close

Вход

Забыли?

вход по аккаунту

?

JP2005286712

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 JP2005286712
PROBLEM TO BE SOLVED: To provide a sound collection device capable of matching the
variations of the characteristics of microphones with high accuracy. SOLUTION: A signal sent
from an m-th receiving circuit DTm to an array signal processing unit 2 through a filter circuit
FTm is multiplied by a characteristic G (f) of a microphone MCm and a filter characteristic H (f) of
a filter circuit FTm. It is done. Since the characteristic H (f) is the product of the characteristics of
all the microphones other than the microphone MCm, G (f) и H (f) is the characteristic H (f)
obtained by multiplying the characteristics of all the microphones equal. Similar results are
obtained for all the microphones MC1 to MCN, so all the characteristics applied to the signal
output from the filter circuit are H (f), and all the characteristics applied to the signal are
corrected to the same characteristics. Be done. [Selected figure] Figure 1
Sound pickup device
[0001]
The present invention relates to a sound collection device having a microphone array, and more
particularly to a sound collection device capable of correcting variations in characteristics of
microphones included in the microphone array.
[0002]
As a technology to extract specific voices under various sound conditions, multiple microphones
are installed in space, and directivity is formed spatially using voice signals obtained from each
microphone Then, a technology (microphone array technology) for separating a target voice has
been proposed.
08-05-2019
1
[0003]
Microphone arrays are widely used in devices that receive audio signals.
Examples of devices that use the microphone array include, for example, a mobile phone, a video
camera, a digital still camera, a voice recorder, and a conference sound collection system.
[0004]
A usage example of the microphone array will be specifically described.
In the case of a voice recorder, the microphone array implements a noise suppression function
that emphasizes only the voice arriving from the direction of the speaker and suppresses the
reception of ambient noise. As another example, in the case of a mobile phone, the microphone
array provides a hands-free call feature that suppresses the reception of ambient noise when the
speaker speaks from a distance from the mobile phone. As yet another example, in the case of a
video camera, the microphone array generates stereophonic signals in the left and right
directions, and uses them as stereo signals to record stereophonic sound. To achieve.
[0005]
For example, an ECM (Electret Condenser Microphone) is used as a microphone constituting the
microphone array. The ECM is an electronic component that converts sound into an electric
signal and outputs the electric signal, and includes a capacitor configured by a pair of electrode
plates and an electronic circuit connected to the capacitor. The above-mentioned capacitor has a
structure in which one of the electrode plates receives a sound and vibrates, and the distance
between the electrode plates changes with the vibration, so the capacitance changes. In addition,
electret (a substance for which electric charge is persistently held) is used for the electrode plate,
and a voltage is generated between the electrode plates, so voltage change and current change
are caused along with capacity change. The resulting sound is converted to an electrical signal.
The above-described electronic circuit is a circuit that outputs a signal by amplifying the
obtained electric signal and adjusting the impedance, and is configured by a transistor, a
capacitor, a resistor, and the like. Also, some ECMs incorporate an A / D converter and digitize
08-05-2019
2
and output an electrical signal.
[0006]
A microphone is not limited to the ECM, but generally has an amplitude-frequency characteristic
and a phase-frequency characteristic unique to the microphone, and makes specific changes in
the amplitude and phase of each frequency component of the input signal. Output the given
signal. Such amplitude-frequency characteristics and phase-frequency characteristics are
hereinafter referred to as frequency characteristics or characteristics.
[0007]
The characteristics of the microphones vary due to various factors such as variations in physical
characteristics and electrical characteristics of components, distortions generated in the
manufacturing process, and so it is difficult to match the characteristics of the respective
microphones. If it is going to reduce the variation in the characteristic of each microphone, the
manufacturing cost of the microphones will rise sharply.
[0008]
Such variations in the characteristics of the microphone degrade the processing accuracy of the
microphone array, making it impossible to obtain a desired effect. For example, even if peripheral
noise suppression processing is performed when recording with a voice recorder, peripheral
noise is not sufficiently suppressed, and a phenomenon occurs such that unclear voice is
recorded.
[0009]
For the purpose of solving such a problem, various methods have been proposed for correcting
variations in microphone characteristics. For example, Japanese Patent Laid-Open No. 7-131886
(Patent Document 1) discloses an array microphone that corrects the sensitivity of each
microphone unit to the same characteristic by amplifying or attenuating the output signal of each
microphone unit, and its sensitivity correction device. Be done. As another example, for example,
08-05-2019
3
in JP-A-2002-99297 (Patent Document 2), the phase-frequency characteristics and sensitivityfrequency of each microphone are provided by providing a characteristic correction unit
configured with a filter circuit or the like at the subsequent stage of the microphones. A
microphone device for correcting characteristics is disclosed. JP-A-7-131886 JP-A-2002-99297
[0010]
However, it is not easy to correct the characteristics of the microphone by the above method, and
there are the following problems.
[0011]
First, in Patent Document 1, correction is performed by amplifying or attenuating the output
signal of each microphone. However, when correcting the characteristics of the microphone by
amplifying or attenuating the signal, the characteristics should be matched over all frequencies.
There is a problem that it can not do.
[0012]
Variations in the characteristics of microphones, including ECM, are generally frequency
dependent.
For example, when comparing the characteristics of two ECMs, one ECM has a higher gain in the
low frequency range than the high frequency gain, and the other ECM has a lower frequency gain
than the high frequency gain. Sometimes it is small.
In such a case, if the characteristics of the microphone are corrected only by signal amplification
or attenuation, only gains at certain specific frequencies can be matched. The array microphone
and its sensitivity correction device disclosed in Patent Document 1 do not propose a solution to
such a problem.
[0013]
Next, the case where correction is performed using a filter as described in Patent Document 2
will be described.
08-05-2019
4
[0014]
In order to correct the variation in the characteristics of each microphone by the filter, it is
desirable to use a filter having an inverse characteristic of the characteristics of each
microphone.
