JP2000172297A - Signal extraction method and apparatus, and medium recording signal extraction program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To extract in real-time only a voice signal component from a signal of one channel that a voice signal and a non-voice signal are mixed by using the signal shifting an input signal by a fixed time and using the high band emphasized signal. SOLUTION: A delayed signal xj is supplied to an adaptive filter 1, and its filter coefficient Wj is successively updated by an output from an adaptive filter coefficient successive renewal part 2. The output signal zj of the adaptive filter 1 is applied to a reduction terminal of a subtracter 3, and is subtracted from the signal yj as it is applied to the non-reduction terminal of the subtracter 3, and the output of the subtracter 3 becomes an output signal ej of an adaptive processing circuit. The output signal ej of the adaptive processing circuit and respective tap output signals xj of the adaptive filter 1 are supplied to the adaptive filter coefficient successive renewal part 2, and the filter coefficient is updated successively based on algorithm by a learning identifying method to be outputted. The successive renewal of the filter coefficient is performed so that a difference ej between the output signal zj and a reference signal yj becomes small.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for picking up a sound signal, and when an unnecessary sound signal is mixed and picked up for some reason, a desired sound signal is extracted from the picked-up signal. The present invention relates to a signal extraction method and apparatus for extracting only a signal in real time and a medium recording a signal extraction program.

[0002]

2. Description of the Related Art As a conventional noise removing method, a so-called one-input method and a two-input method described below are known. First, as a one-input method, for example,
SFBoll, “Suppression of Acoustic Noise in Speech
Using Spectral Subtraction ”IEEE trans., Vol. ASS
P-27, No. 2, pp. 113-120, April (1979). This method is a method of obtaining a short-time amplitude spectrum of a desired voice signal by subtracting a short-time power spectrum of noise known or estimated from a silent section from a spectrum of an input signal.

Next, as a method of the two-input method, for example,
Reference B. Widrow et al., “Adaptive Noise Cancelling
: Principles and Applications ”, Proc.IEEE, Vol. 6
3, No. 12, pp. 1692-1716 (1975). This method is a method capable of removing noise when only a noise component can be separately used as a reference signal in addition to an input signal in which speech and noise are mixed. Since the sound collection system of the reference signal is generally different from the sound collection system of the input signal, the reference signal does not match the noise component included in the input signal, and to extract the audio signal from the input signal, It is necessary to perform a filter process for matching the reference signal and the noise component included in the input signal as much as possible to obtain a difference. Generally, an adaptive filter is used for this filter.

That is, this adaptive signal processing method is a signal processing method using an adaptive filter. As a control algorithm for the adaptive filter, the above-mentioned document B. Widrow et al.,
“Adaptive Noise Cancelling: Principles and Appl
ications ”, Proc.IEEE, Vol. 63, No. 12, pp. 1692-1716
(1975), the literature J. Nagumo et al.,
“A Learning Method for System Identification”, I
EEE Trans., Vol.AC-12, No.3, pp. 282-287 JUNE (196
There is a learning identification method described in 7). Here, an adaptive filter is a kind of learning function that, when an input signal and a signal to be output are given, the coefficients of the adaptive filter are automatically and sequentially corrected, and a signal closer to the desired output is gradually output. It is a filter provided with. Each of the algorithms is an algorithm in which the difference between the filter output and the input is fed back to the filter to converge the value of the filter coefficient, and the learning identification method is LM in that the difference value is normalized.
It is different from S method. The LMS method has the advantage that it is simple and requires a small amount of calculation, so that it is easy to implement. On the other hand, since the difference is not normalized, the dependence on the input signal amplitude is large and the convergence is slow and fast adaptation is performed. Not suitable for the required application. On the other hand, the learning identification method is an improved version of the LMS method, and the amount of calculation for normalization slightly increases, but the convergence is much faster than the LMS method.

[0005]

[0004] Broadcasting, especially news-related materials brought in from the scene, have poor S / N ratios, such as the voice of the kidnapper in a kidnapping case and the voice of speech in the cockpit of an airplane. Not one input signal form. In times of urgency, there is a need to remove noise from such signals in real time,
As described below, the prior art could not meet such a demand.

