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 JP2010226519
To realize a system with improved convergence characteristics and a simplified configuration. An
initial value memory 1 for storing initial values of coefficient values is provided, and at the start
of operation, the initial value of coefficient values is transferred from the initial value memory 1
to the coefficients of the adaptive inverse filter 3 and when adaptive, unknown transfer function
A switching circuit for serially connecting 2 and the adaptive inverse filter 3 is provided, and the
output of the band pass filter 4 and the output of the adaptive inverse filter 3 are subtracted by
the subtracter 5 and input to the coefficient updating arithmetic circuit 6 as an error signal. The
coefficient update calculation circuit 6 calculates a coefficient update value for updating the
coefficient value of the adaptive inverse filter 3 based on the error signal and the output signal of
the unknown transfer function 2, and at the time of non-adaption, the unknown transfer function
2 and The inverse of the adaptive inverse filter 3 is reversely connected, the output of the band
pass filter 4 and the output of the unknown transfer function 2 are subtracted by the subtracter
5, and when the error signal becomes a certain value or more, the switching circuit operates
again as adaptation time. Do. [Selected figure] Figure 1
Inverse filter system
[0001]
The present invention finds an inverse characteristic of an unknown transfer function at high
speed and stably by giving an initial value to the coefficient of the adaptive inverse filter, and
after the coefficient value of the adaptive inverse filter converges, the coefficient value is fixed. In
particular, in the case of external sound localization, the transfer function equivalent to the head
transfer function is created by adaptively obtaining the inverse characteristic of the listener's ear
canal transmission characteristic and fixing the inverse characteristic, and the individual It
effectively corrects for differences and earphone characteristic differences.
08-05-2019
1
[0002]
As a conventional adaptive signal processing system, there are some which are described, for
example in patent document 1 and an adaptive signal processing system.
FIG. 3 is a block diagram of a conventional example. FIG. 4 is an operation occupancy rate chart
showing the occupancy required for each operation processing with respect to the entire
operation processing time when the operation of the conventional example is calculated by
computer simulation.
[0003]
The configuration of FIG. 3 is an example of the adaptive inverse filter system of the patent
application. A detailed explanation of this is given in T. Horiuchi et. Al .: "Adaptive Estimation of
Transfer Functions for Sound Localization Using Stereo Earphone-Microphone Combination",
IEICE, Vol. E85-A, No. 8, pp. 1841-1850, August, 2002.
[0004]
In FIG. 3, it comprises an initial value memory 11, a pre-processing filter 12, an unknown
transfer function 13, an adaptive inverse filter 14, a band pass filter 15, a subtractor 16, and a
coefficient updating operation circuit 17.
[0005]
The initial value memory 11 is a memory for storing a specific initial value as a coefficient value,
and a reverse impulse response sample of a barycentric vector representing the impulse
response {c} of the unknown transfer function 13 of a plurality of listeners as a vector,
Alternatively, it has an inverse impulse response sample {h (n)} of the unknown transfer function
13 closest to the center of gravity as an initial value.
The pre-processing filter 12 is a filter that can vary the filter characteristic with the coefficient
08-05-2019
2
value {h (1)} from the initial value memory 11 or the coefficient value {h (n)} from the adaptive
inverse filter 14. It has the same characteristics as The pre-processing filter 12 is configured to
receive an input signal x (n) and output an output s (n) to the unknown transfer function 13 and
the band pass filter 15.
[0006]
The adaptive inverse filter 14 constitutes the inverse characteristic of the unknown transfer
function 13 and receives the output signal z (n) of the unknown transfer function 13 as its input,
and uses the coefficient update value Δh (n) from the coefficient update operation circuit 17 It is
a filter that can change the characteristics one by one. The band pass filter 15 is a filter
connected in parallel to a circuit in which the unknown transfer function 13 and the adaptive
inverse filter 14 are connected in series, and subjected to delay and band limitation necessary to
obtain the reverse characteristic of the unknown transfer function 13 An output signal s (n) of
the pre-processing filter 12 is input, and the output d (n) thereof is connected to the subtractor
16. The subtractor 16 is a subtractor that subtracts the output y (n) of the adaptive inverse filter
14 from the output d (n) of the band pass filter 15 and outputs an error signal e (n). The
coefficient update calculation circuit 17 is a circuit for calculating the coefficient update value
Δh (n) of the adaptive inverse filter 14 from the output signal z (n) of the unknown transfer
function 13 and the output signal e (n) of the subtracter 6. .