The characteristic A being an inverse characteristic of the characteristic B means that a signal
obtained by applying both the characteristic B and the characteristic A to an arbitrary signal
matches the original signal, that is, the characteristic A is a signal generated by the characteristic
B. It means that it is a characteristic that cancels out the change. The filter having the inverse
characteristic of a certain characteristic is hereinafter referred to as the inverse filter of that
characteristic.
[0015]
By inverse filter correction of the characteristics of each microphone, all the signals are returned
to the previous state affected by the characteristics of the microphones, so that all the
characteristics applied to each signal after correction can be matched.
[0016]
Furthermore, the auditory effect of the correction on the audio signal will also be described.
If the speech signal is distorted due to the characteristics, and if the distortion is not too large, it
is not unnatural, but if it is large, it is unnatural. Feel the nature of feeling. Therefore, if the
characteristics of the microphone and the filter that performs correction give distortion to the
signal so large as to make it unnatural to human hearing, even if the characteristics applied to
the respective signals after correction all matched Even if it reproduces, it becomes an unnatural
signal for human hearing. However, since the signal corrected by the inverse filter of the
microphone characteristic is an audio signal before being subjected to distortion due to the
microphone characteristic, it is an ideal signal with no sense of incongruity for human hearing.
That is, in order to obtain a signal that does not make the human sense of incongruity, it is
necessary to perform correction using a reverse filter or a filter having characteristics close to
the reverse filter.
08-05-2019
5
[0017]
However, the inverse filter has the fundamental problem that stability is not guaranteed. A filter
being stable means that the output signal of the filter does not diverge with respect to any input
signal of finite amplitude, and the inverse filter whose stability is not guaranteed is that the filter
The output may diverge, resulting in an impracticable filter. In addition, even when the condition
for the inverse filter to be stable is satisfied, a difficult problem in mounting tends to occur such
as a long filter length or divergence due to a small calculation error. There is also a method of
approximating the inverse characteristic by an FIR filter or the like whose stability is guaranteed,
but since approximation is performed, there is no guarantee that the characteristic applied to
each signal after correction will match exactly. In other words, even when correction is
performed using a filter, it is not easy to determine a filter that performs correction with a stable
process and with less variation in the characteristics of the microphones with low precision. The
microphone device disclosed in Patent Document 2 does not propose a specific solution to such a
problem.
[0018]
The condition for the filter to be stable is that the sum of absolute values (the integral of the
absolute values in the case of a continuous signal) of the impulse response of the filter (the
output signal when the impulse signal is input to the filter) becomes a finite value It is generally
known to be the case. Furthermore, in particular, when the filter is a FIR (Finite Impulse
Response) filter, the impulse response is completed in a finite time, so that the filter is always
stable. In addition, when the filter is an IIR (Infinite Impulse Response) filter, the condition for
becoming a stable filter is that all the poles (the denominator of the transfer function The point
where it becomes zero is included in the unit circle on the z plane. Also, in the case where the
filter is an analog filter, the condition for becoming a stable filter is that all the poles are included
in the negative region on the s plane when the characteristics are expressed on the s plane of the
Laplace transform. is there.
[0019]
In summary, the present invention is a sound collection device, wherein N is an integer greater
than or equal to 2 and each includes N sound receiving elements for receiving an audio signal,
converts the received audio signal into an electrical signal, and outputs the electrical signal. A
correction circuit that performs first filtering on each of the outputs of the N reception circuits so
as to correct variations in the frequency characteristics of the N reception circuits and the sound
reception elements included in the N reception circuits. And the filter characteristics in the first
08-05-2019
6
filtering process are obtained by combining the characteristics of N-1 sound receiving elements
among N sound receiving elements excluding one sound receiving element corresponding to self.
The signal processing circuit further includes a signal processing circuit that receives the output
of the correction circuit and outputs a result of performing predetermined processing.
[0020]
Preferably, the filter characteristic of the first filtering performed on the output of the mth
receiver circuit among the N receiver circuits is a characteristic Hm (f) according to the following
equation (1): Hm (f) Gm-1 (f) Gm + 1 (f) GN (f) where m represents an integer of 1 or more and N
or less, and f represents a frequency, Gm (f) represents the frequency characteristic of the sound
receiving element included in the m-th receiving circuit.
[0021]
More preferably, the correction circuit includes N filter circuits that receive the outputs of the N
receiving circuits and perform the first filtering.
[0022]
More preferably, the correction circuit further performs a second filtering process in addition to
the first filtering process, and the filter characteristic in the second filtering process is a
characteristic Hall (f) represented by the following equation (2). A characteristic to convert to a
flat characteristic in the frequency domain Hall (f) = G1 (f) .G2 (f)... GN-1 (f) .GN (f) (2).
[0023]
More preferably, the correction circuit includes N filter circuits that respectively receive the
outputs of the N reception circuits and perform the first filtering process and the second filtering
process.
[0024]
More preferably, each of the N filter circuits has a correction filter that performs a first filtering
process and a flattening filter that performs a second filtering process.
[0025]
More preferably, the second filtering process is a stable filtering process.
08-05-2019
7
[0026]
More preferably, the sound collection device further includes a flattening circuit that receives the
output of the signal processing circuit and performs a second filtering process, and the filter
characteristic in the second filtering process is represented by the following equation (3)
Characteristic Hall (f) is a characteristic that converts flat characteristics in the frequency
domain: Hall (f) = G1 (f) и G2 (f) ииииии GN-1 (f) и GN (f) (3) ).
[0027]
More preferably, the planarization circuit includes a planarization filter that receives the output
of the signal processing circuit and performs a second filtering process.
[0028]
More preferably, the second filtering process is a stable filtering process.
[0029]
According to the sound collection device of the present invention, it is possible to match the
variations in the characteristics of the microphones with high accuracy.
Further, according to the sound collection device of the present invention, it is possible to obtain
an audio signal with less discomfort to human hearing.
