close

Вход

Забыли?

вход по аккаунту

?

JP2009037032

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 JP2009037032
An object signal is extracted effectively. SOLUTION: A signal from a specific signal source is
emitted from an observation signal R (p) of a B channel obtained by observing a mixed signal of
signals emitted from a plurality of signal sources by B (B ≧ 2) sensors. The noise signal N (p)
contained in the observation signal R (p) is detected (120, S4), and the noise signal N (p) and the
observation signal R (p) are extracted. The whitening filter coefficient w (p) is calculated (112,
S6), and the noise correlation function which is the inter-channel correlation function of the noise
signal N (p) included in the observed signal of B channel is calculated (130, S14) , And B:
Calculate the observed correlation function which is the correlation function between the
observed signals of the B channel (4, S10), and obtain the inverse filter coefficient c (p) using the
observed correlation function and the noise correlation function (140, S16 ), The inverse filter
coefficient to the observation signal R (p) Crowded only adds (3, S20). [Selected figure] Figure 9
Signal extraction apparatus, method thereof, and program thereof
[0001]
The present invention relates to a signal extraction apparatus, a method thereof, and a program
thereof for well extracting a target signal from mixed signals emitted from a plurality of signal
sources.
[0002]
In recent years, with the advancement of multimedia technology, communication conferences
such as video conferences in the form of a speech communication using a sensor (for example, a
04-05-2019
1
microphone) and a speaker have become widespread.
In that case, natural speech can be made without being aware of the microphones without
placing microphones for the number of speakers on the desk, and noise and reverberation that
degrades the voice quality can be suppressed, and a signal for observing only the target voice
signal An extractor is needed.
[0003]
As such prior art, there is a signal extraction device that suppresses reverberation and noise by
using a plurality of microphones. The details of this signal extraction device are described in
Patent Document 1. FIG. 1 shows an example of the functional configuration of a signal
extraction apparatus 500 according to the prior art. The number of signal sources is A, and
among them, the number of sound sources of the target signal (target sound sources) is one, and
the number of sound sources of noise signals (noise source) is A−1. Let the number of
microphones be B. However, A and B are integers of 2 or more. B microphones 1 b (b = 1,..., B)
are connected to the signal extraction device 500. The signal extraction device 500 includes a
whitening filter coefficient calculation unit 110, an observed correlation function calculation unit
4, an inverse filter coefficient calculation unit 5, B whitening filter units 100b (b = 1,..., B), B
pieces The filter unit 2 b (b = 1,..., B) of FIG. Also, using the discrete time p, the target signal is M
(p), the noise signal is N (p), and the observation signal Rb (p) observed by the microphone 1b.
The observation signal Rb (p) is a signal in which the target signal Mb (p) and the noise signal Nb
(p) are mixed.
[0004]
First, the observation signal Rb (p) from the microphone 1b is input to the whitening filter
coefficient calculation unit 110 and the filter unit 2b. The whitening filter coefficient calculation
unit 110 estimates the average spectrum of the target signal M (p) from the observation signal
Rb (p). Then, the coefficients of the whitening filter are calculated to flatten the average
spectrum. The calculation of the whitening filter coefficients is performed as follows. First, the
signal R1 (p) of the channel (for example, channel 1) specified in advance from the observation
signal Rb (p) is divided into F frames. Let the frame length of all the frames be G (for example,
256). Then, the autocorrelation function U1f (p) of the f (f = 1,..., F) -th frame is calculated by the
following equation. U1f (p) =. SIGMA.qR1 (p) R1 (p + q) (1) where q = G.times. (F-1), G.times. (F-1)
+1,. . . , G × f−2, G × f−1 p = −G,. . . , 0,. . . , G. This is averaged for a frame, and an average
04-05-2019
2
autocorrelation function U <-> 1 (p) is calculated by the following equation. U <-> 1 (p) =. SIGMA.f
= 1 <F> U1f (p) (2)
[0005]
By converting the average autocorrelation function U <-> 1 (p) into the frequency domain, that is,
the following equation (3), the average spectrum V1 (k) of the observed signal can be obtained.
For example, Fourier transform may be used as the conversion method to the frequency domain.