In the conventional spectral subtraction method, a silent section identified in advance is cut out and used to remove noise. Therefore, when the noise component fluctuates with time, adaptability to real-time processing is poor, such as re-detecting a silent section each time. That is, in the conventional one-input method, it is difficult to adaptively adjust parameters, and it is difficult to perform filtering while following only a desired signal component temporally with respect to an input signal of one channel that changes every moment. is there.

Further, in the above-described adaptive signal processing method of the two-input method, although real-time processing can be performed, in this case, since only one channel signal in which voice and noise are mixed is given, it cannot be used. . That is, when trying to follow in time by adaptive signal processing, a signal including only a noise component is required as a reference signal.

As described above, the noise eliminator capable of coping with the noise component fluctuating with time in the case of one input signal form, and further confirming the effect so as to be in the most audible state. It is desired to develop a device capable of adjusting parameters in real time.

Accordingly, an object of the present invention is to provide a method and apparatus capable of adjusting parameters in real time and extracting only an audio signal component from a one-channel signal in which an audio signal and a non-audio signal are mixed, and a signal extraction. It is to provide a medium on which a program is recorded.

[0010]

In order to achieve the above object, the present invention is based on the fact that the adaptive signal processing method used in the conventional two-input method is used as it is. Now, paying attention to the fact that the nature of temporal correlation is different, according to the content of the non-speech signal that can be judged by audibility, a signal obtained by shifting the input signal by a certain time is used as a necessary reference signal. Thus, a signal whose high frequency range is emphasized is used as an input signal.

The former method (using a signal obtained by shifting the input signal by a predetermined time as a necessary reference signal) utilizes the fact that the musical sound has a higher temporal correlation with the voice, and the input signal is different from the musical sound. This is based on the idea that in the case of voice, a musical component having a high temporal correlation is extracted by an adaptive filter process, and this is subtracted from the input signal to extract a voice having a lower correlation.

The latter (in addition to the former, a method in which a high-frequency emphasized signal is used as an input signal) utilizes the fact that speech has a higher temporal correlation between noise and speech, and thus has a higher temporal correlation. By extracting a speech component by an adaptive filter process and using a signal whose high-frequency component has been emphasized in advance as an input signal, it is intended to simultaneously achieve high-frequency emphasis of a speech having a small high-frequency component.

In any case, by sequentially updating the coefficients of the adaptive filter while controlling the convergence speed, only the desired signal component temporally follows the input signal of one channel that changes every moment. While filtering is possible.

Further, in the former, a component having a higher temporal correlation is extracted from an input signal and a signal obtained by delaying the input signal by a predetermined time, and the extracted component is subtracted from the input signal. In the latter, a system for determining a matched filter coefficient in real time by adaptive filter processing from an output signal obtained by subjecting an input signal to high-frequency emphasis processing and a signal obtained by delaying the input signal by a predetermined time, and a determined filter coefficient And a system for filtering and outputting an input signal based on the delay time of a delay circuit for delaying the input signal for a predetermined time, the number of taps of a one-dimensional adaptive filter, a convergence coefficient, and a high value in the latter case. It is characterized in that each of the band emphasis signal gains can be adjusted in real time so that it is most audible.

That is, according to the signal extraction method of the present invention, a desired signal having a low temporal correlation is obtained from a digital input signal of one channel in which a desired signal having a low temporal correlation and an undesired signal having a high temporal correlation are mixed. A method for extracting a signal, wherein the digital input signal is branched into two, and one of the two branched digital input signals is delayed by a designated number N of samples, and delayed by the number N of samples. A digital input signal is input to an adaptive filter having a specified number of taps k at one sample interval, and an output signal of the adaptive filter is set to a set of each tap output signal of the adaptive filter and a set of filter coefficients to be sequentially updated. , And subtracts the result of the computation from the other of the two branched digital input signals. Output as a desired signal having low spatial correlation, and in synchronization with the sequential input of the digital input signal, the subtraction result, the set of tap output signals of the adaptive filter, the set of filter coefficients, and the designated convergence coefficient Based on μ, successively updating the set of filter coefficients by newly generating the set of filter coefficients so that the subtraction result becomes smaller, and repeatedly performing the calculation, the subtraction, and the output. It is assumed that.

