close

Вход

Забыли?

вход по аккаунту

?

DESCRIPTION JP2000338990

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 JP2000338990
[0001]
BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an
audio signal processing apparatus, an audio signal processing method, and an audio signal
processing program for extracting an audio signal having a specific directivity from a plurality of
statistically independent audio signals. The present invention relates to a computer readable
recording medium.
[0002]
2. Description of the Related Art FIG. 8 is a diagram showing a conventional audio signal
processing apparatus disclosed in Japanese Patent Application Laid-Open No. 5-227589 "Audio
Signal Processing Apparatus". In the figure, 1a, 1b and 1c are sound sources present at positions
separated from each other. Reference numerals 2a, 2b and 2c denote nondirectional
microphones provided at observation points at distances La, Lb and Lc or propagation times Ta,
Tb and Tc from the sound source 1a, and collect voices from the sound sources 1a, 1b and 1c.
Then, the observation signals 3a, 3b and 3c are output respectively.
[0003]
Further, in FIG. 8, reference numeral 11 denotes a delay correction unit, which corrects and
operates so that one of the sound sources 1a, 1b, 1c, for example, observation signals 3a, 3b, 3c
10-04-2019
1
from the sound source 1a has the same phase. A plurality of delay control units 11a, 11b and
11c are provided. Reference numeral 12 denotes a correlation unit, and when the observation
signals 3a, 3b, 3c from the sound sources 1a, 1b, 1c are input through the delay correction unit
11, the correlation of the delayed observation signals 3a, 3b, 3c is calculated. Then, the target
extraction signal 9 is taken out.
[0004]
Next, the operation will be described. For example, the case where the signal from the sound
source 1a is extracted as the target extraction signal 9 will be described. In this case, with respect
to the observation signals 3a, 3b, 3c from the sound source 1a, the outputs have the same phase
in advance in consideration of the propagation times Ta, Tb, Tc from the sound source 1a to the
respective microphones 2a, 2b, 2c. The delay amounts are set by the delay control units 11a,
11b, and 11c, respectively.
[0005]
Next, when the observation signals 3a, 3b, 3c from the sound sources 1a, 1b, 1c are input to the
delay correction unit 11, the signals from the sound source 1a are identical by phase control of
the delay control units 11a, 11b, 11c. Although the signals are output in phase, the signals from
the sound sources 1 b and 1 c are not output in the same phase, and are output in different
phases.
[0006]
When the correlation unit 12 receives and synthesizes the observation signals 3a, 3b and 3c
from the delay correction unit 11, the signals from the sound source 1a are emphasized because
they are in the same phase, but the signals from the sound sources 1b and 1c are emphasized.
The signals are suppressed because they are in different phases.
The correlation unit 12 extracts the emphasized signal from the sound source 1 a as the target
extraction signal 9.
[0007]
10-04-2019
2
Even when the signal from the sound source 1b or the sound source 1c is extracted as the target
extraction signal 9, the delay control units 11a, 11b, and 11c control so that the output of the
signal from the sound source 1b or the sound source 1c has the same phase. By doing this, it is
possible to extract the signal from the sound source 1 b or the sound source 1 c. As described
above, by controlling the delay amount of the audio signal from the plurality of microphones, a
specific signal can be extracted.
[0008]
Since the conventional audio signal processing apparatus is configured as described above, when
extracting each of the sound sources 1a, 1b and 1c, each delay control unit 11a of the delay
correction unit 11 should be There is a problem that it is necessary to set the delay amount by
11b and 11c by an instruction from the outside. Further, once set, the delay amount is fixed, and
when extracting signals from other sound sources, there is a problem that the delay amount has
to be reset by an instruction from the outside.
[0009]
The present invention has been made to solve the above problems, and an audio signal
processing apparatus, an audio signal processing method, and an audio signal processing method
capable of autonomously and dynamically controlling the amount of delay when extracting each
sound source An object of the present invention is to obtain a computer readable recording
medium in which an audio signal processing program is recorded.