V1 (k) = FFT (U <-> 1 (p)) (3) However, FFT (U <-> 1 (p)) is Fourier-transformed with respect to U
<-> 1 (p) Yes, k indicates the frequency.
[0006]
As another method of calculating the average spectrum, there is also a method of calculating the
spectrum of the signal R1 (p) for each frame and taking an average for the frame as shown in the
following equation (4). V1 (k) =. SIGMA.f = 1 <F> | FFT (u1 (p)) | (4)
[0007]
Next, the spectrum W1 (k) of the whitening filter is obtained by calculating the reciprocal of the
calculated average spectrum as in the following equation (5). W1 (k) = 1 / V1 (k) (5) The filter
coefficient w1 (p) of the whitening filter performs an inverse Fourier transform IFFT on this
spectrum W1 (k) as in the following equation (6) to obtain a window It will be calculated
separately. There are Hanning window, Hamming window, square window, triangular window,
Kaiser window etc as window types. w1 (p) = window (IFFT (W1 (k))) (6) These processes are
performed for each channel b (b = 1,..., B) to obtain a whitening filter coefficient wb (p).
[0008]
Next, the whitening filter unit 100b (b = 1,..., B) convolves the whitening filter coefficient wb (p)
with the observation signal Rb (p) as shown in the following equation (7). , Whitening the
observation signal Rb (p). Hereinafter, the whitened observation signal is referred to as a
whitened observation signal. The observation correlation function calculator 4 calculates the
04-05-2019
3
correlation function r'11 (p), r'12 (p),... Between the whitened observation signals R'n (p) (n = 1,...,
N). . . , R'1B (p), r'21 (p),. . . , R'2B (p),. . . , R'B1 (p),. . . , R ′ BB (p). Hereinafter, this correlation
function is represented as r 'ij (p) (i = 1, ..., B, j = 1, ..., B). Here, r'ij (p) =. SIGMA.qR'i (q) R'j (q + p)
(8).
[0009]
The inverse filter coefficient calculation unit 5 calculates an inverse filter by solving the following
simultaneous linear equations. The inverse filter coefficient is capable of suppressing the noise
signal and extracting the target signal by convoluting the observation signal in which the target
signal and the noise signal are mixed.
[0010]
Here, R is a matrix of interchannel correlation functions, R ij is a matrix of correlation functions
of the i-th microphone 1i and the j-th microphone 1j, c is a vector of inverse filter coefficients to
be obtained, cb is the n-th inverse filter coefficients , D is the blind objective impulse response
coefficient vector, db is the b-th blind objective impulse response vector, c b (L) is the b-th
inverse filter coefficient, B is the number of microphones, and L is the number of taps of the
inverse filter is there. δ b is “1” when the target sound source 61 is closest to the b-th
microphone 1 b among the microphones, and “0” otherwise.
[0011]
The simultaneous linear equations of equation (9) are solved, and the inverse filter coefficient
vector c is calculated to obtain the inverse filter coefficient cb (p). In order to solve the
simultaneous linear equations of Equation (4), the same conditions as the MINT theory (the
following Equations (10) and (11) must be satisfied. B> A + 1 (10) L = A (K-1) / (N−A) (11) where
A is the number of sound sources and K is the number of impulse response taps. For MINT
theory, see "M. Miyoshi and Y. Kaneda," Invese Filtering of acoustics, "IEEE Trans Acoust. Speech
Signal Process., Vol. ASSP-36, no 2, pp. 145-152, Feb. 1998. "It is described in.
[0012]
04-05-2019
4
Each filter unit 2b obtains the signal Sb (p) by convolving the inverse filter coefficient cb (p) with
the observation signal Rb (p). The output Sb (p) of each filter unit 2b is all added by the adder 3,
and the addition result is output as the target signal S (p). The output signal S (p) suppresses
noise and reverberation and extracts only the target sound. Unexamined-Japanese-Patent No.
2006-66989
[0013]
When implementing the conventional signal extraction apparatus 500, it had to be implemented
in the environment where the microphone closest to the target sound source and the microphone
closest to the noise source exist separately. Under this environment, in the simultaneous linear
equations of the above equation (9), the number of equations was insufficient, and it was not
possible to obtain an appropriate inverse filter coefficient. Therefore, there is a problem that
although the reverberation of the speech to be extracted is suppressed, the noise is not removed.