Further, the signal extraction method of the present invention is characterized in that a desired signal having a high temporal correlation is obtained from a one-channel digital input signal in which a desired signal having a high temporal correlation and an undesired signal having a low temporal correlation are mixed. A signal extraction method for extracting, wherein the digital input signal is branched into three, and a first signal of the three-branched digital input signal is subjected to high-frequency emphasis processing and a designated gain multiplication to perform high-frequency emphasis. And outputs a second signal of the digital input signal divided into three.
Is delayed by a designated number of samples M, a third signal of the three-divided digital input signal is delayed by a designated number of samples L, and the digital input signal delayed by the number of samples M is delayed. Input to an adaptive filter consisting of a specified number of taps k at one sample interval, and output the output signal of the adaptive filter using a set of tap output signals of the adaptive filter and a set of filter coefficients that are sequentially updated. 1 based on the operation expression, and subtracts the operation result based on the first operation expression from the high frequency emphasis signal,
Based on each set of tap output signals of the adaptive filter, the set of filter coefficients, and the designated convergence coefficient μ, the set of filter coefficients is newly generated so that the subtraction result is reduced, and The set is sequentially updated, the digital input signal delayed by the number of samples L is input to a matched filter having the same configuration as the adaptive filter, and the output signal of the matched filter is set to the set of each tap output signal of the matched filter. The adaptive filter that is operated based on a second operation expression using the set of filter coefficients and outputs an operation result based on the second operation expression as the desired signal having a high temporal correlation, and the sequentially updated adaptive filter Is copied to the set of filter coefficients of the matched filter, and synchronized with the sequential input of the digital input signal, Calculation based on the first calculation formula, the subtraction, the sequential update of the set of filter coefficients of the adaptive filter,
The operation based on the second operation expression and the copying of the set of the output and the filter coefficient are repeatedly performed.

Further, the signal extracting apparatus of the present invention converts the desired signal having a low temporal correlation from a digital input signal of one channel in which a desired signal having a low temporal correlation and an undesired signal having a high temporal correlation are mixed. A signal extraction device for extracting, said digital input signal being divided into two by a branching means, and one of the digital input signals bifurcated by the bifurcating means is inputted, and the signal is divided into a designated number of samples N
A delay means for delaying the output signal by an amount corresponding to the number of taps and an output signal of the delay means as an input signal, the adaptive filter comprising k taps designated at one sample interval. An adaptive filter that performs an operation based on a predetermined operation expression using a set of tap output signals of each of the above and a set of filter coefficients that are sequentially updated, and outputs the operation result, Subtracting means for subtracting from the other signal of the digital input signal branched into two by the branching means, and outputting the subtracted result as a desired signal having a low temporal correlation, and synchronizing with the sequential input of the digital input signal Based on the output signal of the subtraction means, the set of tap output signals of the adaptive filter, the set of filter coefficients, and the designated convergence coefficient μ. Wherein the set of filter coefficients is newly generated so that small and is characterized in by comprising comprising at least an adaptive filter coefficient successively updated unit successively updates said set of filter coefficients.

The signal extracting apparatus according to the present invention further comprises a block dividing means for dividing the digital input signal into blocks for every F samples at an input end of the apparatus, and the other of the two branched digital input signals. A first buffer unit having a storage capacity of at least F samples interposed between the two-branching unit and the subtracting unit of the signal system, and sequentially outputs an output signal of the block dividing unit in units of the blocks. (F-1) counting from a first predetermined sample predetermined according to the storage capacity of the first buffer means stored in the first buffer means after the storage, and The storage sample before the sample is input to the subtraction means as the other signal of the digital input signal which is divided into two, and the next following the storage in the block unit Until the storage in the lock unit, a total of F samples are shifted from the storage sample before (F-1) samples by one sample in the direction of the first predetermined sample, counting from the first predetermined sample. A first buffer means for repeatedly inputting samples to the subtraction means, and a second buffer means having a storage capacity of at least (N + k + F) samples instead of the delay means, wherein an output signal of the block dividing means is provided for the block. The second buffer means is sequentially stored in units and the second predetermined buffer is stored in the second buffer means at the same timing as the first predetermined sample after the storage in accordance with the storage capacity of the second buffer means. K times from the stored sample (N + k + F-1) samples before the second predetermined sample counting from the predetermined sample of A set of successive samples is input to the adaptive filter as each tap output signal of the adaptive filter, and the time counted from the second predetermined sample until the storage of the next block after the storage of the block is counted. The total of F sets of the k consecutive samples are shifted in synchronism with the first buffer means by shifting one sample at a time from the storage sample before (N + k + F-1) samples in the second predetermined sample direction. And a second buffer means for repeatedly inputting each tap output signal of the adaptive filter to the adaptive filter.

