close

Вход

Забыли?

вход по аккаунту

?

DESCRIPTION JP2007240209

код для вставкиСкачать
Patent Translate
Powered by EPO and Google
Notice
This translation is machine-generated. It cannot be guaranteed that it is intelligible, accurate,
complete, reliable or fit for specific purposes. Critical decisions, such as commercially relevant or
financial decisions, should not be based on machine-translation output.
DESCRIPTION JP2007240209
An object of the present invention is to obtain a good degree of angular resolution when
estimating the direction of arrival of a signal in three dimensions, which contributes to the
improvement of signal resolution in blind signal separation. A sensor adaptive selection unit 8
adaptively selects three sensors with respect to the signal arrival direction calculated by a first
stage arrival direction estimation unit 7, and the selected three sensors separate by frequency.
And a second-stage arrival direction estimation unit 10 for calculating the signal arrival direction
from the recalculated signal arrival direction for each frequency. [Selected figure] Figure 1
Signal arrival direction estimation apparatus and method, signal separation apparatus and
method, computer program
[0001]
The present invention relates to a signal arrival direction estimation apparatus and method, a
signal separation apparatus and method, and a computer program.
[0002]
Conventionally, a technique of blind signal separation is known, in which Independent
Component Analysis (ICA) is applied in the frequency domain to separate signals.
Then, in order to solve the problem of "permutation: Permutation" when the signal is separated,
the arrival direction of the three-dimensional signal is estimated (see, for example, Non-Patent
10-04-2019
1
Document 1). Tsujihiro, Yamada Makoto, Kawai Tsune, "Direction Estimation of 3D Sound Source
Using ICA", Proceedings of the 2005 Acoustical Society of Japan Annual Conference, 2-2-16, pp.
617-618
[0003]
However, in the prior art described above, when estimating the direction of arrival of a signal in
three dimensions, it may happen that the fixedly selected sensor can not obtain an accurate
direction of arrival over the entire signal spectrum. . In particular, reducing the distance between
sensors to avoid spatial aliasing reduces the separation of direction estimates at low frequencies.
For this reason, it becomes difficult to solve the problem of "Permutation" at low frequencies, and
when using the correlation of inter-frequency signals, there arises a problem that the amount of
calculation in blind signal separation becomes large.
[0004]
The present invention has been made in consideration of such circumstances, and its object is to
obtain good angular resolution when estimating the direction of arrival of a signal in three
dimensions, and to provide a signal in blind signal separation. An object of the present invention
is to provide a signal arrival direction estimation apparatus and method, and a signal separation
apparatus and method which can contribute to improvement of resolution.
[0005]
Another object of the present invention is to provide a computer program for realizing the signal
incoming direction estimation device and the signal separation device of the present invention
using a computer.
[0006]
In order to solve the above problems, a signal arrival direction estimation apparatus according to
the present invention comprises: conversion means for converting a signal detected by a plurality
of non-linearly arranged plurality of sensors into a signal in a frequency domain; A separation
matrix estimation means for blindly calculating a separation matrix for each frequency by
independent component analysis from signals in the frequency domain, and three sensors from
the plurality of sensors are selected, and the selected three sensors select each frequency. A first
direction-of-arrival estimation means for calculating the frequency of the signal arrival from the
separation matrix of the above, a first arrival direction estimation means for calculating the
direction of arrival of the signal from the signal arrival direction for the frequency, and Three
10-04-2019
2
sensors are adaptively selected for the signal arrival direction calculated by the first arrival
direction estimation means, and the selected three sensors recalculate the signal arrival direction
by frequency A second frequency-direction estimation means that, from said re-calculated
frequency-signal incoming direction, characterized in that a second arrival direction estimating
means for calculating the signal arrival direction.
[0007]
In the signal arrival direction estimation apparatus according to the present invention, the first
inter-sensor arrival direction estimation means is a combination in which the inter-sensor
distance is within a half wavelength corresponding to the signal maximum frequency and the
linearity is low. It is characterized in that each sensor is selected.