[0010]
A speech signal processing apparatus according to the present invention comprises a plurality of
observations including a target extraction signal from a plurality of microphones for collecting
speech from a plurality of statistically independent sound sources. A signal is input, and a
predetermined delay amount is given to a plurality of second observation signals other than the
input first observation signal based on the parameter control signal indicating the assumed delay
amount, and the first The observation signal and the second observation signal to which the
predetermined delay amount has been given are superimposed and synthesized, and an
emphasized portion in the synthesized signal is output as an extraction signal, and the extraction
is performed from the first observation signal. A signal extraction unit that outputs a residual
signal obtained by subtracting a signal, and the extracted signal and the residual signal are input,
and the independence between the extracted signal and the residual signal is determined.
Calculate the amount of information, While sequentially changing the assumed delay amount
10-04-2019
3
based on the calculated mutual information amount, the parameter control signal is output to the
signal extraction unit, and the extraction signal from the signal extraction unit and the residual
signal are input, The mutual information amount is optimum, wherein the minimum point of the
mutual information amount is searched by calculating the mutual information amount, and the
extraction signal corresponding to the minimum point of the mutual information amount
searched is output as the extraction signal of the object And a conversion unit.
[0011]
In the audio signal processing device according to the present invention, the signal extraction
unit inputs a plurality of observation signals from the plurality of microphones arranged at equal
intervals along a straight line, and performs parameter control on each of the plurality of second
observation signals. Based on the signal, a delay amount different by an integral multiple of the
delay amount assumed as the predetermined delay amount is given.
[0012]
In the audio signal processing device according to the present invention, the signal extraction
unit transmits, from the plurality of microphones, the first observation signal and the plurality of
second observation signals delayed from the first observation signal by each input delay amount.
In each of the plurality of second observation signals, the product of each input delay amount in
the plurality of second observation signals and the assumed delay amount based on the
parameter control signal is given as a predetermined delay amount. is there.
[0013]
In the audio signal processing device according to the present invention, the mutual information
amount optimization unit assumes three delay amounts sequentially separated by a
predetermined interval, and uses the assumed three delay amounts as a parameter control signal
to the signal extraction unit. The extracted signal and the residual signal are output, each mutual
information amount corresponding to the assumed three delay amounts is calculated, and among
the three delay amounts, the mutual information corresponding to the central delay amount In
the case where the quantity takes the minimum value, it is assumed that three delay amounts are
sequentially separated by an interval smaller than the predetermined interval, and three delay
quantities sequentially separated by an interval smaller than the predetermined interval are
assumed. It outputs to the signal extraction unit as a parameter control signal, inputs the
extraction signal and the residual signal, and calculates the mutual information amounts
corresponding to the delay amounts at three points sequentially separated by an interval
narrower than the assumed predetermined interval. The above mutual information amount by It
is intended to explore the minimum point.
10-04-2019
4
[0014]
The audio signal processing method according to the present invention comprises a first step of
inputting a plurality of observation signals including target extraction signals from a plurality of
microphones that collect speech from a plurality of statistically independent sound sources; A
second step of giving a predetermined delay amount to a plurality of second observation signals
other than the input first observation signal based on a parameter control signal indicating an
assumed delay amount; A third step of superimposing and synthesizing the observation signal
and the second observation signal given the predetermined delay amount, and a fourth step of
outputting an emphasized portion of the synthesized signal as an extraction signal; A fifth step of
outputting a residual signal obtained by subtracting the extraction signal from the first
observation signal, and inputting the extraction signal and the residual signal to determine the
independence between the extraction signal and the residual signal , Assuming the above The
parameter control signal is output while sequentially changing the assumed delay amount based
on the calculated sixth step of calculating the mutual information amount corresponding to the
delay amount and the calculated mutual information amount, and from the second step The sixth
step is repeated to calculate the mutual information amount, thereby the seventh step of
searching the local minimum point of the mutual information amount, and the above
corresponding to the local minimum point of the mutual information amount searched And an
eighth step of outputting the extraction signal as the target extraction signal.