The signal extracting apparatus according to the present invention is characterized in that the signal extracting apparatus further comprises parameter control means for designating the number N of samples of the delay amount, the number k of taps, and the convergence coefficient μ. It is assumed that.

Further, the signal extracting apparatus of the present invention converts the desired signal having a high temporal correlation from a one-channel digital input signal in which a desired signal having a high temporal correlation and an undesired signal having a low temporal correlation are mixed. A signal extraction device for extracting, wherein a three-branch means of the digital input signal and a first signal of a digital input signal branched into three by the three-branch means are input, and the signal is designated as high-frequency emphasis processing. High-frequency emphasizing means for applying a multiplied gain and outputting the signal as a high-frequency emphasizing signal, and a second signal of the digital input signal branched into three by the three-way branching means. First delay means for delaying and outputting the third signal, and a third signal of a digital input signal branched into three by the three-branch means, and delaying the signal by a designated number of samples L And an adaptive filter having an output signal of the first delay means as an input signal and having the number of taps k specified at one sample interval, and outputting an output signal of the adaptive filter. An adaptive filter that performs an operation based on a first operation expression using a set of tap output signals of the adaptive filter and a set of filter coefficients that are sequentially updated, and outputs the output signal of the adaptive filter to the output of the high-frequency emphasis unit. Subtraction means for subtracting from the signal, and as a result of the subtraction by the subtraction means, the subtraction result is reduced based on each set of tap output signals of the adaptive filter, the set of filter coefficients, and the designated convergence coefficient μ. An adaptive filter coefficient successively updating means for newly generating the set of filter coefficients and successively updating the set of filter coefficients; and a filter to which an output signal of the second delay means is inputted. A matched filter, wherein the output signal of the matched filter is operated based on a second operation expression using a set of each tap output signal of the matched filter and a set of filter coefficients, and an operation based on the second operation expression At least a matched filter that outputs a result as the desired signal having a high temporal correlation, and a unit that copies a set of filter coefficients of the adaptive filter that is sequentially updated to a set of filter coefficients of the matched filter. , In synchronization with the sequential input of the digital input signal, an operation based on the first operation expression, the subtraction,
It is characterized in that the updating of the set of adaptive filter coefficients, the operation based on the second operation expression, the output, and the copying of the set of filter coefficients are repeatedly performed.

In the signal extracting apparatus according to the present invention, the high-frequency emphasis means generates a third delay means for delaying the input signal by one sample, and a difference signal between input and output signals of the third delay means. And a multiplying means for multiplying an input signal connected to a stage before the third delay unit or a stage after the differential unit by the gain.