[0014]
An object of the present invention is to provide a signal extraction device, its method, and its
program for effectively removing reverberation and noise of a target signal by obtaining an
inverse filter coefficient in consideration of a noise signal.
[0015]
The present invention is generated from a specific signal source from B-channel observation
signals obtained by respectively observing mixed signals of signals respectively emitted from a
plurality of signal sources with B (B ≧ 2) sensors. It is a signal extraction device that extracts a
target signal.
The signal extraction apparatus includes a noise correlation function calculation unit, an
observation correlation function calculation unit, a weighted inverse filter coefficient calculation
unit, a filter unit, and an addition unit. The noise correlation function calculation unit calculates a
noise correlation function which is an inter-channel correlation function of the noise signal
included in each of the B channel observation signals. The observation correlation function
calculation unit calculates an observation correlation function which is an inter-channel
correlation function of the B-channel observation signal. The weighted inverse filter coefficient
04-05-2019
5
calculation unit obtains the inverse filter coefficient using the observation correlation function
and the noise correlation function. The filter unit convolves inverse filter coefficients into the
observation signal. The addition unit generates the target signal by adding the output of the filter
unit.
[0016]
Furthermore, it may have a whitening filter coefficient calculation unit and a whitening filter unit.
The whitening filter coefficient calculation unit calculates the whitening filter coefficient using
the average spectrum of the observation signal. The whitening filter unit convolves a whitening
filter coefficient to the input signal to whiten the signal. In this case, the observation correlation
function calculation unit calculates an inter-channel correlation function of the B-channel
observation signal whitened by the whitening filter unit, and outputs it as an observation
correlation function.
[0017]
Furthermore, the noise correlation function calculation unit may calculate an inter-channel
correlation function of the noise signal whitened by the whitening filter unit and output it as a
noise correlation function.
[0018]
Furthermore, a noise section detection unit that detects a noise signal included in the observation
signal may be provided.
In this case, the noise correlation function calculator calculates a noise correlation function
between the detected noise signals. Furthermore, in this case, the whitening filter coefficient
calculation unit may calculate the whitening filter coefficient using the average spectrum of the
noise signal detected by the noise section detection unit and the average spectrum of the B
channel observation signal.
[0019]
04-05-2019
6
Further, the signal extraction apparatus of the present invention may include a noise section
detection unit, a noise correlation function calculation unit, an observation correlation function
calculation unit, a weighted inverse filter coefficient calculation unit, a filter unit, and an addition
unit. The noise section detection unit detects noise section information including only a noise
signal from the observation signal of the B channel. The noise correlation function calculation
unit calculates a noise correlation function, which is an inter-channel correlation function of
noise signals respectively included in the B channel observation signal, from the B channel
observation signal and the noise section information. The observation correlation function
calculation unit, the weighted inverse filter coefficient calculation unit, the filter unit, and the
addition unit are the same as described above.
[0020]
Furthermore, in the case of the configuration of this signal extraction device, a whitening filter
coefficient calculation unit and a whitening filter unit may be provided. The whitening filter
coefficient calculation unit calculates a whitening filter coefficient using the average spectrum of
the B-channel observed signal. The whitening filter unit convolutes the whitening filter
coefficients to the input signal. Then, the observation correlation function calculation unit
calculates a correlation function between channels of the observation signal of the B channel
whitened by the whitening filter unit, and outputs the correlation function as the observation
correlation function. The noise correlation function calculation unit calculates a noise correlation
function from the whitened observation signal of the B channel and the noise interval
information.
[0021]
In this case, the noise section detection unit may also detect a noise signal. Then, the whitening
filter coefficient calculation unit calculates a whitening filter coefficient using the detected
average spectrum of the noise signal and the average spectrum of the B channel observation
signal.
[0022]
According to the above configuration, an observed correlation function which is a correlation
function between channels of the observation signal and a noise correlation function which is a
04-05-2019
7
correlation function between channels of the noise signal are obtained. The observed correlation
function is then constrained by the noise correlation function to obtain the inverse filter
coefficients. That is, the equation reflecting the characteristics of the noise signal increases. As a
result, even if the microphone closest to the noise source and the microphone closest to the
target sound source are the same, not only the reverberation of the target signal but also noise
can be effectively suppressed.
