close

Вход

Забыли?

вход по аккаунту

?

DESCRIPTION JP2007288775

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 JP2007288775
A multi-channel echo correction system suitable with improved echo correction is provided. A
multi-channel echo correction system according to the invention receives microphone output
signals from two speaker input channels and at least one microphone connected to the speakers
to provide a speaker input signal emitted by the speaker. A microphone output channel
connected to at least one microphone, a correction channel for each speaker input channel, and a
matching correction filter for each correction channel, wherein the correction output signal is a
microphone for the signal emitted from the speaker A adaptive correction filter, wherein each
adaptive correction filter is configured to filter the signal on an individual correction channel, as
provided to correct the output signal, and a preprocessed speaker input signal on the correction
channel And a pre-processing means for the. [Selected figure] Figure 1
Multi-channel echo correction system and method
[0001]
The present invention relates to multi-channel echo correction systems and methods, and more
particularly to correcting echoes present in microphone signals and arising from multi-channel
sources.
[0002]
The presence of echo in the microphone signal is a problem that arises in different types of
communication systems.
15-04-2019
1
A typical example is hands-free calling in a vehicle such as a vehicle. Such telephone systems
typically include one or more microphones to obtain speech signals from the speakers. However,
some speakers are also mounted in vehicles, such as vehicles. Signals from, for example, a radio
or a CD player are output via these speakers. These signals are also acquired by the microphone
of the hands-free telephone system, which distorts the microphone signal.
[0003]
In order to correct these unwanted signals, adaptive filters are used, which are used to provide a
correction signal corresponding to the unwanted signals contained in the microphone signal. For
this purpose, the adaptive filter uses the input signal of the loudspeaker to determine the
correction signal to be removed from the microphone signal. The structure of a conventional
echo correction system is shown in FIG.
[0004]
In the system shown in FIG. 6, initially, three microphones 601 are provided to obtain speech
signals from the speakers. However, these microphones also obtain the audio signal coming from
the speaker 602. At the microphone output channel 603 connected to the microphone,
beamforming means 604 are provided to combine the microphone output signals in a suitable
manner.
[0005]
The signals hL (n) and hR (n) obtained by the microphone array are mimicked by adaptive
correction filters 605 and 606. These adaptive filters are shown on the loudspeaker signal (d (n)
obtained by the microphone, using the loudspeaker input signals xL (n) and xR (n) (possibly
possible to be amplified in the amplifier 607). Correction signal as close as possible
[0006]
I will provide a. By removing the correction signal from the microphone signal, a portion of the
15-04-2019
2
microphone signal derived from the speaker is removed and the resulting signal e (n) is
minimized.
[0007]
When correcting the signal coming from the car radio 608, the absolute value of the coherence
square can vary considerably.
[0008]
In particular, for signals such as music, the coherence is very small, whereas for news and
interviews etc. the signals xL (n) and xR (n) are (almost) linear dependent and their coherence is
approximately Equal to one.
In the above equations, SxLxR (.OMEGA.), SxLxL (.OMEGA.), And SxRxR (.OMEGA.) Individually
indicate the cross power spectral density and the auto power spectral density of the signals xL
(n) and xR (n). If the coherence is very high, ie, the two signals are approximately linear
dependent, the cost function used in the multi-channel adaptation algorithm does not have a
unique solution. As a result, for example, in the case of an interview, after the speaker changes,
the filter has to be balanced anew, and the echo will reappear for a while.
[0009]
Another example in which echo distorts the microphone signal is a hands-free system provided to
control devices such as car radios, or systems of passenger communication in vehicles such as
vehicles.
[0010]
To date, many efforts have been made to solve the above problems.
According to one proposal, non-linear pre-processing is performed on the radio signal shown in
FIG. 6 by means of corresponding non-linear pre-processing means 609. A possible non-linear
pre-processing can consist of the addition of a half-wave rectifier (see, for example, non-patent
document 1). Another possibility is time variant pre-filtering as disclosed in [2]. J. Benesty, T .;
15-04-2019
3
Gansler, D .; R. Morgan, M .; M. Sondhi, S. L. Gay, Advances in Network and Acoustic Echo
Cancellation, Springer Verlag, Berlin, 2001 A. Sugiyama, Y. Joncour, A .; Hirano, A Stereo Echo
Canceller with Correct Echo-Path Identification Based on an Input-Sliding Technique, IEEE
Transactions on Signal Processing, Vol. 49, Nr. 1, Pages 2577-2587, 2001
[0011]
The disadvantage of these methods is that the signal on the path from the signal source (i.e. car
radio) to the loudspeaker is consequently altered into audible artifacts.