[0007]
First, at the start of operation, the coefficient value {h (1)} of the initial value memory 11 is
copied as the coefficient value of the preprocessing filter 12 and the adaptive inverse filter 14.
Next, when the input signal x (n) is input to the pre-processing filter 12, the convolution
operation result s (n) of the coefficient transferred in advance from the initial value memory 11
and the input signal x (n) is output. . The output signal s (n) of the pre-processing filter 12 is
input to the unknown transfer function 13, and the output signal z (n) of the unknown transfer
function 13 is output. The output signal z (n) of the unknown transfer function 13 is input to the
adaptive inverse filter 14, and the output signal y (n) is output. On the other hand, the output
signal s (n) of the pre-processing filter 12 is input to the band-pass filter 15, and the output
signal d (n) is output. The subtractor 16 subtracts the output signal y (n) of the adaptive inverse
filter 14 from the output signal d (n) of the band pass filter 15. The coefficient updating is
performed by the coefficient updating arithmetic circuit 17 which receives the output signal z (n)
of the unknown transfer function 13 and the error signal e (n) with the output as the error signal
e (n).
08-05-2019
3
[0008]
The above description is expressed by a formula. Since the pre-processing filter 12 is composed
of an M-tap FIR filter, in order to obtain its output signal s (n), the input signal xM (n) and the
impulse response samples h 'M (n) of the pre-processing filter 12 Perform the convolution
operation of xM (n): [x (n), x (n-1), ..., x (n-M + 1)] <T> (1) h'M (n): [h '(n), h '(n-1), ..., h' (n-M + 1)]
<T> (2) s (n) = {h'M} <T> xM (n) (3)
[0009]
Similarly, since the unknown transfer function 13 is composed of an N-tap FIR filter, in order to
obtain its output signal z (n), the input signal sN (n) and the impulse response sample cN (n of
the unknown transfer function 13) Perform the convolution operation of). sN (n): [s (n), s (n-1), ...,
s (n-N + 1)] <T> (4) z (n) = {cN (n)} <T> sN (n) (5)
[0010]
Similarly, since the adaptive inverse filter 14 is composed of an M-tap FIR filter, in order to
obtain its output signal y (n), the input signal zM (n) and the impulse response samples hM (n) of
the adaptive inverse filter 14 Perform the convolution operation of). zM (n) = [z (n), z (n-1),..., z (nM + 1)] <T> (6) y (n) = {hM} <T> zM (n) (7) The subtractor 16 subtracts the output signal y (n) of
the adaptive inverse filter 14 from the output signal d (n) of the band pass filter 15 to obtain e
(n) = d (n) −y (n), which is The error signal e (n) is obtained. Here, as is well known in the LMS
algorithm, Δh M (n) = μ z M (n) e (n) (8)
[0011]
As described above, the inverse filter also has a form in which the transfer function of the band
pass filter 15 is replaced by the problem of identifying the transfer function divided by the
unknown transfer function 13. Obviously, it is possible to apply NLMS other than LMS which is a
sequential algorithm, affine projection (AP) and recursion (RLS). Here, in the affine projection
algorithm, the coefficient update operation circuit 17 is performed as follows. ΔhM (n) = [ZM, r
08-05-2019
4
(n) <T>] <+> er (n) (9) hM (n + 1) = hM (n) + μΔhM (n) (10) where, [ZM , r (n) <T>] <+> = ZM, r
(n) [ZM, r (n) <T> ZM, r (n)] <-1> (11) <> (12) er ( n) = dr (n) -ZM, r (n) <T> hM (n) (13) dr (n) = [d
(n), d (n-1),. r + 1)] <T> (14) Here, r represents the order, and the affine projection algorithm is
generally intended to shorten the convergence time for colored signals such as voice and tone. It
is known that a quadratic function is sufficient. Thus, hM (n) can be obtained sequentially.
[0012]
Japanese Patent Laid-Open No. 2002-95097
[0013]
However, the prior art as described above has the following problems.