Further, the signal extracting apparatus of the present invention further comprises a block dividing means for dividing the digital input signal into blocks for every F samples at an input end of the apparatus, and at least (in place of the third delay means) F + 1)
A first buffer unit having a storage capacity of a sample, wherein the output signal of the block division unit or the output signal of the block division unit multiplied by the gain is sequentially stored in the block unit, and after the storage, The first predetermined time is counted from a first predetermined sample which is predetermined according to the storage capacity of the first buffer means and stored in the first buffer means, and the first predetermined time is counted from a storage sample before F samples in time. A set of two consecutive samples toward the sample is output as an input signal to the difference means, and the time counted from the first predetermined sample is counted from the first predetermined sample to the storage of the next block after the storage of the block. A total of F sets of the two consecutive samples shifted from the storage sample before F samples by one sample in the first predetermined sample direction. A first buffer means for outputting the repeated the difference means, first it has a storage capacity of at least (M + k + F) sample replaces the first delay means 2
Buffer means for sequentially storing output signals of the block dividing means in units of the blocks, and storing the output signals in the second buffer means at the same timing as the first predetermined sample after the storage. Counting from the second predetermined sample predetermined according to the storage capacity of the second buffer means and starting from the storage sample before (M + k + F-1) samples k times toward the second predetermined sample A set of consecutive samples is input to the adaptive filter as each tap output signal of the adaptive filter, and the time counted from the second predetermined sample until the storage of the next block following the storage of the block is counted. (M
+ K + F-1) The k adaptively repeated sets of the k consecutive samples are shifted in synchronization with the first buffer means while shifting in the second predetermined sample direction one sample at a time from the storage sample before the sample. A second buffer means for inputting each tap output signal of the filter to the adaptive filter, and a third buffer means having a storage capacity of at least (L + k + F) samples instead of the second delay means, wherein the block division Means for sequentially storing the output signals of the means in block units, and after the storage, storing the output signals at the same timing as the first predetermined sample.
Counting from a third predetermined sample predetermined according to the storage capacity of the third buffer means stored in the third buffer means and temporally (L + k + F-1) samples from the storage sample before the third sample. A set of k consecutive samples toward a predetermined sample is input to the matched filter as each tap output signal of the matched filter, and the third block is stored until the next block is stored after the block is stored. The total of F sets of the k consecutive samples are shifted from the storage sample before (L + k + F-1) samples one sample at a time in the third predetermined sample direction by counting from the predetermined sample. 1
And a third buffer means for repeatedly inputting each tap output signal of the matched filter to the matched filter in synchronization with the buffer means.

The signal extracting apparatus according to the present invention is further characterized by further comprising parameter control means for respectively specifying the number of samples M and L of the delay amount, the number of taps k, the convergence coefficient μ and the gain. Is what you do.

The medium on which the signal extraction program of the present invention is recorded is a computer in which a desired signal having a low temporal correlation and an undesired signal having a high temporal correlation are mixed.
A medium for recording a program for extracting a desired signal having a low temporal correlation from a digital input signal of a channel, the signal extracting program comprising: a digital input signal delayed by a designated number N of samples; The output signal of the adaptive filter is successively updated with a set of tap output signals of the adaptive filter to an adaptive filter having the number of taps designated at one sample interval and configured in a computer having the signal as an input signal. Using a set of filter coefficients to perform an operation based on a predetermined operation expression, subtracting the operation result from the digital input signal, outputting the subtraction result as the desired signal having a low temporal correlation, and In synchronization with the successive input of the adaptive filter, the set of tap output signals of the adaptive filter,
Based on the set of filter coefficients and the designated convergence coefficient μ, successively updating the set of filter coefficients by newly generating the set of filter coefficients such that the result of the subtraction is small; and The subtraction and the output are repeatedly performed.

The medium on which the signal extraction program of the present invention is recorded is a computer in which a desired signal having a low temporal correlation and an undesired signal having a high temporal correlation are mixed.
A medium for recording a program for extracting a desired signal having a low temporal correlation from a digital input signal of a channel, the signal extracting program comprising: a computer for dividing the digital input signal into blocks for every F samples; The signals are divided and the blocks are sequentially stored in the block units. After the block units are stored, counting is performed from the sample at a predetermined position in the stored block based on the designated number of samples N and the number of taps k. And a set of k consecutive samples in the time direction from the storage sample before (N + k + F-1) samples in time, and (F−
1) One stored sample before the sample is read, and the number of taps k specified at one sample interval based on a predetermined arithmetic expression using the set of read samples and the set of filter coefficients that are sequentially updated is used. An output signal of the adaptive filter is calculated, the calculation result is subtracted from the read sample, the subtraction result is output as a desired signal having a low temporal correlation, and the signal is synchronized with the sequential input of the digital input signal. And generating a new set of filter coefficients based on the subtraction result, the set of read samples, the set of filter coefficients, and a designated convergence coefficient μ such that the subtraction result is reduced. The set of coefficients is successively updated, and the set of k consecutive samples is stored between the block-based storage and the next block-based storage. And a set of a total of F samples and a read of one sample while shifting the one sample by one sample sequentially in the time direction, and a set of the set of the operation, the subtraction, the output, and the filter coefficient. It is characterized in that successive updates are repeatedly performed.