[0012]
In view of the above-mentioned shortcomings of the prior art, the basic problem with the present
invention is to provide a multi-channel echo correction system with improved echo correction.
This problem is solved by the system of claim 1 and the method of claim 17.
[0013]
Thus, the invention provides a multi-channel echo correction system, wherein two speaker input
channels, each speaker input channel being connected to a speaker to provide a speaker input
signal emitted by the speaker , Two speaker input channels and a microphone output channel
connected to the at least one microphone for receiving the microphone output signal from the at
least one microphone, each microphone obtaining a signal emitted from the speaker A
microphone output channel configured and a correction channel for each of the speaker input
channels, each correction channel being connected to an individual speaker input channel and
the microphone output channel; A adaptive correction filter for the correction channel, wherein
each adaptive correction filter is on an individual correction channel such that a correction
output signal is provided to correct the microphone output signal for the signal emitted from the
speaker. Adaptive correction filter, and preprocessing means for preprocessing speaker input
signal on the correction channel, wherein the preprocessing means is configured to filter the two
signals according to a predetermined criterion. Pre-processing means, configured to determine
one of the correlation values of the loudspeaker input signal to the speaker, and to cancel one of
the adaptive correction filters if the determined correlation value passes a predetermined
threshold. Equipped with
[0014]
15-04-2019
4
By releasing one of the adaptive correction filters, the above-mentioned problems of the nonunique solution of the adaptive process are overcome.
[0015]
The predetermined criteria for determining the correlation value may have different forms.
For example, the correlation value may be determined as a coherence value.
In this case, for example, the threshold may be selected as 0.97. If the coherence value is greater
than or equal to this value, the signals on different input channels are highly correlated such that
adaptive filtering to provide a correction signal can be performed using only one filter it is
conceivable that.
[0016]
The multi-channel echo correction system according to the invention has the further advantage
that, in the case of highly correlated signals on different input channels, the required computing
power can be considerably reduced. In particular, in the case of two similar speakers (ie a stereo
system), the computing power required to perform adaptive filtering and filter updating can be
reduced by 50%.
[0017]
The multi-channel echo correction system described above may be associated with the case of
two similar speakers or more than two speakers. In the latter case, a corresponding number of
speaker input channels, correction channels, and adaptive correction filters are provided. If a
correlation value passing a predetermined threshold is determined for two of the loudspeaker
input channels, all but one of the adaptive correction filters may be cancelled.
[0018]
The pre-processing means is configured to provide a linear combination, in particular the
15-04-2019
5
difference and / or the sum of the input channels of the two loudspeakers, to at least one of the
adaptive correction filters.
[0019]
The linear combination of the signals, in particular the differences of the signals, makes it
possible to detect the correlation between the signals in a very simple way.
[0020]
In particular, the pre-processing means provide the sum of the signals of the input channels of
the two speakers for the first filter of the adaptation correction filter, and of the input channels of
the two speakers for the second filter of the adaptation correction filter. It is configured to
provide signal differences.
[0021]
In this way, if the signals of the first and second speaker input channels are highly correlated
(e.g. if the signal output by the speaker is a monaural signal), then the difference signal is very
small or Equal to 0
Therefore, it can be easily determined whether the two input signals are correlated.
[0022]
In this case, in particular, the pre-processing means may be configured to cancel the second filter
of the adaptive correction filter, in case the determined correlation value passes a predetermined
threshold.
[0023]
The preprocessing means are configured to determine the correlation value according to a
predetermined criterion based on the signal power of the signal, in particular to represent the
difference and / or the sum of the signals of the input channels of the two loudspeakers.
[0024]
15-04-2019
6
If the input signals are highly correlated, the difference between these signals is nearly
extinguishing power.
In particular, the signal power may be determined as the norm of the corresponding signal
vector.
That is, a very good indicator of the correlation between the two signals is obtained.
[0025]
In the above-described multi-channel echo correction system, correlation values may be
determined recursively.
This reduces the required computing power to determine the correlation value.
For example, if the signal vector has a length of N, the squared norm of this vector at time n is
the squared norm of the vector at time n-1 + the nth value of the squares of the signal vector-(n
−N) equal to the value of
[0026]
Each adaptive correction filter is configured such that when the adaptive correction filter is
released, the adaptation of the adaptive correction filter is not performed. Such configuration of
the adaptive correction filter further reduces the required computing power and improves the
stability of the filter.
[0027]
In the above-described multi-channel echo correction system, the pre-processing means are two
inputs, each input being two inputs and two outputs connected to individual speaker input
channels. , Each output being connected to an individual adaptive correction filter, a first signal
path connecting two outputs, a first input and a first output, and a second input And a second
15-04-2019
7
signal path connecting the second output, subtraction means on the first signal path, summing
means on the second signal path, a signal on the first signal path, and A third signal path
connecting the first input and the summing means such that the signals on the second signal
path are summed to obtain a summed signal; The signal on the signal path of and the signal on
the second signal path are subtracted As is subtracted in order to obtain a signal, and a fourth
signal path connecting the input and the subtraction means second.