[0015]
In the audio signal processing method according to the present invention, as a first step, a
plurality of microphones arranged at equal intervals on a straight line are used to collect voices
from a plurality of statistically independent sound sources. A plurality of observation signals
including an extraction signal are input, and as a second step, based on a parameter control
signal indicating an assumed delay amount, for a plurality of second observation signals other
than the input first observation signal Thus, as the predetermined delay amount, delay amounts
different by integer multiples are given.
[0016]
In the audio signal processing method according to the present invention, as a first step, a first
observation signal including a target extraction signal from a plurality of microphones that
collects speech from a plurality of statistically independent sound sources. And a plurality of
second observation signals delayed by each input delay amount from the first observation signal,
and as the second step, the plurality of second observation signals are input to each of the
plurality of second observation signals. The product of each input delay amount in the
observation signal of (1) and the assumed delay amount based on the parameter control signal is
given as a predetermined delay amount.
10-04-2019
5
[0017]
A computer-readable recording medium recording an audio signal processing program according
to the present invention is a plurality of observations including a target extraction signal from a
plurality of microphones collecting speech from a plurality of statistically independent sound
sources. A signal is input, and a predetermined delay amount is given to a plurality of second
observation signals other than the input first observation signal based on the parameter control
signal indicating the assumed delay amount, and the first The observation signal and the second
observation signal to which the predetermined delay amount is given are superimposed and
synthesized, and the extraction signal of the highlighted portion in the synthesized signal and the
extraction signal from the first observation signal are subtracted. An audio signal processing
program that performs processing by inputting the residual signal, which is input, the extraction
signal and the residual signal, and determining the independence of the extraction signal and the
residual signal Outputting the parameter control signal while sequentially changing the assumed
delay amount based on the calculated first step of calculating the mutual information
corresponding to the amount and the calculated mutual information, and the extraction signal
and the residual signal Is repeatedly input to calculate the mutual information amount, and the
second step of searching for the minimum point of the mutual information amount, and the
extraction signal corresponding to the minimum point of the mutual information amount
searched for; And the third step of outputting as the target extraction signal.
[0018]
BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, an embodiment of the present
invention will be described.
Embodiment 1
FIG. 1 is a block diagram showing the configuration of an audio signal processing apparatus
according to a first embodiment of the present invention, in which 1a, 1b, 1c,... It is a sound
source that is totally independent.
.. 2v are non-directional microphones provided at observation points of a predetermined distance
from the sound sources 1a, 1b, 1c,. The sound from the sound sources 1a, 1b, 1c,..., 1u is
collected, and the observation signals 3a, 3b,.
10-04-2019
6
[0019]
Further, in FIG. 1, reference numeral 4 denotes a signal extraction unit, which receives
observation signals 3a, 3b,..., 3v, and receives observation signals 3b,. ... A predetermined delay
amount is given to 3 v, and the observation signal 3 a and the observation signals 3 b to which
each delay amount is given are superimposed and synthesized and emphasized in the
synthesized signal While outputting the part as the extraction signal 5, it outputs the residual
signal 6 which subtracted the extraction signal 5 from the observation signal 3a which is the
input original signal.
[0020]
Further, in FIG. 1, reference numeral 7 denotes a mutual information optimization unit, which
receives the extraction signal 5 and the residual signal 6, and determines the independence of
the extraction signal 5 and the residual signal 6, corresponding to the assumed delay amount d.
The parameter control signal 8 is output to the signal extraction unit 4 while the mutual
information amount MI (d) is calculated and the assumed delay amount d is sequentially changed
based on the calculated mutual information amount MI (d), and the signal extraction unit By
inputting the extraction signal 5 and the residual signal 6 from 4 and calculating the mutual
information amount MI (d), the local information amount MI (d) is searched for the minimum
point of the mutual information amount MI (d) The extraction signal 5 corresponding to the
minimum point of d) is output as the target extraction signal 9.