[0023]
The following shows the best mode for carrying out the invention. In addition, the same number
is attached | subjected to the structure part which has the same function, and the process which
performs the same process, and it abbreviate | omits duplication description.
[0024]
In the description of the signal extraction apparatus of this embodiment, although the signal to
be extracted is described as an audio signal, the present invention is not limited to this.
According to the signal extraction apparatus of this embodiment, for example, it is possible to
clearly extract the target electromagnetic wave from the electromagnetic wave buried in the
noise. Further, a sensor that observes a signal is, for example, a microphone. FIG. 2 shows an
example of the functional configuration of the signal extraction device 600-1 of the first
embodiment, and FIG. 3 shows the flow of the main processing of the signal extraction device
600-1. As described in [Background Art], the number of signal sources is A, and among them, the
number of sound sources of the target signal is one, and the number of sound sources of the
noise signal is A-1. Let the number of microphones be B. However, A and B are integers of 2 or
more. B microphones 1 b (b = 1,..., B) are connected to the signal extraction device 600-1. The
signal extraction device 600-1 includes a noise section detection unit 120, a whitening filter
coefficient calculation unit 110, a whitening filter unit 100b (b = 1,..., B), and a filter unit 2b (b =
1,. , B), observation correlation function calculator 4, noise correlation function calculator 130,
and weighted inverse filter coefficient calculator 140.
[0025]
First, when the observation signal Rb (p) is observed by the microphone 1 b (step S 2), the
observation signal Rb (p) is input to the noise section detection unit 120. The noise section
04-05-2019
8
detection unit 120 detects the noise signal Nb (p) included in the observation signal Rb (p) (step
S4). For example, the target signal Mb (p) is considered as voice, and a signal outside the voice
section is output as a noise signal Nb (p) using a general VAD (voice detection technology: voice
activity detection).
[0026]
Further, the whitening filter coefficient calculation unit 110 calculates a whitening filter
coefficient using the average spectrum of the observation signal Rb (p) from the microphone 1b
(step S6). The average spectrum and the whitening filter coefficient may be obtained by using the
above formulas (1) to (6), and the description is omitted here. The whitening filter unit 100b
convolutes the whitening filter coefficient wb (p) into the corresponding observation signal Rb (p)
according to the above equation (7) to generate a whitened observation signal R′b (p) Step S8).
From the above equation (8), the observed correlation number calculation unit 4 obtains the
observed correlation function r'ij (i = 1,...) For the whitened observed signal R'b (p). , B j = 1,..., B)
are obtained (step S10).
[0027]
On the other hand, the whitening filter unit 100b convolutes the whitening filter coefficient wb
(p) corresponding to the noise signal Nb (p) to generate a whitened noise signal N'b (p) (step
S12). The noise correlation function calculation unit 130 calculates noise correlation functions n
′ ij (p) (i = 1,..., B j = 1,...) Which are correlation functions between channels of the whitening
noise signal N ′ b (p). .., B) are obtained (step S14).
[0028]
Here, n'ij (p) =. SIGMA.qN'i (q) N'j (q + p) (12). This addition process adds reverberation time for
q. Although FIG. 2 separately shows the whitening filter portion obtained by whitening the
observed signal Rb (p) and the whitening filter portion obtained by whitening the noise signal Nb
(p), these may be used in combination. . The weighted inverse filter coefficient calculation unit
140 calculates the weighted inverse filter coefficient cb (p) using the noise correlation function N
and the observation correlation function R. Specifically, the inverse filter cb (p) is obtained by
solving the following simultaneous equations (13).
04-05-2019
9
[0029]
Where λ 1 is a correction coefficient, R is a matrix of inter-channel correlation functions, R ij is a
matrix of inter-channel correlation functions of i-th microphone 1 i and j-th microphone 1 j, c is a
vector of inverse filter coefficients, c n is vector of n-th inverse filter coefficient, d is blind
objective impulse response coefficient vector, dn is n-th blind objective impulse response vector,
cb (L) is b-th inverse filter coefficient, B is number of microphones, L is It is the number of
reverse filter taps. δ b is 1 when the target sound source 61 is closest to the b-th microphone 1
b among the microphones, and is 0 otherwise. The elements of the matrix d are BL in d1 to dB,
and the number of "0" s is also BL. Therefore, the number of elements of the matrix d is 2BL.