[0028]
This is a very advantageous implementation in a multi-channel echo correction system.
In particular, if the signals on the two speaker input channels are highly correlated, the signal
power of the subtracted signal is very low. On the other hand, in this case, the summed signals
have a signal power substantially corresponding to twice the signal power of the individual
signals in the loudspeaker input channel. If the signal power of the subtracted signal is below a
predetermined threshold, this is the basis for determining that the correlation is very high. The
adaptive filter connected to the first output of the pre-processing means can then be released.
The correction of the microphone output signal may be based solely on the summed signal input
to the corresponding adaptive correction filter.
[0029]
The pre-processing means in particular multiply the subtracted signal and the summed signal by
a common weighting element in order to multiply the signal on the first signal path and / or the
second signal path by the weighting element , And / or multiplication means on the first signal
path and / or the second signal path.
[0030]
In this way, the signals after summation or subtraction may be considered as being
superpositions of different signals.
In particular, if the signals on the first and second loudspeaker input channels are highly
correlated and summed in the pre-processing means, the corresponding summed signal has twice
as much signal power as the loudspeaker input channel Corresponding to one of the signals. In
15-04-2019
8
that case, a signal substantially corresponding to one of the speaker input channels is derived by
multiplying the signal summed by the factor of 0.5.
[0031]
The pre-processing means is a first adaptive pre-processing filter on a third signal path, the
adaptation of the first adaptive pre-processing filter being between signals on the first signal
path and the third signal path. A first adaptive pre-processing filter based on a difference and a
second adaptive pre-processing filter on a fourth signal path, wherein the adaptation of the
second adaptive pre-processing filter is a second signal path and And a second adaptive preprocessing filter based on the difference between the signals on the fourth signal path.
[0032]
For example, when playing an interview on a stereo channel system, the two speakers are often
placed on different sides acoustically.
In other words, one speaker is present on the first channel and another speaker is present on the
second channel. In such cases, the speech signals on both speaker input channels are highly
correlated, however their difference is non-zero. Providing adaptive pre-processing filters on the
third and fourth signal paths is useful in overcoming this problem.
[0033]
The fit correction filter and the fit pre-treatment filter may be configured such that the fitting of
the pre-treatment filter is performed more slowly than the fitting of the correction filter.
[0034]
Such an arrangement improves the stability of the system.
In particular, the adaptation increment of the pre-processing filter may be smaller than the
increment in the adaptation process of the correction filter.
15-04-2019
9
[0035]
The adaptation increment of the correction filter may be larger than the adaptation increment of
the pre-processing filter. In this way, the adaptation of the preprocessing filter is slower than the
adaptation of the correction filter.
[0036]
The preprocessing means may comprise delay means on the first signal path and the second
signal path, each delay means delaying the signal on the first signal path before reaching the
summing means and subtracting means Are configured to delay the signal on the second signal
path before reaching.
[0037]
In this way, the pre-matching filter overcomes the problem that it does not necessarily converge
to the causal optimal solution.
[0038]
In particular, the delay of the delay means is selected such that the delay of the delay means
corresponds to about half the length of the corresponding adaptive pre-filter.
About half of this adaptive pre-processing filter regenerates the non-causal part.
[0039]
The previously described multi-channel echo correction system may comprise summing means
provided between the adaptive correction filter and the microphone output channel and
configured to sum the signals emitted from the adaptive correction filter.
[0040]
In this way, an advantageous correction signal for the subtraction from the microphone output
signal is provided.
15-04-2019
10
[0041]
The at least one microphone of the multi-channel echo correction system described above may
be configured as an array of at least two microphones connected to the beam forming means on
the microphone output channel.
[0042]
Such an arrangement improves the signal to noise ratio at the microphone output channel.
In particular, the beam forming means produce the proper orientation of the microphone array.
[0043]
The invention also provides a method of correcting echo in a multi-channel system, said multichannel system comprising two loudspeakers, each loudspeaker for providing a loudspeaker
input signal emanating from said loudspeakers Two speakers connected to the input channel and
at least one microphone for obtaining a signal emitted from the speakers, the at least one
microphone being connected to the microphone output channel A microphone and a correction
channel for each of the speaker input channels, each correction channel being a correction
channel connected to an individual speaker input channel and a microphone output channel, and
a matching correction filter for each of the correction channels, Correction output The adaptive
correction filters are configured to filter the signals on the individual correction channels, such
that a signal is provided to correct the microphone output signal to the signal emanating from
the speaker A filter, the method comprising the steps of receiving speaker input signals, the
respective speaker input signals being received on a correction channel, and the speaker input
signals on the correction channel Pre-processing the steps of: determining the correlation value
of the two loudspeaker signals for the two loudspeakers according to a predetermined criterion;
and the determined correlation value passing a predetermined threshold And D. pre-processing,
including releasing one of the adaptive correction filters.
