampling frequency. In Expression (17), an index d that maximizes the CSP coefficient CSP (d, t) is calculated. Then, the arrival time difference τ (t) is calculated by dividing the index d by the sampling frequency. Thus, the arrival time difference τ (t) is calculated based on the cross correlation function of the observed signals X ₁ (ω, t) and X ₂ (ω, t) normalized by the CSP coefficient, that is, the amplitude. . In the CSP method, amplitude information is normalized and CSP coefficients are calculated based on phase difference spectrum information. Therefore, the CSP method is more robust against the influence of reverberation than other sound source direction estimation techniques.

The azimuth estimating unit 28 estimates the azimuth θ (t) at which sound has arrived with respect to the microphones 11 and 12 based on the arrival time difference τ (t). Thereby, the direction of the sound source can be estimated. For example, the azimuth θ (t) can be estimated using Equation (18). C is the speed of sound.

  The determination unit 29 determines whether a target sound source is present in the azimuth θ (t) according to the value of the CSP coefficient CSP. For example, assume that the target sound source is another vehicle. In this case, it is determined whether there is another vehicle in the direction θ (t) according to the value of the CSP coefficient CSP at the index d at which the CSP coefficient is maximum. When the maximum value of the CSP coefficient is larger than the threshold value, the noise component is low and the reliability is high. Therefore, it is determined that another vehicle exists in the direction of θ (t). On the other hand, when the maximum value of the CSP coefficient is smaller than the threshold value, the noise component is high and the reliability is low. Therefore, it is determined that there is no other vehicle in the direction of θ (t).

  Thus, by comparing the CSP coefficient CSP with the threshold value, it can be estimated whether there is a sound source in the azimuth θ (t). The threshold value to be compared with the CSP coefficient CSP may be set in advance by the user according to the experimental result or the like. Whether or not there is a target sound source in the azimuth θ (t) is detected according to the maximum value of the CSP coefficient CSP. Therefore, reliability can be improved.

  Such a determination method based on the CSP coefficient is, for example, “Construction of an approaching vehicle detection system using a microphone array corresponding to a plurality of vehicles” Hideki Sakano et al., IEICE Technical Report; Volume: 2011-3-18, 110, 471; pp13-16 can be used.

  By using the above sound source estimation method, the direction of the target sound source can be estimated more accurately. By introducing the mask M (ω, t), it is possible to reduce the influence of the frequency having a high noise component. Furthermore, by introducing Reliability (t) indicating the reliability of the mask M (ω, t), it is possible to prevent the direction from being estimated at a timing with low reliability. That is, it is possible to estimate the signal component at a high timing. Thereby, the direction of the target sound source can be estimated more accurately.

  In the above description, the mask M (ω, t) is set to binary, that is, binary of (0,1), but the mask M (ω, t) is limited to the binary of (0,1). is not. That is, the value of the mask M (ω, t) may be set stepwise or continuously. For example, the noise N (ω, t) may be compared with a plurality of threshold values, and the mask M (ω, t) may be calculated in multiple stages between 0 and 1. Further, the mask M (ω, t) may be calculated as a continuous value between 0 and 1. Specifically, the mask M (ω, t) can be calculated using the Wiener filter expressed by the following formula (19) or formula (20).

  Note that γ is a parameter that can be set in advance according to an experimental result or the like, and can be a real number other than 2 or 2. In this way, a mask can be generated using a pseudo parametric Wiener filter. Similarly, β is a parameter that can be set in advance according to the experimental result or the like, and can be a real number other than 1 or 1. In this way, a mask M (ω, t) that can eliminate or suppress the influence of a high frequency noise component can be introduced. Even when the value of the mask M (ω, t) is set as a continuous value, Reliability (t) can be calculated using the above equation (15).

  The above-described sound source estimation apparatus is suitable for mounting on a moving object. In a moving body such as an automobile, a mobile robot, and a motorcycle, the direction of the sound source is estimated while moving by itself. Furthermore, in an environment such as a public road where other moving bodies are moving, the other moving bodies that are sound sources also move. In such a case, the sound source estimation apparatus estimates the sound source direction while the moving body moves relative to the target sound source. The sound source estimation process described above is performed in an environment where the target sound source and the sound source estimation apparatus are relatively moving. In the sound source estimation process, since the CSP coefficient subjected to time correction is used, the direction can be estimated more accurately. That is, since the reliability (t) is introduced and the sound source direction is estimated from the observation signal at a highly reliable timing, the estimation accuracy can be improved.

  Hereinafter, an example in which the sound source estimation apparatus is mounted on a vehicle that is a moving body will be described with reference to FIG. FIG. 3 is a block diagram showing a main part of a vehicle equipped with a sound source estimation device. The vehicle 30 includes a vehicle signal acquisition unit 31 and a noise estimator activation unit 32. Further, a mask storage unit 41 and a mask selection unit 42 are added to the sound source estimation apparatus shown in FIG. The short-time DFT unit 21, the short-time DFT unit 22, the time difference estimation unit 27, the direction estimation unit 28, and the determination unit 29 shown in FIG. doing. In the configuration shown in FIG. 3, the mask M (ω, t) is dynamically generated as described below.