[0030]
The equation (13) is shown in detail in FIG. Also, in order to solve equation (13), the MINT theory
described in [Background Art] must hold.
[0031]
The weighted inverse filter coefficient cb (p) thus obtained is input to the corresponding filter
unit 2b. The filter unit 2b convolutes the weighted inverse filter coefficient cb (p) with the
observation signal Rb (p) to generate a signal M '(p). That is, the following equation (14) is
performed. The adder 3 adds the generated signal M'b (p) for all channels to generate a target
signal M (p). That is, it is obtained by the following equation. M (p) =. SIGMA.b = 1 <B> M'b (p)
(15)
[0032]
Further, if the value of λ1 in the above equation (13) is made smaller, the above equation (9)
will be approached, and the reverberation of the target signal can be suppressed, but the noise
signal can not be suppressed. Although the noise signal can be suppressed by increasing the
value of λ1, the reverberation of the target signal becomes large. Therefore, the value of λ1
may be changed as appropriate in consideration of the environment in which the target signal is
extracted. In addition, although λ1 is applied to the matrix N of the noise correlation function in
04-05-2019
10
the matrix RE in the above equation (13), λ1 may be applied to the matrix R of the observed
correlation function. In this case, if the value of λ1 is increased, the reverberation of the target
signal can be suppressed, but the noise signal can not be suppressed. If the value of λ1 is
reduced, the noise signal can be suppressed, but the reverberation of the target signal becomes
large. Alternatively, λ1 may be applied to the matrix N of the observed correlation function, and
a correction coefficient λ1 ′ different from λ1 may be applied to the matrix R.
[0033]
The observation signal Rb (p) includes the noise signal Nb (p) and the target signal Mb (p).
Folding the whitening filter coefficient wb (p) into the observation signal Rb (p) means that the
target signal Mb (p) in the observation signal can be appropriately whitened, but the noise signal
Nb in the observation signal About (p), it has not been able to whiten appropriately. Therefore,
the noise signal Nb (p) from the noise section detection unit 120 is also whitened with wb (p),
and using the whitened observation signal and noise signal, an appropriate weighted inverse
filter can be obtained. it can.
[0034]
When the matrix R used in obtaining the inverse matrix by the signal extraction device 500 is
compared with the matrix RE used in the signal extraction device 600-1, the matrix N of the
noise correlation function is added to the matrix RE. Therefore, the number of equations has
conventionally been increased, and the characteristics of the noise signal Nb (p) are also taken
into consideration. Therefore, even under an environment where the microphone closest to the
noise source and the microphone closest to the target sound source are the same, it is possible to
obtain an inverse matrix more suitable than before, and as a result, noise can be suppressed
effectively.
[0035]
[Modification 1] Next, a signal extraction device 600-2 which is a modification 1 of the
embodiment 1 will be described. An example of functional configuration of the signal extraction
device 600-2 is shown in FIG. The signal extraction device 600-2 differs from the signal
extraction device 600-1 in that the noise signal Nb (p) is not whitened. That is, the noise
correlation function calculation unit 130 obtains the inter-channel correlation function of the
04-05-2019
11
noise signal Nb (p). For example, if the noise signal has a waveform close to the whitened signal,
it is not necessary to whiten. Therefore, if the noise signal has a waveform close to the whitened
signal, this configuration can omit the whitening process of the noise signal, and the effect
similar to that of the signal extraction device 600-1. You can get
[0036]
[Modification 2] Next, a signal extraction device 600-3 which is a modification 2 of the
embodiment 1 will be described. An exemplary functional configuration of the signal extraction
device 600-3 is shown in FIG. The signal extraction device 600-3 differs from the signal
extraction device 600-1 in that the noise segment detection unit 120 is not provided and the
noise signal Nb (p) is not detected. Under circumstances where noise signals can be predicted to
some extent, it may not be necessary to detect noise signals. For example, when it is desired to
extract as a voice signal target signal of the speaker in the conference room, the noise signal is
often air conditioning in the conference room and the like, and the noise signal can be predicted
in advance. Therefore, if the noise signal to be predicted is input in advance and the noise
correlation function for the input noise signal is calculated by the noise correlation function
calculation unit 130, the noise correlation function is obtained without detecting the noise signal.