[0044]
The features and advantages described above in the context of a multichannel echo correction
system also apply in the case of the correction method.
[0045]
The pre-processing means provides for at least one of the adaptation correction filters a linear
15-04-2019
11
combination, in particular the difference and / or the sum of the input channels of the two
loudspeakers, for at least one of the adaptation correction filters Can be included.
[0046]
The step of pre-processing comprises providing the sum of the signals of the first and second
speaker input channels to the first filter of the adaptive correction filter, and the difference
between the signals of the first and second speaker input channels. Providing a second filter of
the adaptive correction filter may be included.
[0047]
The pre-processing step may include the step of canceling the second filter of the adaptive
correction filter if the determined correlation value passes a predetermined threshold.
[0048]
The pre-processing step may include the step of determining the correlation value according to a
predetermined criterion based on the signal power of the signal, in particular representing the
difference and / or the sum of the signals of the two speaker input channels .
[0049]
In the method described above, the correlation value may be determined recursively.
[0050]
The step of pre-processing comprises the steps of: summing speaker input signals on two
correction input signals to obtain a summed signal; and the speakers on the two correction input
signals to obtain a subtracted signal. And subtracting the input signal.
[0051]
The pre-processing may comprise multiplying the summed signal and the subtracted signal by a
weighting factor, in particular a common weighting factor.
[0052]
In the above method, the pre-processing step is filtering the speaker signal of the first speaker
input channel to fit before being added to the speaker signal of the second speaker channel, the
adaptation being The speaker signal of the second speaker input channel, before being
15-04-2019
12
subtracted from the speaker signal of the second speaker channel, and a step based on the
difference between the speaker signals of the first and second speaker input channels. Filtering
to fit, which may be based on the difference between the loudspeaker signals of the first and
second loudspeaker input channels.
[0053]
In particular, the step of adaptively filtering in the pre-processing step may be performed more
slowly than the adaptation of the adaptive correction filter.
For example, the adaptation increment of the correction filter may be selected to be larger than
the adaptation increment of the filtering step to fit in the pre-processing step.
[0054]
The step of pre-processing comprises delaying the speaker signal of the second speaker input
channel before being summed with the matched filtered speaker signal of the first speaker input
channel; Delaying the loudspeaker signal of the first loudspeaker input channel prior to
subtracting the matched filtered loudspeaker signal of the channel.
[0055]
In particular, the delay may be selected to correspond to about half the length of the
corresponding filter in order to filter the delay to fit in the pre-processing step.
[0056]
The method described above may include the step of summing the signals emanating from the
adaptive correction filter.
[0057]
The present invention also provides a computer program product comprising one or more
computer readable media having computer executable instructions for performing the steps of
the above described method when executed on a computer.
[0058]
15-04-2019
13
The present invention further provides the following means.
A multi-channel echo correction system, comprising: two loudspeaker input channels, each
loudspeaker input channel being connected to the loudspeaker to provide a loudspeaker input
signal emitted by the loudspeaker One speaker input channel, and a microphone output channel
connected to the at least one microphone to receive the microphone output signal from the at
least one microphone, each microphone configured to obtain a signal emitted from the speaker A
microphone output channel, and a correction channel for each of the speaker input channels,
each correction channel being connected to an individual speaker input channel and the
microphone output channel, and A adaptive correction filter for the positive channel, each
adaptive correction filter comprising an individual correction channel, such that a correction
output signal is provided to correct the microphone output signal with respect to the signal
emitted from the speaker. An adaptive correction filter, configured to filter the signal on top, and
preprocessing means for a preprocessing speaker input signal on the correction channel, the
preprocessing means comprising the two processing means according to a predetermined
criterion. A pre-processing means configured to determine a correlation value of the speaker
input signal to the speaker and to release one of the adaptive correction filters if the determined
correlation value passes a predetermined threshold. And a multi-channel echo correction system.
(Item 2) The pre-processing means generates a linear combination of at least one of the adaptive
correction filters, in particular, the difference and / or the sum of the input channels of the two
speakers, of the adaptive correction filter. The multi-channel echo correction system according to
claim 1, configured to provide at least one of the following.
(Item 3) The pre-processing means provides the sum of the signals of the input channels of the
two speakers to the first filter of the adaptive correction filter, and the second filter of the
adaptive correction filter. 3. A multi-channel echo correction system according to item 1 or 2,
configured to provide a signal difference of the input channels of two speakers.
