close

Вход

Забыли?

вход по аккаунту

JP2010181635

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 JP2010181635
PROBLEM TO BE SOLVED: To accurately calculate the mutual transfer function from one of the
two receiving means to the other in a system in which signals emitted from one signal source are
received by two different receiving means via two different paths. To identify. SOLUTION: A
sound S (z) emitted from a sound source 16 as a signal source is received by two microphones A
and B as two receiving means. The path from one of the two microphones A and B to the other
has an acoustic transfer function hab (z), and the acoustic transfer function hab (z) is cascaded
with each other, the FIR adaptive filter 50 and the IIR. It is identified by the adaptive filter 52.
Thus, the acoustic transfer function hab (z) is accurately identified by using the FIR adaptive filter
50 and the IIR adaptive filter 52 cascade-connected to each other as means for carrying out the
identification of the acoustic transfer function hab (z). be able to. [Selected figure] Figure 1
Identification apparatus and identification method
[0001]
The present invention relates to an identification device and identification method, and in
particular, in a system in which signals emitted from one signal source are received by two
different reception means via two different paths, the two reception means The present invention
relates to an identification apparatus and an identification method for identifying mutual transfer
functions from one to another based on signals received by each of the two receiving means.
[0002]
This kind of identification device and identification method is used, for example, in a so-called
04-05-2019
1
microphone array system provided with two microphones as two reception means.
One example is disclosed in Non-Patent Document 1. According to the prior art disclosed in this
non-patent document 1, as shown in FIG. 7, two microphones 12 and 14 are provided at a
distance from each other. Then, each of these two microphones 12 and 14 receives the sound S
(z) (z; delay element in z conversion) emitted from the sound source 16 as a signal source.
[0003]
Here, from the sound source 16 to one of the microphones 12 (hereinafter referred to as A, the
microphone A is referred to as a microphone A). The path to) has an unknown acoustic transfer
function of hsa (z). That is, the sound Xa (z) received by the one microphone A is expressed as the
following equation 1 using this acoustic transfer function hsa (z).
[0004]
[0005]
Similarly, the microphone 16 is referred to as the microphone B by using the code from the
sound source 16 to the other microphone 14 (hereinafter, B).
The path to) also has an unknown acoustic transfer function hsb (z). That is, the sound Xb (z)
received by the other microphone B is expressed as the following equation 2 using this acoustic
transfer function hsb (z).
[0006]
[0007]
The output signal Xaj (j; time) of one of the microphones A is input to a non-recursive, so-called
FIR (Finite Impulse Response) type adaptive filter 18.
04-05-2019
2
The adaptive filter 18 uses the output signal Xaj of the one microphone A as a reference signal,
and uses the output signal Xbj of the other microphone B as a desired response to the acoustic
transfer function hab from the one microphone A to the other microphone B z) identify. That is,
the own transfer function Hab (z) is matched with the acoustic transfer function hab (z).
[0008]
For this purpose, the output signal Yj of the adaptive filter 18 is input to the adder (subtractor)
20. At the same time, the output signal Xbj of the other microphone B is also input to the adder
20. The adder 20 subtracts the output signal Yj of the adaptive filter 18 from the output signal
Xbj of the other microphone B, and the output signal (error signal) Ej of the adder 20 is given to
the adaptive filter 18. The adaptive filter 18 updates its filter coefficients so that the output
signal Ei of the adder 20 becomes as small as possible. Thereby, the identification of the abovementioned acoustic transfer function hab (z) by the adaptive filter 18 is realized.
[0009]
The following simulation was performed on the conventional microphone array system 100
shown in FIG. That is, white noise is substituted as speech S (z). Then, the acoustic transfer
function hsa (z) of the path from the sound source 16 to one of the microphones A is defined by
the difference equation expressed by the following equation 3, and the acoustic transfer of the
path from the sound source 16 to the other microphone B The function hsb (z) is defined by the
difference equation represented by Equation 4.
[0010]
[0011]
[0012]
Note that the coefficient a (m) in the equation 3 and the coefficient b (m) in the equation 4 are
both set based on a normal random number that exponentially attenuates, and the maximum
value M of each item number m is any It is assumed that M = 128.