[0008]
In the signal arrival direction estimation apparatus according to the present invention, the second
frequency-specific arrival direction estimation means comprises two sensors having a large
aperture length with respect to the signal arrival direction calculated by the first arrival direction
estimation means. Is selected, and the combination of the two sensors results in the selection of
one sensor which has low linearity.
[0009]
In the signal arrival direction estimation apparatus according to the present invention, the second
frequency-specific arrival direction estimation means is configured to determine the frequency
corresponding to the half wavelength longer than the distance between the three selected
sensors 1. It is characterized in that the calculation result of the arrival direction estimation
means classified by frequency is used.
[0010]
In the signal arrival direction estimation apparatus according to the present invention, the second
frequency-specific arrival direction estimation means determines the inter-sensor frequency for
the frequency corresponding to a half wavelength longer than the inter-sensor distance of the
three selected sensors 1. It is characterized in that the signal arrival direction is recalculated
using a combination of sensors whose distance is within a half wavelength of the frequency.
[0011]
The signal arrival direction estimation apparatus according to the present invention is
characterized in that only the designated frequency is to be calculated as the signal arrival
10-04-2019
3
direction for each frequency.
[0012]
In the signal arrival direction estimation apparatus according to the present invention, when
using signals detected by four or more sensors, one sensor is on a plane different from the plane
on which the other three or more sensors are arranged. When it is arranged in the above, it is
characterized in that the direction of arrival of a signal is estimated in each three-dimensional
coordinate system based on at least four planes formed by at least four sensors including the one
sensor.
[0013]
In the signal arrival direction estimation method according to the present invention, a first
process of converting a signal detected by three or more non-linearly arranged sensors into a
signal in the frequency domain, and independence from the signal in the frequency domain
Second step of calculating the separation matrix for each frequency by component analysis and
three sensors from the plurality of sensors are selected, and the selected three sensors select the
separation matrix for each frequency by frequency Adaptive to the signal arrival direction
calculated in the fourth process, the fourth process of calculating the signal arrival direction from
the signal arrival direction according to the frequency, and the third process of calculating the
signal arrival direction of Direction from the signal arrival direction according to the frequency
according to the fifth step of selecting 3 sensors and recalculating the signal arrival direction
according to frequency with the selected 3 sensors Calculate Characterized in that it comprises a
sixth step of.
[0014]
A computer program according to the present invention has a first function of converting signals
detected by three or more non-linearly arranged sensors into signals in the frequency domain,
and independent component analysis from the signals in the frequency domain. The second
function of calculating the separation matrix for each frequency blindly, and three sensors
selected from the plurality of sensors, the three selected sensors, the signal arrival for each
frequency from the separation matrix for each frequency A third function of calculating a
direction, a fourth function of calculating a signal arrival direction from the signal arrival
direction according to the frequency, and 3 adaptively to the signal arrival direction calculated by
the fourth function The signal arrival direction is selected from the fifth function of selecting the
number of sensors and recalculating the direction of arrival of signals by frequency with the
selected three sensors, and the direction of arrival of signals by frequency recalculated above.
10-04-2019
4
Characterized in that to achieve a sixth feature of the output to the computer.
As a result, the above-described signal arrival direction estimation apparatus can be realized
using a computer.
[0015]
A signal separation apparatus according to the present invention comprises: conversion means
for converting signals detected by a plurality of non-linearly arranged sensors into signals in the
frequency domain; and frequency analysis by independent component analysis from signals in
the frequency domain. Separation matrix estimation means for blindly calculating the separation
matrix for each, three sensors selected from the plurality of sensors, and the signal arrival
directions for each frequency from the separation matrix for each frequency by the selected
three sensors Calculated by the first arrival direction estimation means for calculating the
frequency, the first arrival direction estimation means for calculating the signal arrival direction
from the signal arrival direction for each frequency, and the first arrival direction estimation
means A second frequency-based arrival direction estimation unit that selects three sensors
adaptively to the signal arrival direction and recalculates the signal arrival direction by frequency
using the selected three sensors; The separation for each frequency based on the signal arrival
direction calculated by the second arrival direction estimation means for calculating the signal
arrival direction from the recalculated signal arrival direction for each frequency and the second
arrival direction estimation means And a permutation solution means for solving a permutation
on a matrix.