Noriaki Fujita, Masatoshi Yoshida, Haruhide Sugakari, Shoji Shimada, Shin-ichi Fushino, "Study on
DSP adaptive inverse filter for external auditory canal transfer function compensation," IEICE
Technical Report, EA 2007-70, pp. 37-42 (2007), aiming at real-time processing, an occupancy
rate of a specific operation processing time is first described by computer simulation, and FIG. 4
shows an occupancy time required for operation processing by computer simulation. Indicates
the percentage. The FIR filter is a product-sum operation of convolution, which takes time to
perform the arithmetic processing, and the arithmetic processing of the preprocessing filter 12,
the adaptive inverse filter 14, and the band-pass filter 15 is 29.1%, 13.0%, and 14.1%,
respectively. Considerable difficulties are expected to realize real-time arithmetic processing
using DSP. That is, if the pre-processing filter is eliminated, the load of the arithmetic processing
decreases by about 30%, and other necessary processing becomes possible, and the arithmetic
processing time can be shortened.
[0014]
The present invention adopts the following configuration in order to solve the problems
described above. <Configuration 1>
[0015]
08-05-2019
5
An adaptive inverse filter connected in series with the unknown transfer function at the time of
adaptation to form an inverse characteristic of the unknown transfer function, and obtaining a
coefficient value based on a given coefficient update value, the filter characteristic being variable
according to the coefficient value And a band pass filter connected in parallel to the series
connection circuit of the unknown transfer function and the adaptive inverse filter, and having a
filter characteristic such that an output signal through the series connection circuit becomes
equal to a signal passing through itself. An adaptive inverse filter from an initial value memory
that stores a specific initial value as a coefficient value, a subtractor that subtracts the output
signal of the adaptive inverse filter from the output signal of the band pass filter, and an output
signal of an unknown transfer function and an output signal of the subtractor The coefficient
update arithmetic circuit for calculating the coefficient update value of, and at the start of
operation, the initial value of the coefficient value is transferred from the initial value memory to
the adaptive inverse filter to reduce it. When the output signal (error signal) of the converter falls
below a certain value, the coefficient value of the adaptive inverse filter is fixed and this is taken
as non-adaptive time, and when non-adaptive, it is connected with the adaptive inverse filter and
unknown transfer function. An inverse filter system comprising the switching circuit according to
claim 1. <Configuration 2>
[0016]
In the inverse filter system described in Configuration 1, during adaptation, the update
coefficient value obtained from the coefficient update arithmetic circuit is transferred to the
convergence value memory, and the convergence value memory is transferred to the initial value
memory. Characterized inverse filter system. <Configuration 3>
[0017]
The inverse filter system according to Configuration 1 or 2, wherein the initial value is a band
pass filter impulse response sample. <Configuration 4>
[0018]
In the inverse filter system according to Configuration 1 or 2 or 3, the switching circuit is
configured to be in the operation configuration at the time of adaptation when the error signal of
the output signal of the subtractor at non-adaptation time exceeds a certain value. An inverse
filter system characterized by
08-05-2019
6
[0019]
According to the present invention, after performing the learning operation of finding the
adaptive inverse filter coefficient value by connecting the unknown transfer function and the
adaptive inverse filter in series at the time of adaptation, the coefficient value of the learned
adaptive inverse filter is fixed at the non-adaption time. Since the unknown transfer function and
the adaptive inverse filter are reversely connected and the unknown transfer function is
corrected by the adaptive inverse filter, the pre-processing filter is not required as in the
conventional example, and the convergence characteristic is improved although the configuration
is simplified. Can.
[0020]
FIG. 5 is a block diagram illustrating adaptation of the inverse filter system of the present
invention.
It is a block diagram which shows the non-adaptation time of the reverse filter system of this
invention.
It is a block diagram which shows the adaptive inverse filter system of a prior art example. It is
an operation occupancy rate chart required for operation of each block of a prior art example. It
is principle explanatory drawing which implement ¦ achieves an out-of-head sound image
localization. It is a functional block diagram of only one ear side in an extra-head sound image
localization apparatus. It is characteristic explanatory drawing of an ear canal transfer function
(ECTF). It is explanatory drawing which compares and shows the convergence characteristic of a
prior art example and this example.
[0021]
Hereinafter, embodiments of the present invention will be described.