[0021]
Here, the mutual information amount MI (d) will be described.
The functions of the observation signals 3a, 3b,..., 3v are x1 (t), x2 (t),.
Here, t is the time of observation.
[0022]
Then, assuming the delay amount d, the function f (t, d) of the extraction signal 5 from the signal
10-04-2019
7
extraction unit 4 and the function g (t, d) of the residual signal 6 can be expressed by the
following equation (1) 2) Define as equation.
f(t,d)=(x1(t)+x2(t+d)+・・・+xv(t+(v−1)d))/v
(1)g(t,d)=x1(t)−f(t,d) (2)
[0023]
Next, according to the definition of the distribution function, the distribution pf (m) of the signal f
with respect to the interval m is expressed as the following equation.
pf (m) = # {t | t∈ {1,. . . , T}, f (t) ∈m} / T Here, the number of elements of the set A is written as
#A, and the observation time t is 1 to T.
[0024]
Similarly, the distribution pg (n) of the signal g and the distribution p (m, n) of the signal pair (f,
g) for the section n are respectively expressed as the following equations.
pg(n)=#{t|t∈{1,...,T},g(t)∈n}/T,p(m,n)=#{t|t
∈{1,...,T},f(t)∈m,g(t)∈n}/T/T
[0025]
The mutual information amount MI (d) of the two signals f and g, which is defined from these
distributions, is expressed by the following equation (3).
MI (d) = ∫∫ p (m, n) (log p (m, n)-log pf (m)-log pg (n)) dm dn (3) where log x is the natural
logarithm of the number x It is.
10-04-2019
8
[0026]
The mutual information amount MI (d) represented by the above equation (3) is a non-negative
amount which does not take a negative amount, and when the mutual information amount MI (d)
indicates a positive amount, two signals f and g It is known that the two signals f and g are
statistically independent of each other when they are not statistically independent of one another
and the mutual information amount MI (d) is zero.
As described above, by searching for the minimum value of the mutual information amount MI
(d), the signal f (extraction signal 5) at the time of the minimum value can be extracted as the
target extraction signal 9.
[0027]
Next, the operation will be described.
FIG. 2 is a flowchart showing the process of the signal extraction unit 4. In step ST1, the signal
extraction unit 4 receives the parameter control signal 8 indicating the assumed delay amount d
from the mutual information amount optimization unit 7 with respect to the observation signals
3b,. Then, the delay amount is set as shown in the above equation (1). That is, the delay amount
is 0 for the observation signal 3a, d for the observation signal 3b, 2d for the observation signal
3c (not shown), and thereafter for the observation signal 3v. The delay amount is set to (v-1) d.
[0028]
The signal extraction unit 4 inputs the observation signals 3a, 3b, ..., 3v from the microphones
2a, 2b, ..., 2v in step ST2, and in step ST3, the observation signal 3b except the observation signal
3a, The delay amount set in step ST1 is given to 3v for delay.
[0029]
In step ST4, the signal extraction unit 4 superimposes the observation signal 3a and the
observation signals 3b,..., 3v given the respective delay amounts.
10-04-2019
9
The superimposed signals are emphasized in amplitude according to the propagation time to the
microphones 2a, 2b, ..., 2v of the respective sound sources 1a, 1b, 1c, ..., 1u and the set delay
amounts. The reduced portion and the reduced amplitude generate a suppressed portion. In step
ST5, the signal extraction unit 4 separates and outputs the portion emphasized in step ST4 as the
extraction signal 5, and outputs the remaining signal 6 obtained by removing the extraction
signal 5 from the observation signal 3a which is the original signal input. Output
[0030]
FIG. 3 is a flowchart showing the process flow of the mutual information amount optimization
unit 7. Further, FIG. 4 is a diagram showing a change in the mutual information amount MI (d)
calculated by the mutual information amount optimization unit 7 when sound from four sound
sources is collected by the three microphones, the horizontal axis Represents the delay amount d,
which is represented by the angle of the microphone with respect to each sound source from the
center microphone. As shown in FIG. 4, the mutual information amount MI (d) changes with the
delay amount d with the minimum points K1, K2, K3, and K4 corresponding to the four sound
sources.