[0016]
In the signal separation device according to the present invention, reliability determination
means for determining the reliability of the calculation result of the signal arrival direction for
each frequency based on the signal arrival direction calculated by the second arrival direction
estimation means; Beamforming means for separating a signal corresponding to the signal arrival
direction calculated by the second arrival direction estimation means from the signal in the
frequency domain, or the second arrival direction estimation based on inter-frequency signal
correlation Signal permutation means for rearranging the signals corresponding to the signal
arrival directions calculated by the means; and the permutation solution means comprises the
second arrival direction estimation means in accordance with the determination result of the
reliability. The present invention is characterized in that the beamforming means or the signal
substitution means is properly used to rearrange the separation matrix for each frequency.
[0017]
10-04-2019
5
The signal separation apparatus according to the present invention is characterized in that only
the designated frequency is the calculation target of the signal arrival direction according to the
frequency, and the beamforming means or the signal substitution means is used for the
frequency other than the designation. Do.
[0018]
A signal separation method according to the present invention comprises: a first process of
converting signals detected by three or more non-linearly arranged sensors into signals in the
frequency domain; and independent component analysis from the signals in the frequency
domain. The second process of calculating the separation matrix for each frequency by frequency
and three sensors from the plurality of sensors, and the selected three sensors select signals for
each frequency from the separation matrix for each frequency A third process of calculating an
arrival direction, a fourth process of calculating a signal arrival direction from the signal arrival
direction by frequency, and a signal arrival direction calculated by the fourth process adaptively
The signal arrival direction is calculated from the fifth process of selecting three sensors and
recalculating the signal arrival direction by frequency with the selected three sensors, and the
signal arrival direction by frequency recalculated above. Second Comprising the steps of, based
on said sixth signal incoming direction calculated by the process of, characterized in that it
comprises a seventh step of solving the permutation (Permutation) relates to the separation
matrix for each of the frequencies.
[0019]
A computer program according to the present invention has a first function of converting signals
detected by three or more non-linearly arranged sensors into signals in the frequency domain,
and independent component analysis from the signals in the frequency domain. The second
function of calculating the separation matrix for each frequency blindly, and three sensors
selected from the plurality of sensors, the three selected sensors, the signal arrival for each
frequency from the separation matrix for each frequency A third function of calculating a
direction, a fourth function of calculating a signal arrival direction from the signal arrival
direction according to the frequency, and 3 adaptively to the signal arrival direction calculated by
the fourth function The signal arrival direction is selected from the fifth function of selecting the
number of sensors and recalculating the direction of arrival of signals by frequency with the
selected three sensors, and the direction of arrival of signals by frequency recalculated above.
Allowing a computer to realize a seventh function that solves a permutation (Permutation) on the
separation matrix for each frequency based on a sixth function to be output and a signal arrival
direction calculated by the sixth function. It is characterized by
10-04-2019
6
As a result, the aforementioned signal separation apparatus can be realized using a computer.
[0020]
According to the present invention, a good degree of angular resolution can be obtained when
estimating the arrival direction of a signal in three dimensions, and the estimation accuracy of
the arrival direction of the signal in three dimensions is improved.
This can contribute to the improvement of signal resolution in blind signal separation.
[0021]
Hereinafter, an embodiment of the present invention will be described with reference to the
drawings.
[0022]
[Signal Arrival Direction Estimation Device] FIG. 1 is a block diagram showing a configuration of
a signal arrival direction estimation device 100 according to an embodiment of the present
invention.