, And the same effect as the signal extraction device 600-1 can be obtained.
[0037]
[Modification 3] Next, a signal extraction device 600-4 of Modification 3 will be described. An
example of functional configuration of the signal extraction device 600-4 is shown in FIG. The
signal extraction device 600-4 differs from the signal extraction device 600-1 in that the
whitening filter unit 100b and the whitening filter coefficient calculation unit 110 are not
provided. For example, in the case of a waveform close to a whitened signal for both the noise
signal and the observation signal, it is not necessary to whiten the noise signal and the
observation signal. Therefore, the calculation process of the whitening filter coefficient, the
convolution process of the whitening coefficient to the noise signal, and the convolution process
of the whitening coefficient to the observation signal can be omitted, and the same effect as the
signal extraction device 600-1 can be obtained.
[0038]
04-05-2019
12
[Modification 4] Next, a signal extraction device 600-5 of Modification 4 will be described. Signal
extractor 600-5 is shown in FIG. The signal extraction device 600-5 differs from the signal
extraction device 600-4 in that the noise segment detection unit 120 is not provided. As
described in [Modification 2], if the noise signal can be predicted in advance and both the noise
signal and the target signal have a waveform close to the whitened signal, such as the signal
extraction device 600-5 By using the configuration, the amount of computation can be
significantly reduced, and the same effect as that of the signal extraction device 600-1 can be
obtained.
[0039]
In the configuration of the signal extraction device 600-1 described in the first embodiment,
when the power of the noise signal Nb (p) is larger than the power of the target signal Mb (p), the
whitening filter unit 100b affects the noise signal. As a result, the noise signal and the
observation signal can not be appropriately whitened, which causes a problem that the
dereverberation performance of the target signal and the noise suppression performance are
degraded. Therefore, when calculating the whitening filter coefficients (step S6 in FIG. 3), the
signal extraction device 600-6 according to the second embodiment uses the observation signal
and the noise signal instead of using only the observation signal. , Whitening filter coefficients.
The whitening filter coefficients thus obtained are more accurate than the whitening filter
coefficients obtained in the first embodiment in whitening the target signal.
[0040]
The functional structural example of the signal extraction apparatus 600-6 of Example 2 is
shown in FIG. Signal extraction apparatus 600-6 is different from signal extraction apparatus
600-1 in that observation signal Rb (p) and noise signal Nb (p) from noise section detection unit
120 are input to the whitening filter coefficient calculation unit. It is different. The reference
number of this whitening filter coefficient calculation unit is 112. The whitening filter coefficient
calculation unit 112 uses the average spectrum VR (k) of the observed signal Rb (p) and the
average spectrum VN (k) of the noise signal Nb (p) to calculate the average spectrum of the
target signal Mb (p). Estimate VM1 (k). Specifically, it is obtained by the following equation (16).
VM1 (k) = VR (k) −λ2VN (k) (16) Here, λ2 is a correction coefficient, and 0 <λ2 <1. Although
VN (k) is multiplied by the correction coefficient in Equation (16), VR (k) may be multiplied. Also,
both VN (k) and VR (k) may be multiplied. Then, the whitening filter coefficient wMb (p) of the
target signal can be obtained by calculating V1 (k) in equation (5) as VM1 (k) and using equation
(6). Using the noise signal N'b (p) whitened with this wMb (p) and the observation signal R'b (p),
04-05-2019
13
the noise correlation function calculation unit 130 and the observation correlation function
calculation unit 4 measure the noise correlation function and the observation correlation Find the
function.
[0041]
The observation signal Rb (p) includes the noise signal Nb (p) and the target signal Mb (p).
Convolving the whitening filter coefficient wMb of the target signal into the observation signal
Rb (p) means that the target signal Mb (p) in the observation signal can be appropriately
whitened, but the noise signal Nb in the observation signal About (p), it has not been able to
whiten appropriately. Therefore, the noise signal Nb (p) from the noise section detection unit 120
is also whitened with wMb, and by using the whitened observation signal and noise signal, an
appropriate weighted inverse filter can be obtained.