Furthermore, according to the sound collection device of the present invention, it is possible to
stably perform the correction process of the audio signal.
[0030]
Hereinafter, embodiments of the present invention will be described in detail with reference to
the drawings.
In the drawings, the same reference numerals indicate the same or corresponding parts.
08-05-2019
8
[0031]
First Embodiment FIG. 1 is a block diagram showing a configuration of a sound collection device
according to a first embodiment.
[0032]
Referring to FIG. 1, the sound collection device 1 converts N received signals DT1 into electric
signals and outputs N reception circuits DT1 to DTN, and N filters receiving outputs from the
respective reception circuits. And circuits FT1 to FTN.
Here, N is an integer of 2 or more.
[0033]
The sound collection device 1 further includes an array signal processing unit 2 that outputs a
result of performing predetermined signal processing on the outputs of the filter circuits FT1 to
FTN.
The array signal processing unit 2 performs directivity processing such as, for example, a delaysum processing in which the outputs of the filter circuits FT1 to FTN are delayed by a
predetermined time and added.
The signal output from the array signal processing unit 2 is appropriately processed by the
device using the sound collection device 1.
For example, if the apparatus is a voice recorder, the signal output from the array signal
processing unit 2 is stored in a memory.
If the device is a portable telephone, the signal output from the array signal processing unit 2 is
sent to a voice codec that performs compression processing.
08-05-2019
9
[0034]
The receiving circuit DT1 receives the audio signal SIG1, converts the sound pressure into an
electric signal, and outputs a microphone-type electric signal in analog form, and the A / D
conversion unit AD1 converts the output of the microphone MC1 into a digital signal. including.
The microphone MC1 is configured of, for example, an ECM.
The A / D conversion unit AD1 performs band limitation processing, discretization, and
quantization that allows only an analog signal output from the microphone MC1 to pass a signal
of a specific frequency band, and outputs a digital signal.
[0035]
Since receiving circuits DT2 to DTN have the same configuration as receiving circuit DT1, the
description on the configuration of receiving circuits DT2 to DTN will not be repeated.
[0036]
The filter circuits FT1 to FTN are formed of, for example, an FIR filter that terminates an impulse
response in a finite time.
The filter circuit FTm is a filter having a filter coefficient hm [k] expressed by the following
equation (4).
[0037]
hm [k] = g1 [k] * ... * gm-1 [k] * gm + 1 [k] * ... * gN [k] (4) where m represents an integer of 1 or
more and N or less, and gm [k] ] Indicates an impulse response of the m-th microphone MCm,
and * indicates an operation code indicating a convolution operation (convolution).
08-05-2019
10
Note that since the FIR filter is a digital filter, here, a discrete (sampled) representation in the
time domain is used, and k indicates the index of the sampled signal. It is known that distortion
will not occur even if such discretization is performed if the frequency band of the signal is bandlimited to 1/2 or less of the sampling frequency.
[0038]
The filter coefficient hm [k] is obtained by convolution of the impulse responses of all the
microphones except for the impulse response of the m-th microphone MCm, as shown in
equation (4).
[0039]
The FIR filter is a filter that performs processing by convoluting an input signal with filter
coefficients, and its impulse response matches the filter coefficients.
Therefore, hm [k] expressed by equation (4) is also an impulse response of the filter circuit FTm.
In general, since the microphone impulse responses g1 [k] to gN [k] converge with time and end
in finite time, h1 [k] to hN [k] obtained by convoluting them also take finite time. finish.
Therefore, since the impulse response of the filter circuit FTm ends in a finite time, it can be
realized as an FIR filter.
[0040]
The impulse response converted into the frequency domain is frequency characteristics, and the
convolution operation in the time domain corresponds to multiplication in the frequency domain.
Therefore, the filter circuit FTm is a filter having a characteristic Hm (f) represented by the
equation (1) obtained by converting the equation (4) into a relational expression in the frequency
domain.
[0041]
Hm (f) = G1 (f) ... Gm-1 (f) Gm + 1 (f) ... GN (f) where m is an integer of 1 or more and N or less,
and f is a frequency And Gm (f) represents the characteristics of the m-th microphone MCm.
08-05-2019
11
[0042]
The filter characteristic Hm (f) is a characteristic obtained by multiplying the characteristics of all
the microphones except the m-th microphone MCm, as shown in the equation (1).
[0043]
In addition, hall [k] which is a result obtained by convolving impulse response of all the
microphones MC1 to MCN is shown by the following equation (5).
[0044]
hall [k] = g1 [k] * g2 [k] *... * gN-1 [k] * gN [k] (5) Further, the characteristic Hall (which is
obtained by converting hall [k] into the frequency domain f) is shown by the following formula
(2).
[0045]
Hall (f) = G1 (f) .G2 (f)... GN-1 (f) .GN (f) (2) The characteristic Hall (f) is all as shown in equation
(2). This is a characteristic obtained by multiplying the characteristics of the microphones MC1
to MCN.
[0046]
Note that the filter coefficient hm [k] will be described in more detail in order to clarify the
content represented by the equation (4).
[0047]
When the number of reception circuits is two (N = 2), the filter coefficient hm [k] is expressed as
the following equations (6) to (7).
[0048]
When m = 1: h1 [k] = g2 [k] (6) When m = 2: h2 [k] = g1 [k] (7) The number of receiving circuits
is three (N = 3) In this case, the filter coefficient hm [k] is expressed as the following equations
(8) to (10).
[0049]
If m = 1: h1 [k] = g2 [k] * g3 [k] (8) If m = 2: h2 [k] = g1 [k] * g3 [k] (9) m = In the case of 3: h3 [k]
= g1 [k] * g2 [k] (10) When the number of reception circuits is four (N = 4) or more, the filter
08-05-2019
12
coefficient hm [k] is 11) It is represented as (14).