Signal arrival direction estimation apparatus 100 shown in FIG. 1 includes N (N is an integer of 3
or more) sensors 1-1 to N (hereinafter referred to as “sensor 1” unless otherwise specified),
amplifier 2, and A / D conversion unit 3, frequency domain conversion unit 4, separation matrix
estimation unit 5, frequency-based arrival direction estimation units 6 and 9, first stage arrival
direction estimation unit 7, sensor adaptive selection unit 8, A second stage arrival direction
estimation unit 10 and an estimated direction output unit 11 are provided.
[0023]
Examples of the sensor 1 include a microphone that detects an audio signal and converts it into
an electrical signal.
10-04-2019
7
Examples of the other sensor 1 include an antenna, an optical sensor, a temperature sensor, a
pressure sensor, a flow sensor, and the like.
[0024]
The sensors 1-1 to N are non-linearly arranged so as not to be arranged on a straight line.
Thereby, the sensors 1-1 to N are arranged on at least one plane.
The sensors 1-1 to N form a sensor array.
For example, FIG. 2 shows an arrangement example where six sensors 1-1 to 6 are used.
In FIG. 2, the sensor arrangement is shown by the XY plane coordinate system, and six sensors 11 to 6 are arranged in a non-linear manner in the XY plane.
[0025]
The sensor 1 outputs a detected signal (hereinafter referred to as an observation signal).
The amplifier 2 amplifies the observation signals output from the sensors 1-1 to N, respectively.
The A / D conversion unit 3 converts the observation signals of the respective sensors 1-1 to N
after amplification into digital signals.
[0026]
10-04-2019
8
The frequency domain conversion unit 4 performs Fourier transform processing of converting
each of the digitized observation signals of the sensors 1-1 to N into signals in the frequency
domain from the time domain.
Specifically, a short time FFT (Fast Fourier Transform) is performed on each of observation
signals of the sensors 1-1 to N.
The observation signal is converted into a time signal in the frequency domain by the short time
FFT, and a time signal sequence (observation signal vector) at each frequency is obtained.
The observed signal vector X (f, t) is expressed by equation (1).
[0027]
[0028]
The separation matrix estimation unit 5 blindly calculates a separation matrix for each frequency
by independent component analysis from the observation signal vector X (f, t).
In the separation matrix calculation process, the separation matrix is converged without a
teacher signal for each frequency in the frequency domain.
Specifically, for example, the separation matrix W (f) can be calculated using equation (2).
Thereby, the separation matrix W (f) for each frequency is obtained.
[0029]
[0030]
In the blind signal separation method based on the independent component analysis, the mixed
10-04-2019
9
signal is separated using the separation matrix W (f).
The separated signal vector Y (f, t) is obtained by equation (3).
[0031]
[0032]
The per-frequency arrival direction estimation unit 6 calculates the arrival direction of the signal
for each frequency based on the separation matrix W (f).
[0033]
Each row of the separation matrix obtained by the independent component analysis forms a
separation filter in the frequency domain in which signals coming from one direction are
suppressed while signals coming from one direction are taken out.
Thus, after obtaining the separation matrix (separation filter) at each frequency, it is possible to
estimate the arrival direction of the signal extracted by each row of the separation matrix by the
inverse matrix of the separation matrix.
Further, the phase difference between the sensors 1 can be obtained from each column of the
inverse matrix of the separation matrix at each frequency, and the arrival direction of the signal
can be estimated using the phase difference.
[0034]
Also, the column vector of the inverse matrix W (f) <-1> of the separation matrix W (f) is
approximately proportional to the column vector of the complex mixing matrix H (f). From this, it
is possible to obtain the pseudo inverse matrix W (f) <+> of the separation matrix W (f).
[0035]
10-04-2019
10
The frequency-specific arrival direction estimation unit 6 first obtains a pseudo inverse matrix W
(f) <+> of the separation matrix W (f). Then, for each column of the pseudo inverse matrix W (f)
<+>, the zenith angle θ and the azimuth angle φ are calculated using the elements
corresponding to the three non-linearly arranged sensors 1.