04-05-2019
3
[0013]
Further, the number of taps of the adaptive filter 18 is made four types of 512, 1024, 2048 and
4096.
Then, as a method of updating the filter coefficients of the adaptive filter 18, a known learning
identification method is adopted.
The step size μ for determining the update degree of the filter coefficient is set to μ = 0.1.
Further, disturbance noise (diffuse noise) is not applied to each of the microphones A and B.
[0014]
Under such conditions, in order to evaluate the identification performance of the adaptive filter
18, the output error Ro is obtained based on the following equation (5). In the equation 5, p is an
integer, and J is a constant time represented by a multiple of the time j. That is, according to the
equation (5), the output error Ro is expressed in decibels by averaging the ratio of the output
signal Ej of the adder 20 to the output signal Xbj of the microphone B every fixed time J.
Therefore, the smaller the output error Ro, the higher the accuracy of identification by the
adaptive filter 18.
[0015]
[0016]
FIG. 8 shows the result of obtaining the output error Ro for each of the tap numbers 512, 1024,
2048 and 4096 described above based on the equation (5).
04-05-2019
4
In FIG. 8, curves with individual signs G512, G1024, G2048 and G4096 indicate output errors Ro
when the number of taps is 512, 1024, 2048 and 4096, respectively. Further, in FIG. 8, the time
on the horizontal axis is expressed in units of blocks, and one block corresponds to the time J
described above. Here, in order to suppress fluctuation of the output error Ro and make it easy to
view, the time J is set to J = 4096.
[0017]
As can be seen from FIG. 8, as the number of taps of the adaptive filter 18 decreases, the output
error Ro converges faster. On the other hand, although it is ideal that the output error Ro
decreases as the number of taps increases, such a result can not be recognized from FIG. That is,
even if the number of taps is increased, the output error Ro is reduced to only an insufficient
level of about -10 [dB]. This means that in a reverberant environment, the FIR type adaptive filter
18 can not sufficiently identify the above-described acoustic transfer function hab. The reason is
presumed to be as follows.
[0018]
That is, when the configuration of FIG. 7 mentioned above is shown clearly, it becomes as shown
in FIG. According to the configuration of FIG. 9, the speech S (z) is input to the so-called unknown
system 22 having an unknown transfer function of hsa (z), and the output signal Xa (z) of the
unknown system 22 is an adaptive filter It is input to 18. Then, the output signal Y (z) of the
adaptive filter 18 is input to the adder 20. The input / output signals Xa (z) and Y (z) of the
adaptive filter 18 in FIG. 9 correspond to the input / output signals Xaj and Yj of the adaptive
filter 18 in FIG. 7, respectively. Y j is expressed using the delay element z.
[0019]
Further, the speech S (z) is input to another unknown system 24 having an unknown transfer
function hsb (z), and the output signal Xb (z) of the unknown system 24 is input to the adder 20.
The output signal Xb (z) of the unknown system 24 input from the unknown system 24 to the
adder 20 in FIG. 9 is the output signal Xbj of the microphone B input from the microphone B to
the adder 20 in FIG. It corresponds to
04-05-2019
5
[0020]
Then, the adder 20 subtracts the output signal Yx (z) of the adaptive filter 18 from the output
signal Xb (z) of the unknown system 24, and based on the output signal E (z) of the adder 20,
The filter coefficients are updated and hence the transfer function Hab (z) is updated. The output
signal E (z) of the adder 20 in FIG. 9 corresponds to the output signal Ej of the adder 20 in FIG.
[0021]
According to the configuration of FIG. 9, a table of the following equation 6 is used between the
transfer function Hab (z) of the adaptive filter 18 and the transfer functions hsa (z) and hsb (z) of
the respective unknown systems 22 and 24. The adaptive filter 18 will provide sufficient
identification performance when the relationship is satisfied.
[0022]
[0023]
In other words, the transfer function hab (z) to be identified by the adaptive filter 18 is expressed
by the following equation 7.