[0050]
In the case of m = 1: h1 [k] = g2 [k] * ... * gN [k] (11) In the case of m = 2: h2 [k] = g1 [k] * g3 [k] *
... * gN [K] (12) When m = N-1: hN-1 [k] = g1 [k] * ... * gN-2 [k] * gN [k] (13) When m = N: hN [k] =
g1 [k] *... * gN-1 [k] (14) That is, in the formula (1), N = 2, 3 and m = 1, 2 when N is 4 or more. ,
N?1, N, as in the other cases, the filter coefficient hm [k] convolves the impulse responses of all
the microphones except the m-th microphone MCm. It shows that it can be obtained.
[0051]
Moreover, in order to clarify the content which Formula (1) shows, the filter characteristic Hm (f)
is demonstrated in more detail.
[0052]
When the number of receiving circuits is two (N = 2), the filter characteristic Hm (f) is expressed
as the following equations (15) to (16).
[0053]
In the case of m = 1: H1 (f) = G2 (f) (15) In the case of m = 2 H2 (f) = G1 (f) (16) The number of
receiving circuits is three (N = 3) In this case, the filter characteristic Hm (f) is expressed as the
following equations (17) to (19).
[0054]
In the case of m = 1: H1 (f) = G2 (f) .G3 (f) (17) In the case of m = 2 H2 (f) = G1 (f) .G3 (f) (18) m =
In the case of 3: H3 (f) = G1 (f) и G2 (f) (19) When the number of reception circuits is four (N = 4)
or more, the filter characteristic Hm (f) is given by 20) to 23).
[0055]
In the case of m = 1: H1 (f) = G2 (f) ... ... GN (f) ... (20) In the case of m = 2: H2 (f) = G1 (f) и G3 (f) ...
... GN (F) In the case of m = N-1: HN-1 (f) = G1 (f)... GN-2 (f). GN (f) (22) In the case of m = N: HN
(f) = G1 (f)... GN-1 (f) (23) That is, in the formula (1), N = 2, 3 and m = 1, 2 when N is 4 or more. ,
N?1, N, as in the other cases, the filter characteristics Hm (f) are obtained by multiplying the
characteristics of all the microphones except the m-th microphone MCm. It shows that it is.
08-05-2019
13
[0056]
Also, in order to clarify the content shown by the equation (5), the hall [k] will be described in
more detail.
[0057]
When the number of reception circuits is two (N = 2), hall [k] is expressed as the following
equation (24).
[0058]
hall [k] = g1 [k] * g2 [k] (24) When the number of reception circuits is three (N = 3), hall (f) is
expressed as the following equation (25) .
[0059]
hall [k] = g1 [k] * g2 [k] * g3 [k] (25) That is, equation (2) is the same as in the other cases,
regardless of whether N = 2 or 3. It shows that the hall [k] is obtained by convoluting the impulse
response of all the microphones MC1 to MCN.
[0060]
Also, in order to clarify the content shown by the equation (2), the characteristic Hall (f) will be
described in more detail.
[0061]
When the number of receiving circuits is two (N = 2), the characteristic Hall (f) is expressed as in
the following equation (26).
[0062]
Hall (f) = G1 (f) и G2 (f) (26) When the number of receiving circuits is three (N = 3), the
characteristic Hall (f) is expressed as in the following equation (27) Be
[0063]
Hall (f) = G1 (f) .G2 (f) .G3 (f) (27) That is, in the case of either N = 2 or 3, as in the other cases,
formula (B) The characteristic Hall (f) is a characteristic obtained by multiplying the
characteristics of all the microphones MC1 to MCN.
08-05-2019
14
[0064]
The filter circuits FT1 to FTN read out the respective filter coefficients from the storage means in
which the filter coefficients are previously written.
The storage means is, for example, a ROM (Read Only Memory), and is, for example, a flash
memory.
In FIG. 1, the ROM 3 is shown as an example of a storage device for the filter circuits FT1 to FTN
to read out the filter coefficients.
Although FIG. 1 shows that the ROM 3 is provided outside the sound collection device 1, the
ROM 3 may be included in the sound collection device 1.
[0065]
Note that to describe another configuration, the filter circuits FT1 to FTN of the sound collection
device 1 may be replaced by FIR filters and filter characteristics H1 (f) to HN (f), provided that
conditions for stable operation can be set. It may be an IIR filter having
The condition under which the IIR filter operates stably is that, as described above, when the
characteristics are expressed on the z plane of the z-transform, all the poles (where the
denominator of the transfer function becomes zero) are unit circles on the z plane It is to be
contained within.
[0066]
Furthermore, as another configuration example, the filter circuits FT1 to FTN of the sound
collection device 1 are not limited to digital filters such as FIR filters, and have similar filter
characteristics as long as conditions for stable operation can be set. It may be an analog filter.
08-05-2019
15
The condition under which the analog filter operates stably is that all the poles are included in
the negative region on the s-plane when the characteristics are expressed on the s-plane of the
Laplace transform, as described above.
[0067]
Furthermore, as another configuration, the sound collection device 1 may include a DSP (Digital
Signal Processor) that performs the same processing as the processing of the filter circuits FT1 to
FTN by software instead of the filter circuits FT1 to FTN.
[0068]
The effects of the sound collection device of FIG. 1 will be described.
The characteristic Gm (f) of the microphone MCm and the filter characteristic Hm (f) of the filter
circuit FTm are applied to the signal sent from the m-th receiving circuit DTm to the array signal
processing unit 2 through the filter circuit FTm.
Since the characteristic of the signal is to multiply the signal by the characteristic in the
frequency domain, the characteristic by which the signal is multiplied is Gm (f) и Hm (f).
Since the characteristic Hm (f) is the product of the characteristics of all the microphones other
than the microphone MCm, Gm (f) и Hm (f) is the characteristic Hall (f) obtained by multiplying
the characteristics of all the microphones equal.
Similar results are obtained for all the microphones MC1 to MCN, so all the characteristics
applied to the signal output from the filter circuit are Hall (f), and all the characteristics applied
to the signal are corrected to the same characteristics. Be done.