[0036]
Here, the frequency-based arrival direction estimation unit 6 selects three sensors 1 used for
frequency-based arrival direction estimation from the sensors 1-1 to N. The selection condition is
three sensors in which the inter-sensor distance is within a half wavelength corresponding to the
maximum signal frequency and which is a combination of low linearity. Note that the linearity of
a triangle with three sensors at the top is defined as the difference between the maximum
interior angle and the minimum interior angle of the triangle. Thus, a combination of low
linearity (three sensors) refers to one having a small difference between the maximum internal
angle and the minimum internal angle of a triangle having three sensors as apexes. The least
linear combination is one in which a triangle having three sensors at its apex is an equilateral
triangle.
[0037]
The signal arrival direction by frequency is obtained by the equations (4) to (7). From the pseudo
inverse matrix W (f) <+> of the separation matrix W (f), the equations (4) and (5) hold.
[0038]
[0039]
[0040]
Then, the arrival direction (θ, φ) of the l-th signal is obtained by the equations (6) and (7).
[0041]
10-04-2019
11
[0042]
[0043]
Note that arg (z) represents the phase angle of complex number z (an angle formed by a straight
line connecting a point representing a complex number on the complex plane and the origin with
the real axis).
[0044]
The first stage arrival direction estimation unit 7 calculates the arrival direction of the signal
based on the arrival direction (θ, φ) according to frequency of the estimation result of the
arrival direction estimation unit 6 according to frequency.
Specifically, the arrival direction (θ, φ) estimated for each frequency is statistically processed,
and the arrival direction of the signal is determined from the result of the statistical processing.
For example, θ and φ estimated for each frequency are averaged, and the average value is
adopted as the arrival direction (θ, φ) of the signal.
In addition, you may use methods other than the said average as a method of a statistical
process.
For example, histograms of θ and φ estimated for each frequency may be calculated, and the
arrival directions (θ and φ) of the signal may be determined from the histograms.
Also, the frequencies to be statistically processed may be all frequencies or some frequencies.
[0045]
10-04-2019
12
In the first direction of arrival direction estimation processing described above, in the directionof-arrival estimation by frequency, although the estimation accuracy at high frequencies is good,
the estimation accuracy at low frequencies may not be very good.
For this reason, it is difficult to obtain a good degree of angular resolution. Therefore, the sensor
adaptive selection unit 8, the frequency-based arrival direction estimation unit 9, and the second
stage arrival direction estimation unit 10 perform the second stage arrival direction estimation
processing.
[0046]
The sensor adaptive selection unit 8 adaptively selects three sensors 1 with respect to the signal
arrival direction calculated by the first stage arrival direction estimation unit 7. The selection
condition of these three sensors 1 is to select two sensors 1 having a large aperture length with
respect to a certain signal arrival direction, and further to combine these two sensors 1 with one
sensor having low linearity. Select 1
[0047]
The frequency-based arrival direction estimation unit 9 receives the signal arrival directions (θ,
φ) according to the frequency by the three sensors 1 selected by the sensor adaptive selection
unit 8 in the same manner as the frequency-based arrival direction estimation unit 6 described
above. Calculate). Next, the second stage arrival direction estimation unit 10 generates a signal in
the same manner as the first stage arrival direction estimation unit 7 based on the arrival
direction (θ, φ) according to frequency according to the estimation result of the arrival
direction estimation unit 9 according to frequency. The arrival direction (θ, φ) of is calculated.
[0048]
In the second stage of the arrival direction estimation processing by the sensor adaptive selection
unit 8, the frequency-specific arrival direction estimation unit 9, and the second stage arrival
direction estimation unit 10 described above, each signal arrival calculated by the first stage
arrival direction estimation unit 7 Do each for the direction. In the second direction of arrival
direction estimation processing, the distance between sensors, that is, the distance between
10-04-2019
13
sensors, is obtained by using two sensors 1 having a large aperture length with respect to the
signal arrival direction calculated by the first step arrival direction estimation unit 7. The phase
difference is increased, and the angle resolution in the direction of arrival estimation at low
frequencies is improved. As a result, the accuracy of the estimation of the direction of arrival at
low frequencies that are not good in the first stage of the estimation of the direction of arrival is
improved, so that the estimation result of the second stage of arrival direction estimation unit 10
obtains a good degree of angular resolution. It becomes possible.