  The vehicle signal acquisition unit 31 acquires a vehicle signal related to the vehicle 30. For example, the vehicle signal acquisition unit 31 acquires a control signal or an operation signal of the vehicle 30 as a vehicle signal. Specifically, if the vehicle 30 is an automobile, the on / off state of a wiper or a headlight provided on the vehicle 30 is acquired as a vehicle signal. Furthermore, the traveling speed of the vehicle 30, the depression amount of the brake pedal or the accelerator pedal, map information, or position information from GPS may be used as the vehicle signal. In addition, a recognition result from another sensor from a camera or radar may be used as a vehicle signal. The vehicle signal may be information regarding the operating state of the vehicle 30. The vehicle signal acquisition unit 31 outputs the acquired vehicle signal to the noise estimator activation unit 32 and the mask selection unit 42.

  The noise estimator activation unit 32 activates the noise estimator 23 based on the vehicle signal corresponding to the operation state of the vehicle 30. The noise estimator 23 starts noise estimation in response to an instruction from the noise estimator activation unit 32. When the noise in the environment changes, the noise estimator activation unit 32 activates the noise estimator 23. For example, the noise estimator activation unit 32 may activate the noise estimator 23 at a timing when the vehicle speed becomes a certain speed or less (for example, 20 km / h or less). Thereby, noise estimation is performed at the timing when the vehicle speed becomes a certain speed or less. Alternatively, the noise estimator activation unit 32 may activate the noise estimator 23 based on the depression amount of the brake pedal or the accelerator pedal. Furthermore, the noise estimator activation unit 32 may activate the noise estimator 23 based on the map information and the position information from the GPS. Specifically, when the vehicle 30 approaches a point such as an intersection where traffic accidents frequently occur, a mask may be generated by noise estimation immediately before that. Furthermore, the noise estimator activation unit 32 may activate the noise estimator 23 based on the recognition results of other sensors such as a camera and a laser. In this way, the noise estimator activation unit 32 activates the noise estimator 23 so that noise estimation is performed at the timing when the environment around the vehicle 30 changes or when the operation of the vehicle 30 changes.

  The mask storage unit 41 stores one or more preset masks M (ω, t). For example, a mask is obtained by an experiment or the like at the time of product development, and stored in advance in the mask storage unit 41 at the time of product manufacture. Further, the mask storage unit 41 stores the mask M (ω, t) generated by the mask generation unit 24. Specifically, a noise component in a state where the wiper is operating is collected in advance, and a mask is generated in advance based on the sound collection result. Alternatively, the engine sound of a vehicle traveling at a certain speed is collected, and a mask is generated in advance based on the sound collection result. Such a mask is stored in the mask storage unit 41 in advance.

The mask selection unit 42 selects the following (a) to (c) according to the situation.
(A) The mask generated on the spot (b) The mask (c) mask previously stored at the time of product manufacture is not used in the mask storage unit 41 (that is, all elements of M (ω, t) are always 1) mask)

As described above, the mask of (a) is generated using the observation signal X ₁ (ω, t), the observation signal X ₂ acquired on the spot, and the noise N (ω, t) estimated from them. The The mask (a) is a mask according to the current environment and the operating state of the vehicle 30. On the other hand, the mask storage unit 41 stores in advance a mask that does not depend on the in-situ observation signal.

  The mask selection unit 42 selects any one of the masks (a) to (c) described above based on the vehicle signal. For example, the mask for rainy weather and the mask for sunny weather are switched depending on whether the wiper switch is on or off. Specifically, the mask in the rainy weather can be the mask (b), and the mask in the fine weather can be the mask (a). Furthermore, the mask for the night and the mask for the day are switched depending on whether the headlight is on or off. You may make it switch the mask according to the characteristic of places, such as a city area and a suburb, from map information and the positional information from GPS. As described above, the mask selection unit 42 selects an optimal mask according to the operation state of the vehicle 30.

  As described above, the noise estimator 23 is activated based on the vehicle information indicating the state of the vehicle 30. Accordingly, a noise model, that is, a mask M (ω, t) can be dynamically generated in accordance with a change in the situation of the vehicle 30. Even when the state of noise around the vehicle 30 changes every moment, the mask M (ω, t) can be generated at an appropriate timing. Thereby, the direction of the sound source can be accurately estimated. Furthermore, the mask memorize | stored in the mask memory | storage part 41 and the mask produced | generated on the spot are used properly according to a vehicle signal. As a result, the sound source can be estimated more accurately. By mounting the sound source estimation device on the vehicle 30, it is possible to recognize an approaching vehicle from a side road that becomes a blind spot at an intersection or the like.

  In the above description, the example in which the sound source estimation device is mounted on the vehicle 30 that is an automobile has been described. However, the moving body on which the sound source estimation device is mounted is not particularly limited. For example, a sound source estimation device may be mounted on a motorcycle, a mobile robot, or the like. By mounting a sound source estimation device on a mobile robot, it becomes possible to look back in the direction of the user's voice or detect abnormal sounds.

  The sound source estimation process described above may be realized by causing a computer including a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), a CPU (Central Processing Unit), or a combination thereof to execute a program.