[0069]
Also, as described above, since the impulse response of the microphone generally ends in a finite
time, and the impulse responses h1 [k] to hN [k] obtained by convolving them also complete in a
finite time, the filter circuit FT1 ... FTN can be realized by an FIR filter.
08-05-2019
16
[0070]
FIR filters generally operate stably.
Therefore, the filter circuits FT1 to FTN are also realized as stable filters.
[0071]
The same effect can be obtained when the filter circuits FT1 to FTN are formed of an IIR filter or
an analog filter instead of the FIR filter.
That is, as long as the filter characteristics of the IIR filter or the analog filter are H1 (f) to HN (f),
the characteristics applied to the signal output from the filter circuit are all Hall (the same as the
FIR filter). Since f), all the characteristics applied to the signal are corrected to the same
characteristics.
Also, since the impulse responses h1 [k] to hN [k] corresponding to the filter characteristics H1
(f) to HN (f) are completed in a finite time, the IIR filter or analog filter having these filter
characteristics is also stable. Can be realized as a filter.
[0072]
FIG. 2 is a flowchart for explaining a method of manufacturing the sound collection device 1 of
FIG.
In particular, FIG. 2 describes the process of generating the filter coefficients h1 [k] to hN [k] that
are important in the manufacture of the filter circuits FT1 to FTN.
[0073]
08-05-2019
17
Referring to FIG. 2, first, when the process is started in step S0, the impulse response of
microphones MC1 to MCN is measured in S1 by, for example, input of a test signal.
[0074]
Subsequently, in step S2, filter coefficients h1 [k] to hN [k] are obtained according to equation
(4), for example, by software processing in a computer.
[0075]
Subsequently, in step S3, the filter coefficients h1 [k] to hN [k] are stored for all the filter circuits.
The filter coefficients h1 [k] to hN [k] are written, for example, in the ROM 3 of FIG.
When the writing is completed, the process ends in step S4.
[0076]
In FIG. 2, for convenience of explanation, only the processing of the m-th microphone MCm and
the filter circuit FTm is shown in all steps S1 to S3, but all the microphones and filter circuits are
shown in each step for all the microphones in step S1 to S3. Processing is performed.
[0077]
FIG. 3 is a diagram for explaining the characteristics of each part of the sound collection device 1.
In FIG. 3, the case of N = 3 in the sound collection device 1 of FIG. 1 will be described.
Each portion of the sound collection device 1 exhibits the same characteristics as the
characteristics shown in FIG. 3 even in the case other than N = 3.
08-05-2019
18
[0078]
Referring to FIG. 3, a graph GR1 shows a characteristic G1 (f) of the microphone MC1.
Similar to the graph GR1, the graph GR2 shows the characteristic G2 (f) of the microphone MC2,
and the graph GR3 shows the characteristic G3 (f) of the microphone MC3.
[0079]
As shown in graphs GR1 to GR3, the characteristics are different for each of the microphones
MC1 to MC3.
[0080]
The characteristic H1 (f) of the filter circuit FT1 is shown in the graph GR4.
Similar to the graph GR4, the graph GR5 shows the characteristic H2 (f) of the filter circuit FT2,
and the graph GR6 shows the characteristic H3 (f) of the filter circuit FT3.
[0081]
The characteristic H1 (f) is a characteristic that satisfies H1 (f) = G2 (f) и G3 (f).
Similarly, the characteristic H1 (f) is a characteristic satisfying H2 (f) = G1 (f) и G3 (f), and the
characteristic H3 (f) is H3 (f) = G1 (f) и G2 (f) It is a characteristic that meets.
[0082]
Graphs GR7 to GR9 show characteristics obtained by combining the characteristics of the
microphone and the filter circuit.
08-05-2019
19
That is, a characteristic G1 (f) и H1 (f) obtained by combining the characteristics of the
microphone MC1 and the filter circuit FT1 is shown in the graph GR7, and a characteristic G2 (f)
obtained by combining the characteristics of the microphone MC2 and the filter circuit FT2 is
shown in the graph GR8. H2 (f) is shown, and a graph GR9 shows characteristics G3 (f) и H3 (f)
obtained by combining the characteristics of the microphone MC3 and the filter circuit FT3.
The characteristics G1 (f) и H1 (f), G2 (f) и H2 (f), G3 (f) и H3 (f) all coincide with the characteristic
Hall (f), so the graphs GR7 to GR9 are shown. As such, all have the same characteristics.
[0083]
As described above, even when the characteristics of the microphones MC1 to MC3 are different
from each other, the filter processing of the filter circuits FT1 to FT3 is performed before the
signals input to the microphones MC1 to MC3 are input to the array signal processing unit 2. Is
corrected so as to apply the same characteristic.
[0084]
Second Embodiment The sound collection device of the second embodiment performs correction
processing on the audio signal so that the human hearing is less uncomfortable than the sound
collection device of the first embodiment.
[0085]
FIG. 4 is a block diagram showing the configuration of a sound collection device according to the
second embodiment.
[0086]
1 and 4, sound collection device 1A differs from sound collection device 1 of FIG. 1 in that it
includes filter circuits FTA1 to FTAN instead of filter circuits FT1 to FTN, and the other parts are
collected The device 1A is similar to the sound collection device 1 of FIG.
[0087]
The filter circuit FTA1 includes a correction filter CF1 having a filter coefficient similar to that of
the filter circuit FT1, and a flattening filter SF1 for converting the characteristic Hall (f)
08-05-2019
20
represented by the equation (2) into a flat characteristic in the frequency domain.
An example of a method of determining a flattening filter that converts the characteristic Hall (f)
into a flat characteristic in the frequency domain will be described in detail below.
[0088]
The flattening filter SF1 is formed of, for example, a first-order FIR filter.
As described above, the FIR filter operates stably, and as the filter order is lower, processing can
be performed with a smaller amount of calculation.