(Item 4) The multi according to item 3, wherein the pre-processing means is configured to cancel
the second filter of the adaptive correction filter when the determined correlation value passes
the predetermined threshold. Channel echo correction system.
(Item 5) The pre-processing means determines the correlation value according to a
15-04-2019
14
predetermined standard based on the signal power of the signal, and in particular, represents the
difference and / or the sum of the signals of the input channels of the two speakers. Item 5. The
multi-channel echo correction system according to any one of Items 1 to 4, which is configured
to.
(Item 6) The multi-channel echo correction system according to any one of items 1 to 5, wherein
the correlation value is determined recursively.
(Item 7) Each of the adaptive correction filters is configured to prevent the adaptation of the
adaptive correction filter from being performed when the adaptive correction filter is released.
Channel echo correction system.
(Item 8) The pre-processing means is two input units, each input unit being two input units and
two output units connected to an individual speaker input channel. The outputs are connected to
respective adaptive correction filters, a first signal path connecting two outputs, a first input and
a first output, and a second input and a second A second signal path connecting the outputs,
subtraction means on the first signal path, summing means on the second signal path, a signal on
the first signal path, and the second A third signal path connecting the first input and the
summing means such that the signals on the signal path are summed to obtain a summed signal;
And the signal on the second signal path are subtracted to obtain a subtracted signal. Item 8. A
multi-channel echo correction system according to any of the preceding items, comprising a
second input and a fourth signal path connecting the subtracting means.
9. The pre-processing means, in particular, the signal subtracted by a common weighting
element, in order to multiply the signal on the first signal path and / or the second signal path by
a weighting element. 9. A multi-channel echo correction system according to claim 8, comprising
multiplication means on the first signal path and / or the second signal path to multiply the
summed signals.
(Item 10) The pre-processing means is a first adaptive pre-processing filter on the third signal
path, wherein the adaptation of the first adaptive pre-processing filter is the first signal path and
the first adaptive signal. A first adaptive pre-processing filter based on the difference between the
signals on the three signal paths, and a second adaptive pre-processing filter on the fourth signal
path, the second adaptive pre-processing filter 10. A multi-channel echo according to item 8 or 9,
comprising a second adaptation pre-processing filter, wherein the adaptation of the filter is based
on the difference between the signals on the second signal path and the fourth signal path.
15-04-2019
15
Correction system.
11. The multi-function of claim 10, wherein the adaptive correction filter and the adaptive preprocessing filter are configured such that the adaptation of the pre-processing filter is performed
more slowly than the adaptation of the correction filter. Channel echo correction system.
12. The multi-channel echo correction system of claim 11, wherein the adaptation increment of
the correction filter is greater than the adaptation increment of the pre-processing filter.
(Item 13) The pre-processing means includes delay means on the first signal path and the second
signal path, and each delay means is arranged on the first signal path before reaching the
summing means. Item 12. The multi-channel echo correction system according to any one of
items 10 to 12, configured to delay a signal of B. and delay the signal on the second signal path
before reaching the subtracting means. .
14. The multi-channel echo correction system of claim 13, wherein the delay of the delay means
is selected such that the delay of the delay means corresponds to about half of the length of the
corresponding adaptive pre-processing filter.
(Item 15) Any one of the items 1 to 14, comprising summing means provided between the
adaptive correction filter and the microphone output channel and configured to add together the
signals emitted from the adaptive correction filter. Multi-channel echo correction system as
described in.
The at least one microphone may be configured as an array of at least two microphones
connected to beam forming means on the microphone output channel. Channel echo correction
system. 17. A method of correcting echo in a multi-channel system, the multi-channel system
comprising two speakers, each speaker being a speaker input for providing a speaker input
signal emitted from the speakers. Two speakers connected to the channel and at least one
microphone for obtaining the signal emitted from the speakers, the at least one microphone
being connected to the microphone output channel A correction channel for each speaker input
channel, each correction channel being a correction channel connected to an individual speaker
input channel and a microphone output channel, and a matching correction filter for each
correction channel, Out Each adaptive correction filter is configured to filter the signal on an
individual correction channel, such that a signal is provided to correct the microphone output
15-04-2019
16
signal to the signal emanating from the speaker. And the method comprises the steps of:
receiving a speaker input signal; wherein each speaker input signal is received on a correction
channel; and the speaker input signal on the correction channel. Pre-processing the steps of:
determining the correlation value of the two loudspeaker signals for the two loudspeakers
according to a predetermined criterion; and the determined correlation value passing a
predetermined threshold Pre-processing, including releasing one of the adaptive correction
filters. Method. (Item 18) The pre-processing includes providing a linear combination, in
particular, the difference and / or the sum of the input channels of the two speakers, to at least
one of the adaptive correction filters. The method according to item 17. (Item 19) The preprocessing step provides the sum of the signals of the first and second speaker input channels to
the first filter of the adaptive correction filter, and the first and second speaker inputs. 19. A
method according to item 17 or 18, comprising the step of providing the signal difference of the
channel to a second filter of the adaptive correction filter.