[0024]
[0025]
However, the transfer function hab (z) expressed as a fraction like this number 7 is generally
indivisible and a so-called remainder is generated.
In particular, in an environment where the above-described reverberation occurs, the excess
always occurs.
The transfer function hab (z) causing the remainder in this manner is sufficiently identified by
the adaptive filter 18 of the FIR type, that is, the transfer function hab (z) and the transfer
function Hab (z) of the adaptive filter 18 are matched with each other. This is basically
04-05-2019
6
impossible even if the number of taps of the adaptive filter 18 is increased.
Therefore, in the prior art, sufficient identification performance can not be obtained as described
above.
[0026]
Hirofumi Nakano, Kensaku Fujii, Michiji Yasuyasu, "A Study on Noise Reduction Method with
Two Microphone System", IEICE Technical Report, The Institute of Electronics, Information and
Communication Engineers, January 2007, vol. 106, no. 483, EA 2006-100, p. 15-20
[0027]
That is, the problem to be solved by the present invention is that, in a system in which signals
emitted from one signal source are received by two different receiving means via two different
paths, the prior art does not. It means that the mutual transfer function from one of the receiving
means to the other can not be sufficiently (correctly) identified.
In particular, in an environment with reverberation, this problem inevitably arises.
[0028]
Therefore, an object of the present invention is to provide an identification apparatus and an
identification method capable of sufficiently identifying the mutual transfer function from one of
the two reception means to the other even in a reverberant environment.
[0029]
In order to achieve this object, according to a first aspect of the present invention, as described
above, a system in which a signal emitted from one signal source is received by two different
receiving means via two different paths. On the premise of an identification device for identifying
the mutual transfer function from one of the two receiving means to the other based on the
received signal by each of the two receiving means, and under this premise, the mutual transfer
function is identified An adaptive filter means is provided.
04-05-2019
7
The adaptive filter means includes FIR filter means connected in cascade with each other and socalled recursive so-called IIR (Infinite Impulse Response) filter means.
[0030]
That is, in the first aspect of the present invention, the adaptive filter means for identifying the
mutual transfer function includes the FIR filter means and the IIR filter means connected in
cascade.
Here, it is known that all transfer functions including mutual transfer functions can be
represented by a cascade connection of a non-recursive model and a recursive model. Therefore,
as in the first invention, the mutual transfer function can be sufficiently identified by providing
the adaptive filter means in which the non-recursive FIR filter means and the recursive IIR filter
means are cascaded with each other. it can. Generally speaking, even if a part that can not be
identified by the FIR filter means arises in the mutual transfer function, that part is identified by
the IIR filter means. Therefore, sufficient identification performance can be obtained as the entire
adaptive filter means including the FIR filter means and the IIR filter means even under the
above-mentioned reverberant environment. Moreover, as described later, it was also confirmed
by experiments that the sufficient identification performance can be obtained even with a
relatively small number of taps.
[0031]
In the first aspect of the present invention, the mutual transfer function may be equivalent to the
transfer function of one path divided by the transfer function of the other path. Even in such a
case, according to the first aspect of the present invention, the mutual transfer function can be
sufficiently identified.
[0032]
Also, the order of the cascade connection of the FIR filter means and the IIR filter means is not
particularly limited. For example, when the signal received by one of the receiving means is input
to the FIR filter means, that is, when the FIR filter means is provided in front of the IIR filter
04-05-2019
8
means, the output signal of the FIR filter means is the input of the IIR filter means The signal is
input to the IIR filter means as a signal. On the contrary, when the signal received by one of the
receiving means is input to the IIR filter means, that is, when the IIR filter means is provided at
the front stage of the FIR filter means, the output signal of the IIR filter means is FIR It is input to
the said FIR filter means as an input signal of a filter. In any case, the signal received by the other
receiving means is input to the IIR filter means as a feedback signal to the IIR filter means.
[0033]
Furthermore, as a signal emitted from a signal source, there is a sound signal such as voice. In
this case, each receiving means is constituted by a microphone.