[0049]
In order to avoid spatial aliasing at high frequencies, the frequency-specific arrival direction
estimation unit is used for the frequency corresponding to a half wavelength longer than the
distance between the sensors of the three sensors 1 selected by the sensor adaptive selection
unit 8 It is preferable to use the calculation result of 6. Or about the said frequency, it is
preferable to recalculate a signal arrival direction using the combination of the sensor 1 whose
distance between sensors is less than the half wavelength of the said frequency.
[0050]
The estimated direction output unit 11 outputs the signal arrival direction (θ, φ) obtained by
the second stage arrival direction estimation unit 10.
[0051]
According to the embodiment described above, it is possible to obtain a good degree of angular
resolution when estimating the direction of arrival of a signal in three dimensions.
This improves the estimation accuracy of the signal arrival direction in three dimensions. As a
result, it can contribute to the improvement of signal resolution in blind signal separation.
[0052]
Note that the frequency to be subjected to the estimation of the direction of arrival of the above
signal may be determined in advance based on actual measurement. The estimation accuracy for
10-04-2019
14
each frequency changes in accordance with the arrangement interval of the sensor 1. Therefore,
it is preferable to estimate the direction of arrival using the actual signal at each frequency after
determining the placement of the sensor 1, evaluate the estimation result, and select the
frequency to be included in the estimation target of the direction of arrival. As a result,
frequencies that can not be expected to have a desired estimation accuracy are excluded from
estimation targets to reduce the amount of calculation.
[0053]
Also, in the case of using four or more sensors including at least three sensors arranged in a nonlinear manner, when all are arranged on the same plane, the estimation method in the case of
using three sensors described above is applied as it is it can. On the other hand, in the case of
using four or more sensors including at least three sensors arranged non-linearly, one sensor is
arranged on a plane different from the plane on which the other three or more sensors are
arranged. In this case, at least four planes are formed by at least four sensors including the one
sensor. Then, the arrival direction of the signal can be estimated for each three-dimensional
coordinate system based on each plane.
[0054]
[Signal Separation Device] Next, a signal separation device to which the signal incoming direction
estimation device 100 according to the above-described embodiment is applied will be described.
Hereinafter, parts corresponding to the parts in FIG. 1 are given the same reference numerals. In
the signal separation device according to the present embodiment, the separation matrix W (f) is
rearranged using the arrival direction (θ, φ) of the estimation result of the signal arrival
direction estimation device 100. The procedure is shown below.
[0055]
Step S1; Based on the arrival direction (θ, φ) obtained by the second stage arrival direction
estimation unit 10, the reliability of the arrival direction for each frequency estimated by the
arrival direction estimation unit 9 for each frequency is determined. Specifically, the arrival
direction estimated by the second stage arrival direction estimation unit 10 is compared with the
arrival direction estimated for each frequency, and if the absolute difference between the two is
within a predetermined range, the estimation result of the frequency is relied upon If the
10-04-2019
15
absolute difference between the two is out of a predetermined range, it is determined that the
estimation result of the frequency is not reliable.
[0056]
Step S2: With respect to the frequency determined to be reliable in step S1, the signal Y (f, t) is
rearranged in accordance with the estimated direction of arrival. The signal Y (f, t) is calculated
by the above equation (3). As a result, for the frequency determined to be reliable, a signal Y (f, t)
in which the problem of "Permutation" has been solved is obtained.
[0057]
Step S3; The signal Y (f, t) is newly separated by beamforming for the frequency determined to be
unreliable in step S1. Also, the signal Y (f, t) is separated by beamforming even for frequencies
that were not used to estimate the direction of arrival of the signal.