20. The method of claim 19, wherein the pre-processing comprises releasing a second filter of
the adaptive correction filter if the determined correlation value passes the predetermined
threshold. (Item 21) The step of pre-processing determines the correlation value according to a
predetermined standard based on the signal power of the signal, and in particular, represents the
difference and / or the sum of the signals of the input channels of the two speakers. 21. A
method according to any one of items 17-20, comprising a step. 22. The method of any one of
claims 17-21, wherein the correlation value is determined recursively. 23. The pre-processing
step comprises summing the speaker input signals on two correction input signals to obtain a
summed signal, and the two corrections to obtain a subtracted signal. The method according to
any one of items 17 to 22, comprising subtracting the speaker input signal on the input signal.
24. The method according to claim 23, wherein the pre-processing step comprises the step of
multiplying the summed signal and the subtracted signal by a weighting factor, in particular a
common weighting factor. 25. The pre-processing step may include filtering to match the
speaker signal of the first speaker input channel before being summed with the speaker signal of
the second speaker channel, the adaptation being A step based on the difference between the
loudspeaker signals of the first and second loudspeaker input channels, and the loudspeaker
signal of the second loudspeaker input channel before being subtracted from the loudspeaker
signal of the second loudspeaker channel Filtering to fit, wherein the fit is based on the
difference between the loudspeaker signals of the first and the second loudspeaker input
channels, including the steps of the method of. 26. The method of claim 25, wherein the step of
adaptively filtering in the pre-processing step is performed more slowly than the adaptation of
the adaptive correction filter. 27. The method of claim 25 or 26, wherein the adaptation
increment of the correction filter is selected to be greater than the adaptation increment of the
adaptively filtering step of the pre-processing step. 28. The pre-processing step includes delaying
the speaker signal of the second speaker input channel before being summed with the matched
filtered speaker signal of the first speaker input channel. And delaying the speaker signal of the
15-04-2019
17
first speaker input channel prior to subtracting the matched filtered speaker signal of the first
speaker input channel. The method described in.
29. The method of claim 28, wherein the delay is selected to correspond to about half the length
of the corresponding filter to filter to match in the pre-processing step. Method. 30. A method
according to any one of claims 17 to 29, comprising summing the signals emitted from the
adaptive correction filter. 31. One or more computer readable media having computer executable
instructions for performing the steps of the above method according to any one of items 17 to
30 when executed on a computer. Computer program product.
[0059]
The present invention relates to a multi-channel echo correction system, comprising two
loudspeaker input channels, each loudspeaker input channel being connected to a loudspeaker
providing a loudspeaker input signal emitted by the loudspeakers. A speaker input channel, a
microphone output channel connected to the at least one microphone for receiving a microphone
output signal from the at least one microphone, and a correction channel for the respective
speaker input channel, each microphone being the speaker A microphone output channel and a
correction channel for each of the speaker input channels, each correction channel being an
individual speaker input channel and the microphone output A correction channel connected to
the channel and a matching correction filter for each of the correction channels, such that a
correction output signal is provided to correct the microphone output signal relative to the signal
emitted from the speaker Wherein each said adaptive correction filter is adapted to filter the
signal on the respective correction channel, and a pre-processing means for pre-processing
speaker input signal on said correction channel, The pre-processing means determines a
correlation value of the speaker input signal to the two speakers according to a predetermined
criterion, and one of the adaptive correction filters if the determined correlation value passes a
predetermined threshold. And a pre-processing means configured to release
[0060]
FIG. 1 schematically shows an embodiment of a multi-channel echo correction system according
to the invention.
In this example, a stereo signal source such as a car radio (not shown) outputs radio signals to
the left speaker channel xL (n) and the right speaker channel xR (n). These radio signals are
emitted by two loudspeakers 102 and obtained by a microphone array consisting of three
15-04-2019
18
microphones 101. The transmission from the loudspeaker to the microphone may be described
in finite impulse response, for example in a car in a car or the like.
[0061]
The variable n indicates the time dependence of the coefficients.
[0062]
The signal obtained by the microphone 101 is output to the microphone output channel 103
where processing means 104 is provided.
The processing means may perform linear time invariant processing, for example, as done by a
beamformer or high pass filter. The microphone output signal (in the example after processing
means 104) is denoted d (n).