[0034]
A second invention of the present invention is a method invention corresponding to the first
invention, wherein the mutual transfer function is identified by adaptive filter means including
FIR filter means and IIR filter means connected to each other.
[0035]
As described above, according to the present invention, since the adaptive filter means
responsible for identification of the mutual transfer function includes the FIR filter means and
the IIR filter means cascaded together, the mutual transfer function is sufficiently identified. can
do.
In particular, sufficient identification performance can be obtained even in an environment with
reverberation that can not be handled by the above-described prior art.
[0036]
It is a block diagram showing a schematic structure of a microphone array system concerning
one embodiment of the present invention. It is a block diagram which shows the structure of FIG.
1 briefly. It is a block diagram which shows the detailed structure of the FIR adaptive filter in FIG.
It is a block diagram which shows the detailed structure of the IIR adaptive filter in FIG. It is an
04-05-2019
9
illustration figure which shows the simulation result in the embodiment. FIG. 6 is a block diagram
showing an example of configuration different from FIG. 4 of the IIR adaptive filter in FIG. 2; It is
a block diagram which shows schematic structure of the conventional microphone array system.
It is an illustration figure which shows the simulation result in the prior art of FIG. FIG. 8 is a
block diagram briefly showing the configuration of FIG. 7;
[0037]
One embodiment of the present invention will be described by taking a microphone array system
as an example.
[0038]
As shown in FIG. 1, the microphone array system 10 according to this embodiment is an FIR
adaptive filter as FIR filter means connected in cascade with each other, instead of the adaptive
filter 18 in the conventional microphone array system 100 shown in FIG. 50 and an IIR adaptive
filter 52 as IIR filter means, and also uses the output signal Xbj of the microphone B as a
feedback signal to the IIR adaptive filter 52 as described later.
The remaining structure is similar to that shown in FIG. 7 and, therefore, like parts will be
denoted by the same reference characters and detailed description thereof will be omitted.
[0039]
That is, in the configuration of FIG. 1, the output signal Xaj of the microphone A is input to the
FIR adaptive filter 50. Then, the output signal Yfj of the FIR adaptive filter 50 is input to the IIR
adaptive filter 52. Further, the output signal Xbj of the microphone B is input to the IIR adaptive
filter 52 as its feedback signal. Then, the output signal Yij of the IIR adaptive filter 52 is input to
the adder 20. The adder 20 subtracts the output signal Yij of the IIR adaptive filter 52 from the
output signal Xbj of the microphone B. Then, the filter coefficients α j (n) and β j (n) (n; tap
number) of the FIR adaptive filter 50 and the IIR adaptive filter 52 are updated so that the output
signal E j of the adder 20 becomes as small as possible. The respective transfer functions Hf (z)
and Hi (z) are updated.
[0040]
04-05-2019
10
Specifically, the filter coefficient αj (n) of the FIR adaptive filter 50 is updated based on the
following equation 8, that is, the updated filter coefficient αj + 1 (n) is obtained. Then, the filter
coefficient βj (n) of the IIR adaptive filter 52 is updated based on the equation 9, that is, the
updated filter coefficient βj + 1 (n) is obtained. In these equations (8) and (9), μ is a constant
called the above-mentioned step size, and the updating degree of the respective filter coefficients
αj (n) and βj (n) is controlled by this step size μ.
[0041]
[0042]
[0043]
Furthermore, the configuration of FIG. 1 is as shown in FIG.
In FIG. 2 as well as in FIG. 9 described above, the input and output signals Xa (z) and Yf (z) of the
FIR adaptive filter 50 are expressed using a delay element z.
Then, the configuration of the FIR adaptive filter 50 is shown in detail as shown in FIG. 3
following the expression of FIG.
[0044]
That is, as shown in FIG. 3, the FIR adaptive filter 50 includes delay elements 60, 60,... For the
number N of taps, multipliers 62, 62, ... for the number one more than the number N of taps. It is
a so-called direct type provided with adders 64, 64,... For N taps. Then, the filter coefficient α j
(n) is set to each multiplier 62. This filter coefficient α j (n) is updated based on the equation 8,
as described above. Then, the output signal Yf (z) of the adder 64 of the final stage (rightmost in
FIG. 3) is input to the IIR adaptive filter 52.