[0042]
In the signal extraction device 600-1 of the first embodiment, since the whitening filter is
calculated including the characteristics of the noise signal Nb (p), the whitening of the target
signal has not been achieved. However, by subtracting the average spectrum VN (k) of the noise
signal from the noise section detection unit 120 from the average spectrum VR (k) of the
observation signal by the configuration of the signal extraction device 600-6, a more accurate
target signal Estimate the average spectrum VM (k) of the target signal whitening of the target
signal. As a result, the accuracy of the inverse filter coefficient is improved, and the performance
of noise suppression and dereverberation of the target signal is improved. Therefore, even if the
power of the noise signal Nb (p) is large, noise suppression and reverberation suppression of the
target signal are possible.
[0043]
FIG. 10 shows a functional configuration example of the signal extraction device 600-7 of the
third embodiment. The noise section detection unit 120 in the signal extraction device 600-7
receives noise section information Tb (b = 1,..., B) and noise signal for each channel used to detect
the noise signal Nb (p). It also outputs Nb (p). The noise interval information Tb is, for example,
the time interval t1 to t2 of the noise signal included in the observation signal. The noise interval
information Tb is input to the noise correlation function calculation unit 130, and the noise
04-05-2019
14
signal Nb (p) is input to the whitening filter coefficient calculation unit 112. Then, the processes
of the whitening filter coefficient calculation unit 112 and the whitening filter unit 100b are
performed.
[0044]
Here, the noise signal included in the whitened observation signal R′b (p) from the whitening
filter unit 100 b is also whitened. Also, the time interval of the noise signal included in the
observed signal Rb (p) before being whitened and the time of the noise signal N′b (p) included
in the whitened observed signal R′b (p) It is almost equal to the target section. Therefore, by
using the noise segment T detected by the noise segment detection unit 120, the whitened noise
signal N'b (p) can be detected from the whitened observation signal R'b (p). Although the noise
correlation function calculation unit 130 performs this detection process in the third
embodiment, the present invention is not limited to this. Therefore, it is not necessary to
whitenize the noise signal Nb (p) from the noise zone detection unit 120, and the same effect as
the signal extraction device 600-6 of the second embodiment can be obtained.
[0045]
Further, as a modification of the third embodiment, as in the signal extraction device 600-8
shown in FIG. 11, the whitening filter coefficient calculation unit 112 is replaced with the
whitening filter coefficient calculation unit 110, and only the observation signal Rb (p) The
whitening filter coefficient may be determined using The whitening filter coefficient calculation
unit 110 and the whitening filter unit 100b may be omitted from the signal extraction device
600-8 as in the signal extraction device 600-9 shown in FIG. In this case, the noise correlation
function calculation unit 130 may detect the observation signal using the noise section
information T from the noise section detection unit 120, and obtain the noise correlation
function of the detected observation signal.
[0046]
The figure which shows the function structural example of the conventional signal extraction
apparatus 500. FIG. FIG. 2 is a diagram showing an example of a functional configuration of a
signal extraction device 600-1 according to the first embodiment. The figure which shows the
flow of the main processes of the signal extraction apparatus 600-1. The figure which showed
04-05-2019
15
the detail of Formula (13). FIG. 2 is a diagram showing an example of a functional configuration
of a signal extraction device 600-2 according to the first embodiment. FIG. 2 is a diagram
showing an example of a functional configuration of a signal extraction device 600-3 according
to the first embodiment. FIG. 2 is a diagram showing an example of a functional configuration of
a signal extraction device 600-4 according to the first embodiment. FIG. 2 is a diagram showing
an example of a functional configuration of a signal extraction device 600-5 according to the first
embodiment. FIG. 7 is a diagram showing an example of a functional configuration of a signal
extraction device 600-6 according to a second embodiment. FIG. 16 is a diagram showing an
example of a functional configuration of a signal extraction device 600-7 according to a third
embodiment. FIG. 16 is a diagram showing an example of a functional configuration of a signal
extraction device 600-8 according to a third embodiment. FIG. 16 is a diagram showing an
example of a functional configuration of a signal extraction device 600-9 according to a third
embodiment.
04-05-2019
16
Документ
Категория
Без категории
Просмотров
0
Размер файла
28 Кб
Теги
jp2009037032
1/--страниц
Пожаловаться на содержимое документа