[0031]
The mutual information amount optimization unit 7 calculates the mutual information amounts
MI (d0), MI (d1), MI (d2) at the delay amounts d0, d1, d2 of three points separated by a
predetermined interval D, and calculates According to the calculation result, the mutual
information amounts MI (d0), MI (d1), MI (d2) are calculated sequentially while shifting the delay
amounts d0, d1, d2 of the three points by the interval D, as shown in FIG. The minimum points
K1, K2, K3, and K4 shown in FIG.
[0032]
In step ST11 of FIG. 3, as shown in FIG. 4, the mutual information amount optimization unit 7
sets the intervals D between the minimum value Dmin and the maximum value Dmax of the
parameter range to be searched and the delay amounts d0, d1 and d2. In addition to the
determination, the initial values of the delay amounts d0, d1 and d2 are assumed as follows.
d0 = Dmin-Dd1 = d0 + Dd2 = d1 + D However, the initial values of the delay amounts d0, d1 and
10-04-2019
10
d2 are positions moved by the interval D to the left than the positions shown in FIG.
[0033]
Here, the interval D is determined by the resolution L determined from the setting condition of
the microphone, and the maximum value Dmax and the minimum value Dmin of the delay
amount d are determined from the arrangement of the microphone and the speed of sound.
[0034]
In step ST12 of FIG. 3, the mutual information amount optimization unit 7 moves the delay
amounts d0, d1, and d2 by the interval D.
That is, it is assumed that d0 = d0 + Dd1 = d1 + Dd2 = d2 + D. The positions of the delay amounts
d0, d1 and d2 are the positions shown in FIG. 4, and the position of the delay amount d0 is the
position of Dmin.
[0035]
The mutual information amount optimization unit 7 checks whether the delay amount d2
exceeds the maximum value Dmax of the parameter range in step ST13 of FIG. 3. If not, the
mutual information amount MI (d0) in step ST14. , MI (d1), MI (d2).
[0036]
In this process, mutual information amount optimization unit 7 sequentially provides parameter
control signal 8 indicating delay amounts d0, d1 and d2 to signal extraction unit 4, and signal
extraction unit 4 starts step ST1 shown in FIG. The processing up to ST5 is repeated three times,
and mutual information optimization section 7 receives extracted signal 5 and residual signal 6
in delay amounts d0, d1, d2 and determines the independence of extracted signal 5 and residual
signal 6 The mutual information amounts MI (d0), MI (d1), MI (d2) are calculated.
[0037]
In step ST15 of FIG. 3, the mutual information amount optimization unit 7 calculates the mutual
information amounts MI (d0), MI (d1), MI (d2) calculated in step ST14 as MI (d0) ≧ MI (d1) MI.
10-04-2019
11
(D2) It is checked whether ≧ MI (d1).
That is, among the mutual information amounts MI (d0), MI (d1), MI (d2) for the three delay
amounts d0, d1, d2, the mutual information amount MI (d1) corresponding to the central delay
amount d1 is minimum. It is checked whether it is a value, that is, whether the three points (d0,
d1, d2) are convex with respect to the mutual information amount MI (d).
[0038]
FIG. 5 is a diagram showing a state in which the delay amount d is moved to search for the
minimum point K1 of the mutual information amount MI (d).
Then, in step ST15 of FIG. 3, as shown in FIG. 5A, when the mutual information amount MI (d1)
is not the minimum value, the process returns to step ST12 to set the delay amounts d0, d1, d2
to FIG. Respectively, by an interval D, and steps ST13 to ST15 are repeated.