[0022]
Hereinafter, embodiments of the present invention will be described in detail using specific
examples.
08-05-2019
7
[0023]
<< Specific Example >> <Configuration> FIG. 1 is a block diagram showing a specific example of
the inverse filter system of the present invention.
The illustrated system comprises an initial value memory 1, an unknown transfer function 2, an
adaptive inverse filter 3, a band pass filter 4, a subtractor 5, a coefficient update calculation
circuit 6, an error detection circuit 7, and a convergence value memory 8.
[0024]
The initial value memory 1 is a memory for storing a specific initial value as a coefficient value,
and in this specific example, has an impulse response sample of the band pass filter 4 as an
initial value.
Thereby, the amplitude characteristic of the unknown transfer function 2 starts from the flat
state (characteristic of through).
[0025]
The adaptive inverse filter 3 forms the inverse characteristic of the unknown transfer function 2
and receives the output signal z (n) of the unknown transfer function 2 as an input, and uses the
coefficient update value Δh (n) from the coefficient update operation circuit 6 It is a filter that
can change its characteristics sequentially. The band pass filter 4 is a filter connected in parallel
to a circuit in which the unknown transfer function 2 and the adaptive inverse filter 3 are
connected in series, and subjected to delay and band limitation necessary to obtain the inverse
characteristic of the unknown transfer function 2 An input signal x (n) is input, and an output d
(n) thereof is connected to the subtractor 5. The subtractor 5 is a subtractor that subtracts the
output y (n) of the adaptive inverse filter 14 from the output d (n) of the band pass filter 15 and
outputs an error signal e (n). The coefficient update calculation circuit 6 is a circuit for
calculating the coefficient update value Δh (n) of the adaptive inverse filter 3 from the output
signal z (n) of the unknown transfer function 2 and the output signal e (n) of the subtractor 5 is
there. The error detection circuit 7 is a circuit that outputs a command signal to stop the
coefficient update value when the error signal e (n) falls below a certain value. The convergence
08-05-2019
8
value memory 8 is a memory for temporarily storing the adaptive inverse filter coefficient value
after convergence by the command signal from the error detection circuit 7 at the time of
adaptation. Thereafter, the convergence value memory 8 is configured to be transferred to the
initial value memory 1, and the convergence characteristic is further improved by transferring
the coefficient value of the convergence value memory 8 to the impulse response coefficient of
the adaptive inverse filter 3 at the start of operation.
[0026]
At the start of operation, the coefficient value of the initial value memory 1 is transferred to the
impulse response coefficient of the adaptive inverse filter 3, and the switch S1 S2 S3 S4 S5 of the
switching circuit of FIG. 1 is turned to the a side. Therefore, in this configuration, the output
signal y (n) of the adaptive inverse filter 3 is input from the input signal x (n) to the input signal
of the subtracter 5 via the unknown transfer function 2 in parallel with the input signal x (n). And
the output signal d (n) of the band pass filter 4 becomes the input signal of the subtractor 5, and
the output signal of the subtractor 5 becomes the error signal e (n). On the other hand, the
output signal s (n) of the unknown transfer function is switched to the input signal z (n) of the
adaptive inverse filter 3 by the switching circuit, and this input signal z (n) and the error signal e
(n) A coefficient update value Δh M (n) is output by the coefficient update calculation circuit 6
which receives as an input. This output signal .DELTA.hM (n) is input to the coefficient value of
the adaptive inverse filter, and coefficient updating is performed.
[0027]
Further, in FIG. 2, the switches S1 S2 S3 S4 S5 of the switching circuit at the time of nonadaptation fall to the b side. That is, the input signal x (n) is connected to the adaptive inverse
filter 3 and the unknown transfer function 2 and is connected to the input of the subtractor 5,
while the input signal x (n) is also subtracted via the band pass filter 4 Since it is connected to the
5 input, the output signal of the subtractor 5 is the error signal e (n), and if the error signal e (n)
is constantly monitored, the value will become high again if it exceeds a certain value. If the
switch S1 S2 S3 S4 S5 of the switching circuit is turned to the a side to start the adaptive
operation, the operation can be adaptively performed again. Since these operations are actually
performed using a DSP, they can be easily realized by software.