In addition, the flattening filters SF1 to SFN have the same filter coefficients.
Furthermore, the configuration of filter circuits FTA2-FTAN is similar to that of filter circuit
FTA1, and correction filters CF2-CFN included in filter circuits FTA2-FTAN have filter coefficients
similar to those of filter circuits FT2-FTN, respectively. .
Therefore, the description of the configuration of each of filter circuits FTA2-FTAN will not be
repeated hereinafter.
[0089]
Defining the characteristic Hall (f) to be converted into a flat characteristic in the frequency
domain for the filter coefficient of the flattening filter SF1 is, for example, as follows.
[0090]
First, assuming that the flattening filter SF1 is a first-order FIR filter, the relationship between the
time-series signal input to the flattening filter SF1 and the time-series signal output is
represented by the following equation (28).
08-05-2019
21
The frequency characteristic A (f) of the flattening filter SF1 depends on the values of the filter
coefficients a0 and a1, and follows the following equation (29).
[0091]
y [k] = a0x [k] + a1x [k-1] (28) A (f) = a0 + a1exp (-2?if / Fs) (29) where x [k] is input A timeseries signal is shown, and y [k] is an output time-series signal.
Also, as in the equations (4) and (5), k is an index of the sampled signal.
In equation (29), i is an imaginary unit, Fs is a sampling frequency of the A / D conversion unit, f
is a frequency, and the units are both in Hz.
[0092]
The filter coefficients a0 and a1 of the flattening filter SF1 are determined so as to minimize the
evaluation function L represented by the following equation (30).
As a method for minimizing the evaluation function L, for example, an optimization method such
as a gradient method for obtaining a gradient of the evaluation function L and finding an optimal
solution in the direction in which the gradient becomes steep is used.
[0093]
L = ? || Hall (f) и A (f) | ?1 | <2> df (30) In the equation (30), the integration range is an entire
frequency band equal to or less than 1?2 of the sampling frequency.
[0094]
Moreover, Formula (29) is substituted and used for A (f) in Formula (30).
08-05-2019
22
As for the characteristic Hall (f), since hall [k] representing this in the time domain is the
convolution of impulse responses of all the microphones, the impulse response of each
microphone is measured in advance, The hall [k] is calculated by the convolution operation (for
example, calculated in steps S1 and S2 shown in the flowchart of FIG. 2), and the characteristic
Hall (f) converts the calculated hall [k] into the frequency domain It is determined by
[0095]
The completely flat frequency characteristic value is 1 in the entire frequency band.
That is, the evaluation function L of equation (30) is a value obtained by accumulating the square
error between the completely flat frequency characteristic and the characteristic shown by | Hall
(f) и A (f) | over the entire frequency band. is there.
Therefore, minimizing the value of the evaluation function L means making the characteristics of
the function | Hall (f) и A (f) | as flat as possible in the frequency domain, that is, the amplitude
characteristic of Hall (f) It is to find A (f) to be converted into flat characteristics in the frequency
domain.
[0096]
The effects of the sound collection device of the second embodiment will be described.
The signals output from the correction filters CF1 to CFN are aligned in a state in which all the
signals have the same characteristic Hall (f). However, when the characteristic Hall (f) is not flat
but largely distorted in the frequency domain, the signals output from the correction filters CF1
to CFN are audio signals that sound unnatural to humans. In such a case, since the characteristics
of Hall (f) are converted to be closer to flatter characteristics by the flattening filters SF1 to SFN,
the characteristics applied to the respective signals are all aligned. Furthermore, it is corrected to
a sound signal that is good for human hearing.
[0097]
08-05-2019
23
The reason why the error of the amplitude characteristic is used in the evaluation function L of
the equation (30) is that human hearing is more sensitive to the change of the amplitude
characteristic than the change of the phase characteristic. When the amplitude characteristic
changes, the reproduced voice becomes a voice with distorted high- and low-pitched parts, and
human hearing responds sensitively. That is, since the evaluation function L is an evaluation
function that corrects the amplitude characteristic more accurately than the phase characteristic,
correction suitable for human auditory characteristics is performed. It should be noted that when
the phase characteristics change, it may sound reverberant to humans, but if the change is not so
large, it hardly affects human hearing.
[0098]
Also, the above-mentioned flattening processing is performed by a first-order FIR filter. In
general, FIR filters are known to operate stably. In addition, the lower the FIR filter, the smaller
the amount of calculation. Therefore, the stability of the process and the simplicity of the process
can be ensured.
[0099]
The evaluation function L is not limited to the form of equation (30), and may be, for example, a
function represented by the following equations (31) to (33): L = L | Hall (f) и A (f) -1 | <2> df (31)
L = ? | Hall (f) и A (f) | <2> -1 | df (32) L = ?w (f) || Hall (f) и A (f) | ?1 | <2> df (33) Here, in the
equation (33), the function w (f) is a predetermined weight function.
[0100]
Note that the flattening filters SF1 to SFN are not limited to filters determined by optimizing the
evaluation function L as described in Equations (30) to (33), and the characteristic Hall (f) Any
filter that converts to flat characteristics in the frequency domain may be used.
[0101]
Other configuration examples will be described below.
08-05-2019
24
Although FIG. 4 shows the flattening filter SF1 in the filter circuit FT1 as a configuration for
receiving the output of the correction filter CF1, the order of receiving the signals may be
changed.
That is, the flattening filter SF1 may receive the output of the A / D conversion unit AD1, and the
output of the flattening filter SF1 may be received by the correction filter CF1.
[0102]
In general, when processing is performed by combining linear filters such as an FIR filter, an IIR
filter, and an analog filter, the result of the filter processing is the same even if the processing
order is changed. Therefore, as long as the flattening filter and the correction filter are both
composed of these filters, the signal receiving order may be changed.