[0039]
In step ST15, as shown in FIG. 5B, when the mutual information amount MI (d1) is the minimum
value, that is, three points (d0, d1, d2) relate to the mutual information amount MI (d) If it is
convex, the local minimum point K1 is present in the segment I in the range of the delay amount
d0 to d2, so the process proceeds to step ST16.
[0040]
In step ST16, as shown in FIG. 5C, the mutual information amount optimization unit 7 calculates
the delay amounts d10, d11, d12, d13 and d14 in the segment I in the range of the delay
amounts d0 to d2 as follows. Let's assume.
d11=(d0+d1)/2d13=(d1+d2)/ 2
[0041]
10-04-2019
12
The mutual information amount optimization unit 7 calculates the mutual delay amounts MI
(d10), MI (d11), MI (MI) for the partial segment I0 in the range of the delay amounts d10 to d12
in the same manner as step ST14. d12) is calculated, and MI (d11) is the minimum value as in
step ST15, that is, three points (d10, d11, d12) of partial segment I0 are convex with respect to
mutual delay amount MI (d) Check if it is. Similarly, it is checked whether three points for partial
segment I in the range of delay amount d11 to d13 and partial segment I2 in the range of delay
amount d12 to d14 are convex with respect to mutual delay amount MI (d) .
[0042]
Then, the mutual information optimization unit 7 divides the partial segment that is convex with
respect to the mutual delay amount MI (d) into partial segments in a narrower range, and
becomes convex with respect to the mutual delay amount MI (d) Check if it is. In the example of
FIG. 5 (c), the partial segment I1 is further divided into partial segments in a narrow range, and it
is checked whether the partial segments in the narrow range are convex with respect to the
mutual delay amount MI (d). The division of this partial segment is performed within the
resolution range determined by the sampling frequency of the observation signals 3a, 3b,..., 3v to
search for the local minimum point K1 of FIG.
[0043]
In step ST17 of FIG. 3, the mutual information amount optimization unit 7 stores the extraction
signal 5 at the minimum point K1 searched in step ST16 as the target extraction signal 9, and
returns to step ST12 to start the next minimum point Search for K2.
[0044]
As described above, as shown in FIG. 4, four minimum points K1, K2, K3, and K4 are searched,
and in step ST13 of FIG. 3, the delay amount d2 exceeds the maximum value Dmax of the
parameter range. In this case, in step 18, the mutual information amount optimization unit 7
outputs the target extraction signal 9 stored in step ST17.
[0045]
The delay amounts d corresponding to the four minimum points K1, K2, K3 and K4 searched as
shown in FIG. 4 correspond to the directions of the respective sound sources, and the extraction
signals 5 corresponding to the respective delay amounts d The function f (t, d) of is extracted by
emphasizing the audio signal from the direction of each sound source.
10-04-2019
13
[0046]
FIG. 6 is a diagram showing an example of an observed signal when four human voices are
collected by three microphones as a sound source, and FIG. 6 (a) is a waveform of a function x1
(t) of the observed signal 3a FIG. 6 (b) shows the waveform of the function x2 (t) of the observed
signal 3b, and FIG. 6 (c) shows the waveform of the function x3 (t) of the observed signal 3c.
[0047]
FIG. 7 shows the waveforms of the original sound of one of the four human voices in FIG. 6 (FIG.
7 (a)) and the results of the search for the local minimum by the mutual information optimization
unit 7; It is a figure which shows the waveform (FIG.7 (b)) of the extraction signal 9 of the
objective corresponding to Fig.7 (a), and it turns out that the human voice of Fig.7 (a) is
extracted.
[0048]
In the above example, as shown in FIG. 1, the microphones 2a, 2b, ..., 2v are arranged at equal
intervals on a straight line, but the microphones 2a, 2b, ..., 2v are arranged at arbitrary positions
Even in this case, the target extraction signal 9 can be extracted.