[0028]
08-05-2019
9
<Effects> Next, in order to clarify the effects of the above specific example, a case where it is
specifically applied to the outside sound localization will be described. First, an explanation will
be given of the localization of the extra-sound image. FIG. 5 is an explanatory view of the
principle for realizing the out-of-head sound image localization. In FIG. 5, (a) represents listening
by a speaker, and (b) represents listening by a binaural earphone. In the figure, 100 is a sound
source signal, 101 is a speaker, 102 is a listener, 103 is a digital filter, 104 is an earphone, and
105 is a microphone installed in the ear canal of the listener. The suffixes L and R such as SSTFL
and SSTFR indicate the left side and the right side.
[0029]
The principle of the out-of-head sound image localization will be described below with reference
to FIG. By applying the same stimulation as the sound stimulation from the sound source in the
space (speaker 101 in FIG. 5A) to the tympanic membrane from the earphone 104 to the
tympanic membrane, the sound source can be perceived as being outside. However, since it is
extremely dangerous to capture vibration stimulation signals on the tympanic membrane by
sound waves from a living body as an electrical signal, the transfer function of the electric signal
from the sound source signal lOO to the tympanic membrane in FIG. It is not possible.
[0030]
Therefore, a very small microphone lO5 is attached to the ear canal of both ears, and the transfer
function from the sound source signal 100 radiated from the speaker lOl to the output of the
microphone 105, that is, the space acoustic transfer function in the left and right ears Measure
Transfer Function).
[0031]
Here, the documents: Iens Brauert, Masayuki Morimoto, Toshiyuki Goto "Spatial sound", as
described in the sound propagation in the ear canal, section 2.4 of Kashima Publishing, pp. 2028, the sound waves in the ear canal Since it is expressed by a one-dimensional model, it is
obvious that the electric signal by the vibration on the tympanic membrane becomes equivalent
to the left and right in FIG. 5 if the installation position of the microphone 105 is fixed.
Therefore, even if the microminiature microphone 105 is fixed to the ear canal and the transfer
08-05-2019
10
function, that is, the impulse response in the time domain is measured, the result is equivalent.
[0032]
However, as pointed out in the above document, since the speaker 101 has frequency
characteristics, the true transfer function of the electric signal from the input of the speaker 101
to the output of the microphone 105 is the transfer function of the speaker 101 If it is set as
LSTF (Loud-Speaker Transfer Function), it is SSTF / LSTF. This value is almost equivalent to head
related transfer function HRTF (Head Related Transfer Function). LSTF is a speaker transfer
function for correcting the frequency characteristic of the speaker 101. The head related transfer
function HRTF is a characteristic change from the sound source when the sound is reproduced
from the speaker 101 in free space to the eardrum position of the listener, and if the HRTF is
directly reproduced by the earphone 104, the same sound as the speaker 101 in the earphone
104 Can be reproduced.
[0033]
HRTF is the ratio of the transfer function when the head is present to the transfer function when
the head is not present, as defined in the above document, and the function when the head is not
present is the function of space propagation and the LSTF Since the space propagation is only a
delay amount, HRTF ≒ SSTF / LSTF.
[0034]
On the other hand, in FIG. 5B, in order to create a transfer function equivalent to this using
binaural earphones (or stereo headphones) lO4, the output of the microphone 105 attached to
the external ear canal from the input of the binaural earphones 104 If the transfer function of
the product of the ECTF and the transfer function of the digital filter 103 matches the transfer
function SSTF / LSTF, the ear canal is measured. The same listening signal as the speaker
listening can be reproduced at the location of the microphone 105 installed in
That is, if the SSTF / LSTF is used as the target frequency characteristic of the binaural earphone
104, it is possible to realize the out-of-head sound localization and listening.
[0035]
08-05-2019
11
As described above, the transfer function of the digital filter 103 can be determined by a
computer. That is, assuming that this transfer function (out-of-head sound localization transfer
function) is SLTF (Sound-image Localization Transfer Function), SLTF = {SSTF / (ECTF · LSTF)} ≒
HRTF / ECTF, and each term of the right side of this equation Are all determined by
measurement, and after that, SLTF can be determined by performing mathematical operations.