[0103]
In addition, as another configuration example, the flattening filters SF1 to SFN are not limited to
first-order FIR filters, and may be higher-order FIR filters. In this case, although the amount of
operation of the flattening filter is increased, it is possible to flatten the frequency characteristic
with high accuracy even if the characteristic Hall (f) is a more complicated characteristic.
[0104]
Further, as another configuration example, the flattening filters SF1 to SFN may be IIR filters
instead of FIR filters as long as conditions for stable operation can be set. As described above, the
condition under which the IIR filter operates stably means that when the characteristics are
expressed on the z plane of the z-transform, all poles (where the denominator of the transfer
function becomes zero) are unit circles on the z plane Is included in the
[0105]
In addition, as another configuration example, the flattening filter is not limited to a digital filter
08-05-2019
25
such as an FIR filter, and may be an analog filter as long as a condition for stable operation can
be set. As described above, the condition under which the analog filter operates stably is the case
where all the poles are included in the negative region on the s plane, when the characteristics
are expressed on the s plane of the Laplace transform.
[0106]
Further, as another configuration example, the sound collection device 1A may be replaced with
the filter circuits FTA1 to FTAN and may be provided with a DSP that performs the same
processing as the processing of the filter circuits FTA1 to FTAN by software.
[0107]
FIG. 5 is a diagram showing the characteristics of each part of the sound collection device 1A.
[0108]
Referring to FIG. 5, a graph GR10 shows the characteristics after the correction processing has
been performed by the correction filter.
The characteristics shown by the graph GR10 are the same as the characteristics of each of the
graphs GR7 to GR9 in FIG. 3 and are equal to Hall (f).
[0109]
The graph GR11 also shows the characteristic A (f) of the flattening filter.
The graph GR13 shows the characteristics of Hall (f) и A (f) obtained by synthesizing the
characteristics of GR10 and GR11. It is shown that the characteristics of the graph GR13 are flat
characteristics in the frequency domain as compared with the characteristics of the graph GR10.
[0110]
08-05-2019
26
On the other hand, as a characteristic of an ideal flattening filter, the graph GR12 shows the
characteristic of a filter (inverse filter) having the inverse characteristic of the characteristic
indicated by GR10. The graph GR14 is a characteristic indicated as Hall (f) и Hall <?1> (f)
obtained by synthesizing the characteristics of GR10 and GR12. The characteristics are
completely flattened in the graph GR14.
[0111]
As shown in the graph GR14, ideal characteristics can be obtained by using an inverse filter as
the flattening filter. However, using such an inverse filter may cause the operation of the filter to
become unstable. On the other hand, the characteristic does not have to be strictly flat as long as
an audio signal which does not make the human auditory sense of strangeness be obtained. In
the sound pickup apparatus of the second embodiment, by using a filter with guaranteed stability
as the flattening filter, it is possible to obtain a sound signal that is comfortable for human
hearing with simple processing while performing stable operation. It is.
[0112]
Third Embodiment A sound collection device according to a third embodiment is a sound
collection device whose circuit scale is reduced as compared with the sound collection device
according to the second embodiment.
[0113]
FIG. 6 is a block diagram showing the configuration of a sound collection device according to the
third embodiment.
[0114]
4 and 6, sound collection device 1B differs from sound collection device 1A of FIG. 4 in that it
includes N filter circuits FTB1 to FTBN instead of filter circuits FTA1 to FTAN.
[0115]
The filter circuit FTB1 is one filter obtained by combining the correction filter CF1 included in
the filter circuit FTA1 of FIG. 4 and the flattening filter SF1.
08-05-2019
27
Similarly, filter circuits FTB2 to FTBN are each a combination of the correction filter and the
flattening filter included in filter circuits FTA2 to FTAN in FIG.
[0116]
In general, two FIR filters are combined into one FIR filter having coefficients obtained by
convolving the filter coefficients.
As an effect of the sound collection device of the third embodiment, the circuit scale of the sound
collection device 1B of FIG. 6 is larger than that of the correction filter and the flattening filter
separately configured by dedicated hardware as in the sound collection device 1A of FIG. It
becomes smaller.
[0117]
In addition, the sound collection device 1B may include a DSP that performs the same filtering
process by software instead of the filter circuits FTB1 to FTBN.
When filtering is performed by the DSP, the amount of computation is greatly reduced.
[0118]
FIG. 7 is a diagram showing the characteristics of each part of the sound collection device 1B.
[0119]
Although FIG. 7 describes the case where N = 3 in the sound collection device 1B of FIG. 6, each
portion of the sound collection device 1B has characteristics similar to the characteristics shown
in FIG. .
[0120]
08-05-2019
28
Referring to FIG. 7, the graph GR20 shows the characteristic H1 (f) of the correction filter CF1 of
FIG. 4, the graph GR21 shows the characteristic H2 (f) of the correction filter CF2 of FIG. 4, and
the graph GR22 shows the correction of FIG. The characteristic H3 (f) of the filter CF3 is shown.
The graph GR20 is identical to the graph GR4 of FIG.
Similarly, the graph GR21 is identical to the graph GR5 of FIG. 3, and the graph GR22 is identical
to the graph GR6 of FIG.
[0121]
Also, GR23 indicates the characteristic A (f) of the flattening filter. The graph GR23 is identical to
the graph GR11 of FIG.
[0122]
Graphs GR24 to GR26 are the characteristics of the filter circuits FTB1 to FTB3. The graph GR24
shows the characteristics of the filter circuit FTB1. Similarly, the graph GR25 shows the
characteristic of the filter circuit FTB2, and the graph GR26 shows the characteristic of the filter
circuit FTB3.
[0123]
The characteristic indicated by the graph GR24 is a characteristic H1 (f) и A (f) combining the
correction filter CF1 of FIG. 4 and the flattening filter, and the characteristic indicated by the
graph GR25 is a combination of the correction filter CF2 of FIG. 4 and the flattening filter The
characteristic H2 (f) и A (f) is the characteristic H2 (f) и A (f), and the characteristic indicated by
the graph GR26 is the characteristic H3 (f) и A (f) obtained by combining the correction filter CF3
and the flattening filter in FIG.