[0049]
The observation signal 3b from the microphone 2b is input as the input delay amount d2 with
respect to the observation signal 3a from the microphone 2a, and the observation signal 3c from
the microphone 2c not shown is the input delay amount d3. If the observed signal 3v of is input
as the input delay amount dv, the extraction function f (t, d) and the residual function g (t, d)
shown in the following equations (4) and (5) can be used: good.
f(t,d)=(x1(t)+x2(t+d×d2)+・・・+xv(t+d×dv))/v
(4)g(t,d)=x1(t)−f(t,d) (5)
[0050]
That is, the delay amount obtained by the product of the input delay amount d2 and the assumed
delay amount d by the parameter control signal 8 is given to the observation signal 3b without
giving the delay amount to the observation signal 3a, and For the observation signal 3v, a delay
10-04-2019
14
amount determined by the product of the input delay amount dv and the delay amount d
assumed by the parameter control signal 8 is given, and the extraction function f shown in the
above equations (4) and (5) The extracted signal 5 of (t, d) and the residual signal 6 of the
residual function g (t, d) are extracted.
[0051]
In the above embodiment, the processing for searching for the minimum point and extracting the
target extraction signal 9 as shown in FIG. 3 and FIG. 4 is realized by a program for audio signal
processing, and this program is a recording medium Are provided to record.
[0052]
As described above, according to the first embodiment, assuming the delay amount d, the
extracted signal 5 and the residual signal 6 are separated, and the mutual information amount MI
for determining the independence of the extracted signal 5 and the residual signal 6 is
determined. By calculating (d) and searching for the minimum point of the mutual information
amount MI (d), control of the delay amount d is performed autonomously and dynamically to
extract the target extraction signal 9 The effect of being able to
[0053]
Second Embodiment
An application example of the algorithm in the first embodiment will be specifically described.
First, the case of application to a car navigation system mounted on a car will be described.
For example, it is effective when the driver gives a voice instruction to the car navigation system
in an environment where there is noise from a car or noise from the car.
In this case, the voice from the driver and the noise from the car and the noise from the car are a
plurality of signals that are statistically independent, and as shown in the first embodiment,
assuming the delay amount d, the mutual information amount By calculating MI (d), searching
for a local minimum point, extracting an audio signal from the driver of interest, and inputting it
10-04-2019
15
to the car navigation system, it is possible to reliably give the driver's instruction to the car
navigation system Become.
[0054]
In addition to the driver's voice, even when there are voices from a plurality of passengers in the
car, they are a plurality of signals that are statistically independent, as shown in FIG. 5 of the first
embodiment. As described above, it is possible to search for a plurality of minimal points and
take out the voice signals of the driver and the plurality of passengers, and take out only the
voice signal of the driver according to the angle of the driver and the microphone stored
beforehand.
Furthermore, in this case, the mutual information amount MI (d) is calculated assuming only the
delay amount d in the range corresponding to the angle between the driver and the microphone,
and searching for the minimum point is performed more efficiently. Only the driver's voice signal
can be extracted.
[0055]
Next, the algorithm of the first embodiment can be applied to a hands-free phone mounted on a
car.
In Europe, the driver is obliged to use a hands-free phone with a fixed phone, and the use
environment in a car with high noise and noise is the same as that of the above car navigation
system, and only the driver's voice signal is taken out Can be input to the hands-free phone.
[0056]
Furthermore, the algorithm in the above-described Embodiment 1 can be applied to an image
signal.
For example, in an X-ray real-time imaging apparatus, it is possible to capture a plurality of
10-04-2019
16
moving objects moving independently and to extract an image signal of a target moving object
from a plurality of image signals which are statistically independent. It is.
[0057]
Furthermore, the algorithm of the first embodiment described above can be applied to radio
signals such as radar signals.
For example, when a chaff (metal foil piece) is scattered from an aircraft being tracked by a radar
device, a desired radar signal is reflected from the tracked aircraft and a radar signal reflected
from the chaff. It is also possible to extract a radar signal reflected from an airplane.