[0036]
However, even if it is possible to obtain the above SLTF, the transfer functions ECTF and HRTF
differ depending on the size of the ear canal of the listener, the size of the ear, and the size of the
face. That is, unless the transfer function conforms to the face shape of the individual, an
accurate sound image can not be localized outside the head, and in the worst case, the sound
image can not be determined before and after, and may not be localized outside the head.
[0037]
For this reason, since it is not used widely for an unspecified number of listeners if localization is
not made out of head due to individual differences in transfer functions, a hard solution method
to select several kinds of transfer functions measured in advance, Or a system that can measure
the transfer function for each individual is needed. However, the above-described hardware
solution causes an increase in the amount of hardware, and the method using a system that
measures the transfer function for each individual makes such a measurement system very
expensive.
[0038]
By the way, studies on the individuality of the transfer functions ECTF and SSTF have already
been made. For example, according to S. Yano, H. Hokari, and S. Shimada; "A Study on Personal
difference in the Transfer Functions of Sound Localization", AES the IO 6th Convention, NO.
4922, May 8-11 (1999). It is concluded that the function ECTF is more individual than the
transfer function SSTF.
08-05-2019
12
[0039]
Therefore, by introducing the adaptive signal processing technology, a method for obtaining the
transfer function of the ECTF matched to oneself while listening to the musical tone and voice
without measuring the transfer function of the ECTF for each individual has already been
disclosed in Japanese Patent Application No. No. 10-261292 filed in "Earphone and extra-head
sound localization device".
[0040]
FIG. 6 is a functional block diagram of only one ear side in the extra-head sound image
localization apparatus.
The illustrated system shows a block on only one ear side for realizing the function of FIG. 5 (b),
in which only the inverse transfer function of ECTF is to be adaptively equalized. It is a wellknown fact that adaptive equalization takes time (convergence time) to identify the transfer
function when the impulse response length of the unknown transfer function is long. Therefore,
it is desirable to adapt short impulse response lengths. The above HRTF has a time length of
about 30 msec or more, and the impulse response length of ECTF is about 5 msec in comparison,
and it is sufficient to estimate the inverse impulse response length of ECTF by at most twice as
much. It is desirable to target it for adaptation. As described above, it is better to store the HRTF
transfer function in a direction known in advance or the transfer function of SSTF / LSTF in a
memory, and first adaptively equalize 1 / ECTF, which has remarkable differences among
individuals. It turns out that it is preferable practically.
[0041]
The system shown in FIG. 6 includes a digital filter 201, an adaptive filter 202, an earphone
(speaker) 203, a microphone 204, a band pass filter 205, and a subtractor 206. The digital filter
201 calculates the impulse response of the transfer function of SSTF / LSTF calculated in
advance. It is a digital filter stored. The adaptive filter 202 is a digital filter connected in series
with ECTF which is an unknown transfer function. The band pass filter 205 is a band pass filter
that passes a predetermined frequency band with respect to the output of the digital filter 201.
[0042]
08-05-2019
13
The reason why the band pass filter 205 is provided is as follows. That is, the adaptive filter 202
and ECTF are connected in series, and if this output signal is an impulse, the transfer function of
the adaptive filter 202 is the inverse transfer function (= 1 / ECTF) of ECTF. However, the ECTF
includes the earphone 203 and the microphone 105, and causes attenuation outside the band.
For this reason, the transfer function of the adaptive filter 202, which is the inverse function of
ECTF, has a large gain outside the band. That is, the gain becomes infinite outside the band.
[0043]
Therefore, if the convolution operation result of each impulse response of the adaptive filter 202
and ECTF is used as the impulse response of the band pass filter 205, the tap coefficient value of
the adaptive filter 202 or the impulse response value can be stably obtained. That is, if the band
of the band pass filter 205 is made to pass a band narrower than the band of the adaptive filter
202, the subtractor 206 cancels out-of-band part of the transfer function from the adaptive filter
202 and stabilizes the solution. You can ask for
[0044]
Furthermore, in order to clarify some of the typical characteristics of this ECTF, impulse response
characteristics and frequency amplitude characteristics in the time domain of ECTF are shown.