[0058]
In the beamforming, in order to form a directional beam in the arrival direction estimated by the
second stage arrival direction estimation unit 10, at a frequency determined to be unreliable and
a frequency not used to estimate the arrival direction of the signal. Calculate the weight
coefficient matrix of Then, it is used for estimation of the frequency determined to be unreliable
from the signal in the frequency domain (the output signal of the frequency domain conversion
unit 4) and the arrival direction of the signal by beamforming using the calculated weighting
coefficient matrix. The signal Y (f, t) of the frequency not present is separated.
[0059]
In step S3, the separation matrix may be rearranged based on inter-frequency signal correlation,
instead of separating signals by beamforming. In this case, the correlation of the signals between
the frequencies is calculated, and the signals corresponding to the arrival direction estimated by
the second stage arrival direction estimation unit 10 are rearranged based on the inter-frequency
signal correlation of the calculation result. As a result, a signal Y (f, t) in which the problem of
10-04-2019
16
“Permutation” has been solved is determined for the frequency determined to be unreliable
and the frequency not used to estimate the direction of arrival of the signal.
[0060]
Step S4; The signals Y (f, t) in all frequency bands are aligned by the signal Y (f, t) obtained in
step S2 and the signal Y (f, t) obtained in step S3. The signal Y (f, t) in the entire frequency band
is inverse Fourier transformed to be converted into a time domain signal. This allows each
incoming signal to be separated in the time domain.
[0061]
According to the embodiment described above, it is possible to obtain a good degree of angular
resolution when estimating the direction of arrival of a signal in three dimensions, and to
improve the estimation accuracy of the direction of arrival of a signal in three dimensions.
Improves the signal resolution in
[0062]
In addition, since blind signal separation and beamforming are selectively used according to the
reliability of the estimation result of the arrival direction of the signal in the frequency domain, it
can contribute to maintenance and improvement of the separation accuracy of the signal.
[0063]
FIGS. 3 and 4 are plots showing the estimation results of the direction of arrival of the signal
according to frequency according to the present embodiment.
In FIG. 3 and FIG. 4, the lower the frequency number, the lower the frequency, and the higher the
frequency number, the higher the frequency.
The embodiment according to FIG. 3 and FIG. 4 uses three microphones as sensors, and the three
microphones detect audio signals generated from two sound sources. As the sound source, the
voice of the ATR phoneme balanced sentence is used. The sampling frequency is 8 kHz, and the
FFT size is 1024 points. And, the separation matrix is obtained by FastICA. The coordinate values
10-04-2019
17
(x, y, z) of the three microphones are (0, 0, 0), (4, 0, 0), (0, 4, 0) centimeters, respectively (see FIG.
2). Corresponds to sensors 1-1, 1-2 and 1-4).
[0064]
In FIG. 3, the estimation result of the azimuth angle φ for each frequency is plotted. As shown in
FIG. 3, it can be seen that the azimuth angle φ of the two sound sources can be accurately
estimated. In FIG. 4, the estimation results of the zenith angle θ corresponding to the two
azimuthal angles φ are plotted.
[0065]
Also, in FIGS. 5 and 6, in the case of detecting voice signals generated from three sound sources
by six microphones under the same conditions as FIG. 3 and FIG. The results (azimuth angle φ
only) are plotted. The coordinate values (x, y, z) of the six microphones are (0, 0, 0), (4, 0, 0), (8,
0, 0), (0, 4, 0), ( 4, 4, 0), (8, 4, 0) centimeters (corresponding to sensors 1-1, 1-2, 1-3, 1-4, 1-5, 16 in FIG. 2). In the case of FIGS. 5 and 6, the separation matrix is determined by the six
microphones. In addition, three microphones having the coordinate values (x, y, z) of (0, 0, 0), (4,
0, 0), (0, 4, 0) centimeters (the sensor 1-1 of FIG. 1, 1-2 and 1-4), the direction of arrival of the
signal according to frequency is estimated. 5 and 6, the lower the frequency number, the lower
the frequency, and the higher the frequency number, the higher the frequency.