[0058]
As described above, according to the second embodiment, it is possible to input a target audio
signal to a car navigation system or hands-free phone, and the signal to be handled is not limited
to the audio signal, and Also in an image signal, a radar signal, etc., an effect is obtained that a
target signal can be obtained from a plurality of statistically independent signals.
[0059]
As described above, according to the present invention, a plurality of observation signals
including target extraction signals from a plurality of microphones for collecting speech from a
plurality of statistically independent sound sources are input. And a predetermined delay amount
is given to a plurality of second observation signals other than the input first observation signal
based on the parameter control signal indicating the assumed delay amount, and a
predetermined observation signal The second observation signal given the delay amount of 1 is
superimposed and synthesized, and the emphasized portion in the synthesized signal is output as
an extraction signal, and the residual signal obtained by subtracting the extraction signal from
the first observation signal The mutual information amount calculated by calculating the mutual
information amount corresponding to the assumed delay amount which receives the extraction
signal and the residual signal and outputs the extracted signal and the residual signal to
determine the independence of the extraction signal and the residual signal Sequentially change
the delay amount assumed based on While the parameter control signal is output to the signal
extraction unit, the extraction signal and the residual signal from the signal extraction unit are
input and the mutual information amount is calculated to search for the minimum point of the
mutual information amount By providing an extraction signal corresponding to the minimum
point of the searched mutual information amount as a target extraction signal and providing a
10-04-2019
17
mutual information amount optimization unit, control of the delay amount is performed
autonomously and dynamically. There is an effect that the target extraction signal can be taken
out.
[0060]
According to the present invention, the signal extraction unit inputs a plurality of observation
signals from the plurality of microphones arranged at equal intervals on a straight line, and
based on the parameter control signal, each of the plurality of second observation signals By
giving delay amounts different by integral multiples of the delay amount assumed as the
predetermined delay amount, it is possible to extract a target extraction signal from observation
signals from a plurality of microphones arranged at equal intervals on a straight line. It has the
effect of
[0061]
According to the present invention, the signal extraction unit receives the first observation signal
and the plurality of second observation signals delayed from the first observation signal by each
input delay amount from the plurality of microphones, and Arbitrary positions can be obtained
by giving the product of each input delay amount in the plurality of second observation signals
and the assumed delay amount based on the parameter control signal as a predetermined delay
amount to each of the plurality of second observation signals. Among the observation signals
from a plurality of microphones arranged in the above, there is an effect that a target extraction
signal can be taken out.
[0062]
According to the present invention, a first step of inputting a plurality of observation signals
including target extraction signals from a plurality of microphones collecting speech from a
plurality of statistically independent sound sources; A second step of giving a predetermined
delay amount to a plurality of second observation signals other than the input first observation
signal based on the parameter control signal indicating the amount; a first observation signal and
a predetermined From the third step of superimposing and synthesizing the second observation
signal to which the delay amount is given, the fourth step of outputting the highlighted portion of
the synthesized signal as the extraction signal, and the first observation signal A fifth step of
outputting a residual signal obtained by subtracting the extraction signal, and an amount of
mutual information corresponding to the assumed delay amount which receives the extraction
signal and the residual signal and determines the independence of the extraction signal and the
residual signal. 6th step to calculate Outputting the parameter control signal while sequentially
changing the assumed delay amount based on the calculated mutual information amount and
calculating the mutual information amount by repeating the second step to the sixth step By
10-04-2019
18
providing the seventh step of searching for the minimum point of the mutual information amount
and the eighth step of outputting the extraction signal corresponding to the minimum point of
the searched mutual information amount as the target extraction signal, There is an effect that
control of the delay amount can be performed autonomously and dynamically to extract a target
extraction signal.
10-04-2019
19
Документ
Категория
Без категории
Просмотров
0
Размер файла
31 Кб
Теги
jp2000338990, description
1/--страниц
Пожаловаться на содержимое документа