FIG. 7 is an explanatory view showing impulse response characteristics in the time domain of
ECTF and frequency amplitude characteristics. According to these characteristics, since the
impulse response length is 128 samples at 44.1 kHz sampling, taking its multiple as a safety
factor, it is approximately 5 msec or less, and this characteristic is mainly the ear ear
characteristics of the listener (the ear canal Volumes and types of earphones (electrodynamic and
electromagnetic types, open air and closed types, "intra-conca-type" to be inserted into the ear,
"Supraconca-type" to be placed on the ear, "surcam oral-type" covering the ear, etc.) It depends
on the electro-acoustic characteristics that depend on the Therefore, this ECTF corresponds to
the unknown transfer function 2 of the present invention shown in FIG.
[0045]
The effects of applying this specific example to the above-described extra-head sound image
localization technology will be described below. It is important to show convergence
08-05-2019
14
characteristics in order to clarify the performance of this example and the conventional example.
Therefore, the estimation error ε (n) is defined as follows. (15) However, B (f), C (f) and H (f, n)
in equation (15) are the frequency characteristics of the band pass filter 4, the unknown transfer
function 2 and the adaptive inverse filter 3, respectively. The estimation error ε (n) indicates the
estimation error at each time of convergence, and the product of the transfer function (1 / ECTF)
of the adaptive inverse filter 3 and the unknown transfer function (ECTF) is the target transfer
function. It indicates that a certain band pass filter 4 is approached.
[0046]
FIG. 8 is a convergence characteristic diagram of the estimation error ε (n). In the figure, (a)
shows the conventional example shown in FIG. 3, and (b) shows the convergence characteristic of
this specific example shown in FIG. Further, it represents an estimation error obtained by
applying the affine projection (AP) algorithm to the coefficient updating operation when a
colored signal (pink noise signal) is input. The horizontal axis shows time, and the sampling
frequency is 44.1 kHz. Each figure from the left 21845 samples (about 0.5 seconds), 43690
samples (about 1.0 seconds), 65535 samples (about 1.5 seconds), the upper side is the
convergence characteristic, the lower side is the frequency amplitude characteristic at each
sample time (elapsed time) FIG.
[0047]
As shown in the drawing, the convergence time is about 0.5 seconds in the conventional example,
which is -15 dB, and the present invention is -20 dB. It can be seen that the conventional example
(a) is clearly inferior to the inventive example (b) in convergence characteristics. In the input
signal, the case of the relaxation coefficient μ = 0.5 is shown. In the example of the present
invention, the impulse response of the band pass filter 4 is set to the coefficient value of the
adaptive inverse filter 3 as the initial value.
[0048]
The embodiment of the present invention has a configuration essentially different from that of
the prior art. That is, there is no pre-processing filter. In the conventional construction method,
the input signal x (n) is connected to the unknown transfer function and the adaptive inverse
filter through the pre-processing filter, and the coefficients are such that the unknown transfer
08-05-2019
15
function and the adaptive inverse filter have the same characteristics as the band pass filter.
Since the adaptive inverse filter coefficient value is first obtained by using the update, and this
coefficient value is transferred to the coefficient value of the pre-processing filter, the
convergence characteristic is deteriorated. This is apparent from the computer simulation of FIG.
In the construction method according to the present invention, the adaptive inverse filter
coefficient value is obtained at the time of adaptation, and connected at the time of nonadaptation with the coefficient updated adaptive inverse filter and the unknown transfer
function, so a pre-processing filter is not required. It can be seen that the result is almost the
same as the operation result to be identified.
[0049]
In the configuration example of the present invention, the coefficient value of the adaptive
inverse filter 3 is obtained at the time of adaptation, and when the output signal e (n) of the
subtracter 5 becomes a certain value or less, the non-adaptation time is assumed. Since the
coefficient value of the filter 3 is fixed and the output signal y (n) becomes the connection of the
input signal u (n) of the unknown transfer function 2 (ECTF), the transfer function of the adaptive
inverse filter 3 (SLTF = {SSTF / Transfer function (SLTF · ECTF = SSTF / LSTF) of the product with
(ECTF · LSTF)} matches the target transfer function (SSTF / LSTF) It can be reproduced, and the
listener is configured to compensate for the ear canal transfer function ECTF.
[0050]
1: Initial value memory 2: Unknown transfer function 3: Adaptive inverse filter 4: Band pass filter
5: Subtractor 6: Coefficient update operation circuit 7: Error detection circuit 8: Convergence
value memory
08-05-2019
16
© Copyright 2021 DropDoc