[0063]
Microphones are mainly used to obtain speech signals from speakers, for example for hands-free
telephone systems or for special-device hands-free control systems. With this in mind, it is
desirable to reduce the radio signal component in the microphone output signal d (n). In order to
reduce these components, two adaptive correction filters 105 and 106 are provided, whose
impulse responses are
[0064]
Given by
[0065]
In general, the order of the adaptive filter N is smaller than the order of the impulse response.
As an example, 300 to 500 coefficients at a sampling rate of 11 kHz may be used for the
15-04-2019
19
adaptive filter. The adaptive filter provides a correction signal that is removed from the
microphone output signal d (n). The result is sometimes a signal e (n), called the error signal,
which takes the form
[0066]
This output signal is used to adapt the adaptive correction filter. Filter fit is estimated impulse
response
[0067]
Are performed as close as possible to the true impulse responses hL, i (n) and hR, i (n), and in a
way that a large number of coefficients are estimated. Filter adaptation may be performed via the
NLMS algorithm.
[0068]
In the example of FIG. 1, each speaker input channel 111 and 112 is connected to the
microphone output channel 103 by a corresponding correction channel 113 and 114 such that
each correction channel receives an individual speaker input signal. The pre-processing means
110 comprises two inputs 115 and 116, each input being connected to an individual speaker
input channel via a correction channel. The pre-processing means 110 further comprises two
outputs 117 and 118, which are connected to the respective adaptive correction filters 105 and
106.
[0069]
In the pre-processing unit 110, there is a first signal path 119 connecting the first input unit 115
and the first output unit 117, and further, the second input unit 116 and the second output unit
118 are connected. There is a second signal path to In the first signal path 119 subtraction
means 121 and in the second signal path 120 summing means 122 are provided.
15-04-2019
20
[0070]
The third signal path 123 connects the first input 115 and the summing means 122, while the
fourth signal path 124 connects the second input 116 and the second subtracting means 121.
On the first and second signal paths, after the summing means and subtracting means, a
multiplying means 125 follows and multiplies the summed signal and the subtracted signal by
the common weighting factor of 0.5. As a result, the following linear combination of the
loudspeaker input channels originates from the preprocessing means at the outputs 117 and
118:
[0071]
In a conventional signal processor, weighting by a factor of 0.5 can be realized by shifting the
result in the accumulation register with one bit.
[0072]
FIG. 2 shows the change in time of a typical stereo radio signal.
The upper graph corresponds to the left speaker input channel, and the lower graph corresponds
to the right speaker input channel. In the left part of each graph the power spectrum for the
reproduction of news is represented, the right part corresponds to the reproduction of classical
music.
[0073]
FIG. 3 shows the signals xS (n) and xD (n) corresponding to the signals of FIG. The upper graph
shows xS (n) weighted by an element of 0.5, and the lower graph shows xD (n) weighted by an
element of 0.5. Since the signal corresponding to the reproduction of the news is usually a
monaural signal, the difference signal xD (n) disappears, or at least almost disappears in this
period. In classical music, the summed and subtracted signals differ only slightly from the
original signal.
[0074]
15-04-2019
21
Returning to FIG. 1, as soon as the difference signal xD (n) has a power spectrum below a
predetermined threshold, the corresponding adaptive correction filter 105 is released. The
adaptation of this adaptive correction filter is then not performed such that the computing power
required to adapt the filter is halved. However, although only one of the adaptive correction
filters is enabled, the resulting echo reduction is not significantly altered.
[0075]
This is illustrated in FIG. In FIG. 3, the change of the microphone output signal d (n) in time is
shown without further echo correction. Furthermore, although the conventional echo correction
shown in FIG. 6 has been performed, the output signal e (n) is represented as an example for
comparison. Furthermore, the graph also shows the short-term output of the output signal e (n)
for which the echo correction according to FIG. 1 has been performed. However, for better
illustration, 4 dB is added to the value of the last curve. In other words, without the additional 4
dB, the curves of the output signal using conventional and new echo correction methods are
almost indistinguishable.
[0076]
It should be pointed out that the echo correction resulting from the new method is almost the
same as that of the conventional method, and the required computing power is considerably
reduced. As can be seen, after some adjustment, the radio signal can be reduced to around 30 dB.
[0077]
In FIG. 1, the pre-processing means 110 and the adaptive filters 105 and 106 are shown as
separate elements. The control of the adaptive filter by the pre-processing means, in particular
the release of one of the adaptive filters, has to be carried out via the corresponding control
connection (not shown). However, it should be understood that checking whether the correlated
value is less than a predetermined threshold may also be performed at the input of the adaptive
filter. In this case, the pre-processing means also include part of the adaptive filter.