[0124]
The characteristics of the microphones MC1 to MC3 included in the receiving circuits DT1 to
08-05-2019
29
DT3 are the same as the graphs GR1 to GR3 in FIG.
The result obtained by combining the characteristics of the microphones MC1 to MC3 and the
characteristics of the filter circuits FTB1 to FTB3 is Hall (f) и A (f) shown by the graph GR13 in
FIG.
[0125]
[Fourth Embodiment] Similar to the sound collection device of the third embodiment, the sound
collection device of the fourth embodiment is a sound collection device whose circuit scale is
reduced as compared with the sound collection device of the second embodiment. is there.
[0126]
FIG. 8 is a block diagram showing a configuration of a sound collection device of the fourth
embodiment.
[0127]
With reference to FIGS. 1 and 8, the sound collecting apparatus of FIG. 1 in that sound collecting
apparatus 1C includes M flattening filters SF1 to SFM that receive the output of array signal
processing unit 2 and flatten the characteristics. Different from 1.
The filter coefficients of the flattening filters SF1 to SFM are the same as the filter coefficients of
the flattening filters SF1 to SFN of FIG.
M is an integer of 1 or more and N or less. The flattening filters SF1 to SFM are formed, for
example, of a first-order FIR filter, similarly to the flattening filter of FIG.
[0128]
Note that to describe another configuration example, the filter order of the flattening filters SF1
to SFM is not limited to one, and may be a higher-order FIR filter. In this case, although the
amount of operation of the flattening filter is increased, it is possible to flatten the frequency
08-05-2019
30
characteristic with high accuracy even if the characteristic Hall (f) is a more complicated
characteristic.
[0129]
As another configuration example, the flattening filters SF1 to SFM may be IIR filters instead of
FIR filters, as long as conditions for stable operation can be set. As described above, the condition
under which the IIR filter operates stably means that when the characteristics are expressed on
the z plane of the z-transform, all poles (where the denominator of the transfer function becomes
zero) are unit circles on the z plane Is included in the
[0130]
In addition, as another configuration example, the flattening filter is not limited to a digital filter
such as an FIR filter, and may be an analog filter as long as a condition for stable operation can
be set. As described above, the condition under which the analog filter operates stably is the case
where all the poles are included in the negative region on the s plane, when the characteristics
are expressed on the s plane of the Laplace transform.
[0131]
In addition, as another configuration, the sound collection device 1C may include a DSP that
performs the same filtering process as the flattening filters SF1 to SFM by software, instead of
the flattening filters SF1 to SFM.
[0132]
The effects of the sound collection device of the fourth embodiment will be described.
The signals input from the reception circuits DT1 to DTN to the array signal processing unit 2
are synthesized by the array signal processing unit 2. When performing array signal processing
such as directivity processing, generally, many microphones are often used to generate one or
two audio signals, and the number of output signals is smaller than the number of input signals.
In such a case, the number of filtering processes required can be reduced by performing the
08-05-2019
31
flattening process on the signal after array signal processing.
[0133]
By providing a filter for performing flattening processing after the array signal processing unit 2,
when the flattening filter is dedicated hardware, the number of filter circuits can be reduced, and
the circuit scale can be reduced. is there.
[0134]
Further, when the filtering process is performed by the DSP instead of the flattening filter, the
amount of operation in the DSP can be reduced, and high speed processing becomes possible.
[0135]
FIG. 9 is a diagram showing the characteristics of each part of the sound collection device 1C.
[0136]
Referring to FIG. 9, a graph GR30 shows a characteristic Hall (f).
[0137]
The graph GR31 is one of the characteristics of the flattening filters SF1 to SFM.
The characteristics shown by the graph GR31 are the same as the characteristics shown by the
graph GR11 of FIG.
[0138]
The graph GR32 is a characteristic represented by Hall (f) и A (f).
The characteristics of the graph GR32 are identical to the characteristics of the graph GR13 of
FIG.
08-05-2019
32
As compared with the characteristics of the graph GR30, the characteristics of the graph GR32
are flat characteristics in the frequency domain.
Therefore, it is possible to obtain an audio signal with less discomfort for human hearing.
[0139]
It should be understood that the embodiments disclosed herein are illustrative and nonrestrictive in every respect. The scope of the present invention is indicated not by the above
description but by the claims, and is intended to include all the modifications within the meaning
and scope equivalent to the claims.
[0140]
FIG. 1 is a block diagram showing a configuration of a sound collection device of a first
embodiment. It is a flowchart explaining the manufacturing method of the sound collection
apparatus 1 of FIG. It is a figure explaining the characteristic of each part of sound collection
device 1. FIG. 7 is a block diagram showing a configuration of a sound collection device of a
second embodiment. It is a figure which shows the characteristic of each part of 1 A of sound
collection apparatuses. FIG. 13 is a block diagram showing a configuration of a sound collection
device of a third embodiment. It is a figure which shows the characteristic of each part of the
sound collection apparatus 1B. FIG. 16 is a block diagram showing a configuration of a sound
collection device of a fourth embodiment. It is a figure which shows the characteristic of each
part of 1 C of sound collection apparatuses.
Explanation of sign
[0141]
1, 1A to 1C sound pickup device, 2 array signal processing unit, 3 ROM, AD1 to ADN A / D
conversion unit, CF1 to CFN correction filter, DT1 to DTN reception circuit, FT1 to FTN, FTA1 to
FTAN, FTB1 to FTBN Filter circuit, GR1 to GR32 graph, MC1 to MCN microphone, S0 to S4 step,
SF1 to SFN flattening filter.
08-05-2019
33
08-05-2019
34
Документ
Категория
Без категории
Просмотров
0
Размер файла
47 Кб
Теги
jp2005286712
1/--страниц
Пожаловаться на содержимое документа