[0066]
In FIG. 5, the azimuth angle φ for each frequency in the estimation result of the frequency
direction arrival direction estimation unit 6 is plotted in the first direction of arrival direction
estimation processing. In this first step, since the zenith angle θ of the sound source 1
(corresponding to φ1) and the sound source 3 (corresponding to φ3) is small, the estimated
value of the azimuth angle φ in the low frequency band has a large variation and 0 degrees It
shows a tendency towards approaching.
[0067]
10-04-2019
18
In FIG. 6, the azimuth angle φ for each frequency in the estimation result of the frequency
direction arrival direction estimation unit 9 is plotted in the second direction of the arrival
direction estimation processing. At this second stage, the variation of the estimated values is
generally smaller than at the first stage. In particular, the low frequency part of the sound source
1 (corresponding to φ1) and the sound source 3 (corresponding to φ3) having a small zenith
angle θ is significantly improved. This is because a combination of microphones with large intersensor phase difference is used. By improving the angular resolution in this way, the problem of
"Permutation" in the low frequency part can be easily solved.
[0068]
A program for realizing the functions of the signal incoming direction estimation apparatus and
the signal separation apparatus according to the present embodiment is recorded in a computer
readable recording medium, and the computer system reads the program recorded in the
recording medium. The signal arrival direction estimation process and the signal separation
process may be performed by performing the process. Note that the “computer system”
referred to here may include an OS and hardware such as peripheral devices. The "computer
system" also includes a homepage providing environment (or display environment) if the WWW
system is used. In addition, “computer readable recording medium” refers to flexible disks,
magneto-optical disks, ROMs, writable nonvolatile memories such as flash memories, portable
media such as CD-ROMs, hard disks incorporated in computer systems, etc. Storage devices.
[0069]
Furthermore, the “computer-readable recording medium” is a volatile memory (for example,
DRAM (Dynamic Memory) inside a computer system that becomes a server or a client when a
program is transmitted via a network such as the Internet or a communication line such as a
telephone line). As Random Access Memory), it is assumed that the program which holds the
program for a fixed time is included. The program may be transmitted from a computer system
in which the program is stored in a storage device or the like to another computer system via a
transmission medium or by transmission waves in the transmission medium. Here, the
“transmission medium” for transmitting the program is a medium having a function of
transmitting information, such as a network (communication network) such as the Internet or a
communication line (communication line) such as a telephone line. Further, the program may be
for realizing a part of the functions described above. Furthermore, it may be a so-called
difference file (difference program) that can realize the above-described functions in combination
with a program already recorded in the computer system.
10-04-2019
19
[0070]
The embodiment of the present invention has been described in detail with reference to the
drawings, but the specific configuration is not limited to this embodiment, and design changes
and the like within the scope of the present invention are also included.
[0071]
FIG. 1 is a block diagram showing a configuration of a signal incoming direction estimation
apparatus 100 according to an embodiment of the present invention.
It is a two-dimensional coordinate figure showing the example of arrangement at the time of
using six sensors 1-1-6 concerning the embodiment. It is a plot which shows the estimation result
of the arrival direction of the signal according to frequency based on one Embodiment of this
invention. It is a plot which shows the estimation result of the arrival direction of the signal
according to frequency based on one Embodiment of this invention. It is a plot which shows the
estimation result of the arrival direction of the signal according to the frequency according to
one embodiment of the present invention. It is a plot which shows the estimation result of the
arrival direction of the signal according to frequency of the 2nd step concerning one
embodiment of the present invention.
Explanation of sign
[0072]
1 (1-1 to N) ... sensor, 2 ... amplifier, 3 ... A / D conversion unit, 4 ... frequency domain conversion
unit, 5 ... separation matrix estimation unit, 6, 9 ... arrival direction estimation unit by frequency,
7 ... First stage arrival direction estimation unit, 8: sensor adaptive selection unit, 10: second
stage arrival direction estimation unit, 11: estimated direction output unit, 100: signal arrival
direction estimation apparatus
10-04-2019
20
Документ
Категория
Без категории
Просмотров
0
Размер файла
32 Кб
Теги
description, jp2007240209
1/--страниц
Пожаловаться на содержимое документа