15-04-2019
22
[0078]
Determining the correlation value according to a predetermined criterion may be performed by
determining the squared norm (ie at the signal power) of the signal vector at the output of the
pre-processing means or at the input of the adaptive correction filter. This can be done in a
recursive manner.
[0079]
The corresponding release variables aS (n) and aD (n) may be set to 0 if the norm determined in
this way is less than a predetermined threshold.
[0080]
For example, the predetermined threshold may be set to 0.03.
The determination of the output signal e (n) after subtraction of the correction signal is then:
[0081]
In this equation, summation (corresponding to convolution) is determined only if the
corresponding cancellation variable is non-zero. Correspondingly, the adaptation of the adaptive
correction filter is performed only under these conditions.
[0082]
The conventional echo correction method according to FIG. 6 has the following convergence
behavior for the adaptive correction filter:
[0083]
Under this condition, the signals are not perfectly correlated.
15-04-2019
23
The new echo correction method according to FIG. 1 has the following focusing behavior, even if
the input signals are not completely correlated.
[0084]
Thus, there is no uniqueness even with the new method.
[0085]
An extension of the multi-channel echo correction system shown in FIG. 1 is schematically
represented in FIG.
In FIG. 5 the structure of the corresponding preprocessing means is shown.
[0086]
This embodiment is particularly useful when, for example, in an interview, one of the speakers is
placed (voicefully) to the left and the other is placed (voiceally) to the right. This can be done by
changing the amplification of the left and right channels, or by inserting delay time components,
or a combination of both. Additionally, a filter that modifies the tone of the beep may be used.
[0087]
In such cases, a modified pre-processing means as shown in FIG. 5 is advantageous. In this
embodiment, the output signal of the preprocessing means has the following equation:
[0088]
Here, further adaptive pre-processing filters 526 and 527 are provided along the third signal
path 523 and the fourth signal path 524 individually. In addition, on the first signal path 519
15-04-2019
24
and the second signal path 520, delay means 528 and 529 are provided individually. These delay
means are provided before the summing and subtracting means in the direction of the signal
flow.
[0089]
The adaptation of preprocessing filters 526 and 527 may be performed via the NLMS algorithm,
as in the case of the correction filter. First of all, two error signals for the fit are determined as
follows.
[0090]
Filter adaptation is then performed according to the following equation:
[0091]
Preferably, the adaptation of the preprocessing filters 526 and 527 is performed more slowly
than the adaptation of the correction filters 105 and 106.
This may be achieved, for example, by selecting smaller increments in the case of adaptation of
the pre-processing filter.
[0092]
If the filters 526 and 527 do not always converge to the causal optimum solution, then delay
elements 528 and 529 are selected in such a way that the delay time of the NV cycle is selected
such that about half of the corresponding filters reproduce the noncausal part. Can be
configured.
[0093]
In the case of a mono signal (xL (n) = xR (n)), both filters focus on the optimal solution using the
transfer function.
[0094]
15-04-2019
25
The signal output by the preprocessing means then has the following equation:
[0095]
Thus, for NV cycle delay, the adaptive pre-processing of FIG. 5 corresponds to the fixed preprocessing shown in FIG.
[0096]
In the above example, the multi-channel echo correction system represents the case of a twochannel system (e.g. stereo).
However, extensions to more than one channel are also possible.
In this case, for example, the pre-processing means may be configured to release all but one of
the adaptive filters if the correlation value passes a predetermined threshold.
[0097]
Further modifications and health of the present invention will be apparent to those skilled in the
art in view of the present description.
Accordingly, the description is to be construed as illustrative only and is for the purpose of
teaching those skilled in the art the general manner of carrying out the present invention.
It is to be understood that the forms of the invention shown and described herein are to be taken
as the presently preferred embodiments.
[0098]
1 shows the structure of a multi-channel echo correction system. Fig. 6 shows an example of the
15-04-2019
26
change in time of a radio signal on two speaker input channels. The change in time of the sum
and difference of the radio signal is shown. 7 shows the structure of another multi-channel echo
of different signals. 7 shows the structure of another multi-channel echo correction system. 1
shows the structure of a prior art multi-channel echo correction system.
Explanation of sign
[0099]
DESCRIPTION OF SYMBOLS 101 Microphone 102 Speaker 105, 106 Conformation correction
filter 110 Preprocessing means 111, 112 Speaker input channel 113, 114 Correction channel
115 1st input part 116 2nd output part 117 1st output part 118 2nd output part 119 First
signal path 120 Second signal path 121 Subtracting means 122 Combining means
15-04-2019
27
Документ
Категория
Без категории
Просмотров
0
Размер файла
40 Кб
Теги
description, jp2007288775
1/--страниц
Пожаловаться на содержимое документа