  In the above example, a program including a group of instructions for causing a computer to perform sound source estimation processing is stored using various types of non-transitory computer readable media and supplied to the computer. can do. Non-transitory computer readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (for example, flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (for example, magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / W, semiconductor memory (for example, mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (Random Access Memory)) are included. The program may also be supplied to the computer by various types of transitory computer readable media. Examples of transitory computer readable media include electrical signals, optical signals, and electromagnetic waves. The temporary computer-readable medium can supply the program to the computer via a wired communication path such as an electric wire and an optical fiber, or a wireless communication path.

11 microphone 12 microphone 13 microphone amplifier 14 microphone amplifier 15 A / D converter 16 CPU
DESCRIPTION OF SYMBOLS 21 Short time DFT part 22 Short time DFT part 23 Noise estimator 24 Mask generation part 25 Reliability generation part 26 CSP coefficient calculation part 27 Time difference estimation part 28 Direction estimation part 29 Judgment part 30 Vehicle 31 Vehicle signal acquisition part 32 Noise estimator starting Unit 41 Mask storage unit 42 Mask selection unit

Claims (14)

A sound source estimation apparatus that estimates the direction of a sound source using observation signals acquired by at least two microphones,
A noise estimation unit for estimating a noise component included in the observation signal;
A mask generation unit that generates a mask based on the noise component estimated by the noise estimation unit;
A reliability calculation unit for calculating the reliability of the mask generated by the mask generation unit;
A CSP coefficient calculation unit that calculates a CSP coefficient corrected by the mask and the reliability of the mask;
A sound source estimation apparatus comprising: an estimation unit that estimates a direction of a sound source based on the corrected CSP coefficient. Based on an index that maximizes the corrected CSP coefficient, a difference in arrival time at which sound arrives at the two microphones is calculated,
Based on the arrival time difference, the estimation unit estimates the direction of the sound source,
The sound source estimation apparatus according to claim 1, wherein it is determined whether or not a target sound source exists in the estimated direction according to the corrected maximum value of the CSP coefficient. The mask is a discrete value according to frequency,
The sound source according to claim 1 or 2, wherein the corrected CSP coefficient is calculated by performing inverse Fourier transform on a product of a cross-correlation of observation signals from two microphones and the mask. Estimating device.   The sound source estimation apparatus according to claim 3, wherein the corrected CSP coefficient is calculated by multiplying the inverse Fourier transform value by the reliability. It is a moving body carrying a sound source estimating device in any 1 paragraph of Claims 1-4,
The moving body, wherein the noise estimation unit performs noise estimation based on a vehicle signal corresponding to an operating state of the moving body. A mask storage unit in which the moving body stores a mask in advance;
The apparatus further comprises: a mask stored in the mask storage unit and a mask selection unit that selects one of the masks generated by the mask generation unit based on a vehicle signal corresponding to an operation state of the moving body. Item 6. A moving object according to item 5 A sound source estimation method for estimating the direction of a sound source using observation signals acquired by at least two microphones,
Estimating a noise component included in the observed signal;
Generating a mask based on the noise component;
Calculating a reliability of the mask;
Calculating a CSP coefficient corrected by the mask and the reliability of the mask;
A sound source estimation method comprising: estimating a sound source direction based on the corrected CSP coefficient. Based on an index that maximizes the corrected CSP coefficient, a difference in arrival time at which sound arrives at the two microphones is calculated,
Based on the arrival time difference, the estimation unit estimates the direction of the sound source,
The sound source estimation method according to claim 7, wherein whether or not a target sound source exists in the estimated direction is determined according to the corrected maximum value of the CSP coefficient. The mask is a discrete value according to frequency,
9. The sound source according to claim 7, wherein the corrected CSP coefficient is calculated by performing inverse Fourier transform on a product of a cross-correlation of observation signals from two microphones and the mask. Estimation method.   The sound source estimation method according to claim 9, wherein the corrected CSP coefficient is calculated by multiplying the inverse Fourier transform value by the reliability. A sound source estimation program for estimating the direction of a sound source using observation signals acquired by at least two microphones,
Against the computer,
Estimating a noise component included in the observed signal;
Generating a mask based on the noise component;
Calculating a reliability of the mask;
Calculating a CSP coefficient corrected by the mask and the reliability of the mask;
Estimating the direction of the sound source based on the corrected CSP coefficient;
Sound source estimation program that executes Against the computer,
Based on the index that maximizes the corrected CSP coefficient, the arrival time difference at which sound arrives at the two microphones is calculated,
Based on the arrival time difference, the estimation unit estimates the direction of the sound source,
The sound source estimation program according to claim 11, wherein it is determined whether or not a target sound source exists in the estimated direction according to the corrected maximum value of the CSP coefficient. The mask is a discrete value according to frequency,
The sound source according to claim 11 or 12, wherein the corrected CSP coefficient is calculated by performing inverse Fourier transform on a product of a cross-correlation of observation signals from two microphones and the mask. Estimation program.   The sound source estimation program according to claim 13, wherein the corrected CSP coefficient is calculated by multiplying the inverse Fourier transform value by the reliability.

Images