04-05-2019
11
[0045]
In FIG. 2, the input and output signals Yf (z) and Yi (z) of the IIR adaptive filter 52 are also
represented using a delay element z. Then, the configuration of the IIR adaptive filter 52 is
shown in detail as shown in FIG. 4 according to the expression of FIG.
[0046]
As shown in FIG. 4, the IIR adaptive filter 52 includes adders 70, 70,... For N taps, multipliers 72,
72,. And the elements 74, 74,... However, as described above, the output signal Xb (z) of the
microphone B is used as a feedback signal, and in detail, the delay element 74 (rightmost in FIG.
4) that constitutes the feedback unit is used as the feedback signal. The output signal Xb (z) of
the microphone B is input. By using the output signal Xb (z) of the microphone B as a feedback
signal as described above, the operation of the IIR adaptive filter 52 is stabilized. Then, a filter
coefficient β j (n) is set to each multiplier 72, and this filter coefficient β j (n) is updated based
on the above-mentioned equation (9). Then, the output signal Yi (z) of the adder 70 of the final
stage (rightmost in FIG. 4) is input to the adder 20 in FIG.
[0047]
The following simulation was performed about the microphone array system 10 of this
embodiment comprised in this way. That is, the number N of taps of each of the FIR adaptive
filter 50 and the IIR adaptive filter 52 is set to N = 128. Then, white noise is substituted as
speech S (z), and disturbance noises Na (z) and Nb (z) for the microphones A and B are given
mutually uncorrelated white noise. The power ratio of the white noise as the voice S (z) and the
white noise as the disturbance noise Na (z) and Nb (z) is −40 [dB]. The other conditions are the
same as in the above-described simulation in the prior art. Then, the output error Ro was
obtained based on the above-mentioned equation 5. However, in calculating the output error Ro,
disturbance noise (diffuse noise) was removed from the numerator and denominator of the
equation (5). Further, as a reference, the output error Ro was obtained under the same conditions
in the above-described conventional technology. The results are shown in FIG.
[0048]
04-05-2019
12
In FIG. 5, a curve labeled with G1 indicates a simulation result for the present embodiment, and a
curve labeled with G2 indicates a simulation result for the prior art. As apparent from FIG. 5,
according to the present embodiment (curve G1), the output error Ro is reduced to a sufficient
degree of about -40 [dB]. On the other hand, according to the prior art (curve G2), it is also
reduced to an insufficient level of about -10 [dB]. That is, according to the present embodiment, it
has been proved that the transfer function hab (z) from one microphone A to the other
microphone B can be identified with extremely high accuracy as compared with the prior art.
This also means that according to the present embodiment, sufficient identification performance
can be obtained even in a reverberant environment. Moreover, such high accuracy identification
can be realized with a relatively small number of taps N such as N = 128. Although not described
in detail here, the same result as shown in FIG. 5 is obtained also when human's voice is used as
the voice S (z).
[0049]
As described above, according to the present embodiment, the FIR adaptive filter 50 and the IIR
adaptive filter 52, which are cascade-connected to each other, are used as means for identifying
the transfer function hab (z) from one microphone A to the other microphone B. Therefore, the
transfer function hab (z) can be identified with extremely high accuracy than in the past. In
particular, high-accuracy identification performance can be obtained even in the presence of
reverberation. And this was confirmed also from the experimental result.
[0050]
Such an embodiment is applied to, for example, a so-called speech separation technique for
separating speech Xa (z) received by one microphone A and speech Xb (z) received by the other
microphone B. Can. Of course, application to techniques other than speech separation can also be
expected.
[0051]
Although the FIR adaptive filter 50 has the direct type configuration shown in FIG. 3 in the
present embodiment, the present invention is not limited to this. For example, a configuration
other than the direct type, such as a cascade type, a parallel type, or a lattice type, may be
employed.
04-05-2019
13
[0052]
Similarly, the IIR adaptive filter 52 is not limited to the direct type shown in FIG. 4, and may be a
cascade type, parallel type, or grid type other than the direct type. In particular, if a lattice type
filter as shown in FIG. 6 is adopted as this IIR adaptive filter 52, for example, the amount of
operation for updating the filter coefficient can be reduced, and the adaptive operation is further
stabilized. It is expected that you can.
[0053]
The filter 52 of FIG. 6 will be described in more detail. The filter 52 is a so-called 2-multiplier
structure, that is, an adder 80 provided on the forward side for each tap. An adder 82 provided
on the feedback side, a multiplier 84 for performing multiplication according to the filter
coefficient βj (n) from the feedback side to the forward direction, and a filter coefficient −βj (n)
of the opposite sign to the opposite direction And a delay element 88 provided on the feedback
side. Also in this grid type filter 52, the output signal Xb (z) of the microphone B as a feedback
signal is input to the delay element 88 (rightmost in FIG. 6) that constitutes the feedback unit.
Furthermore, the filter coefficient βj (n) is updated based on the above-mentioned equation 9,
and complemented with this, the filter coefficient −βj (n) of the opposite sign is updated. The
output signal Xb '(z) of the adder 82 of the final stage (the leftmost in FIG. 6) on the feedback
side is not particularly used. Also, although not shown in the figure, a one multiplier structure
may be employed instead of this two multiplier structure.
[0054]
Further, in the present embodiment, the output signal Xb (z) (Xbj) of the microphone B is adopted
as the feedback signal to the IIR adaptive filter 52, but the present invention is not limited to this.
For example, as generally known, the output signal Yi (z) (Yij) of the IIR adaptive filter 52 itself
may be adopted as the feedback signal. However, as described above, by employing the output
signal Xb (z) of the microphone B as the feedback signal, the operation of the IIR adaptive filter
52 can be stabilized.
[0055]
04-05-2019
14
Furthermore, the order of the cascade connection of the FIR adaptive filter 50 and the IIR
adaptive filter 52 may be reversed to that described in the present embodiment. That is, the
output signal Xa (z) (Xaj) of the microphone A may be input to the IIR adaptive filter 52 by
locating the IIR adaptive filter 52 at the front stage of the FIR adaptive filter 50. In this case, the
output signal Yi (z) (Yij) of the IIR adaptive filter 52 is input to the FIR adaptive filter 50. Then,
the output signal Yf (z) (Yfj) of the FIR adaptive filter 50 is input to the adder 20. The output
signal Xb (z) (Xbj) of the microphone B is input to the IIR adaptive filter 52 as its feedback signal.
Alternatively, as described above, the output signal Yi (z) of the IIR adaptive filter 52 itself may
be used as the feedback signal.
[0056]
Then, one or more FIR adaptive filters other than these or one or more IIR adaptive filters may be
further cascade-connected to the cascade connection of the FIR adaptive filter 50 and the IIR
adaptive filter 52. However, basically, the cascade connection of only the FIR adaptive filter 50
and the IIR adaptive filter 52 can sufficiently identify all the transfer functions including the
above-described transfer function hab (z).
[0057]
The FIR adaptive filter 50 and the IIR adaptive filter 52 can be configured by, for example, a CPU
(Central Process Unit) or a DSP (Digital Signal Processor). The adder 20 is also similar.
[0058]
And although this embodiment explained the case where the present invention was applied to the
microphone array system 10 which handles sound S (z), it does not restrict to this. For example,
the present invention can be applied to a system that handles signals such as radio waves and
various electric signals.
[0059]
In short, two receiving means are provided, and an unknown transfer function exists between
04-05-2019
15
each of the two receiving means and the signal source, in other words, the signal emitted from
the signal source is received directly by the respective receiving means. In such a configuration,
the present invention can be applied to an application of identifying a transfer function from one
receiving means to the other receiving means.
[0060]
Reference Signs List 10 microphone array system 12, 14 microphones 16 sound source 20 adder
50 FIR adaptive filter 52 IIR adaptive filter
04-05-2019
16
1/--страниц
Пожаловаться на содержимое документа