The medium on which the signal extraction program of the present invention is recorded is a computer in which a desired signal having a high temporal correlation and an undesired signal having a low temporal correlation are mixed.
A medium storing a program for extracting a desired signal having a high temporal correlation from a digital input signal of a channel, the signal extracting program comprising: a digital input signal delayed by a designated number M of samples; The output signal of the adaptive filter is successively updated with a set of tap output signals of the adaptive filter to an adaptive filter having the number of taps designated at one sample interval and configured in a computer having the signal as an input signal. The first arithmetic expression is calculated from a high-frequency emphasized signal obtained by performing a high-frequency emphasis process and a designated gain multiplication on the digital input signal using the set of filter coefficients. The result of the subtraction, the set of tap output signals of the adaptive filter, the set of filter coefficients, and the specified Based on the coefficient μ, the set of filter coefficients is newly generated so that the result of the subtraction is reduced, and the set of filter coefficients is sequentially updated. The digital input signal delayed by the specified number of samples L is input to the input signal. In a matched filter having the same configuration as that of the adaptive filter configured in the computer, a second operation is performed using the set of tap output signals of the matched filter and the set of filter coefficients using the output signal of the matched filter. An arithmetic operation is performed based on an expression, an arithmetic result based on the second arithmetic expression is output as a desired signal having a high temporal correlation, and a set of filter coefficients of the adaptive filter that is sequentially updated is determined by the matched filter. Let it be copied to a set of filter coefficients,
In synchronization with the sequential input of the digital input signal, the operation based on the first operation expression, the subtraction, the successive update of the set of filter coefficients of the adaptive filter, the operation based on the second operation expression, the output and The method is characterized in that copying of the set of filter coefficients is performed repeatedly.

The medium on which the signal extraction program of the present invention is recorded is a computer in which a desired signal having a high temporal correlation and an undesired signal having a low temporal correlation are mixed.
A medium storing a program for extracting a desired signal having a high temporal correlation from a digital input signal of a channel, the signal extracting program comprising: a computer for dividing the digital input signal into blocks for every F samples After dividing the signal into blocks, the signals are sequentially stored in block units, and after storing in block units, counting is performed from a sample at a predetermined position in the stored block based on the designated number of samples M and the number of taps k. The first set of k samples in the time direction from the stored sample before (M + k + F-1) samples in time, the number of samples L specified and the number of samples from the predetermined position are counted based on the k. Is a set of k second samples consecutive in the time direction from the storage sample before (L + k + F−1) samples. And reading out a set of two consecutive third samples in the time direction from the stored sample temporally before F samples counted from the sample at the predetermined position, and sequentially updating the read first set of samples. And a set of filter coefficients to be used,
An output signal of the adaptive filter having the number of taps k specified at the sample interval is calculated, and a calculation result based on the first calculation expression is multiplied by a specified gain to form the third sample set 2 Subtraction from a difference value between two samples, and the filter is configured to reduce the subtraction result based on the subtraction result, the read first set of samples, the set of filter coefficients, and a designated convergence coefficient μ. A set of coefficients is newly generated to sequentially update the set of filter coefficients, and the same as the adaptive filter based on a second operation expression using the read second set of samples and the set of filter coefficients. The output signal of the matched filter having the configuration is calculated, the calculation result based on the second calculation expression is output as the desired signal having a high temporal correlation, and the storage is continued in the block unit. A total of F sets of the first set of samples, the second set of samples, and the third set of samples are sequentially shifted by one sample in the time direction until the storage of each block unit. , And repeatedly performing the calculation, the subtraction, the output, and the successive update of the set of filter coefficients of the adaptive filter.

[0028]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail below based on embodiments of the present invention with reference to the accompanying drawings. [First Invention] FIG. 1 shows a case where a desired signal (for example, voice) having low temporal correlation and an undesired signal (for example, voice) having high temporal
4 is a flowchart showing a signal processing procedure according to the signal extraction method of the present invention for extracting a desired signal having a low temporal correlation from a digital input signal of one channel mixed with a musical tone.

In FIG. 1, a signal (sound) having a low temporal correlation and a signal (tone) having a high temporal correlation are mixed.
First, a channel signal (analog signal) is input (step S1). This signal is, for example, an audio signal including a musical tone picked up by a microphone. This signal is first digitized for the adaptive processing described below (step S2). The digitized signal is branched into two as shown in the figure, one of which is left as it is, and the other digital signal is a designated number of samples (delay amount) N
(Step S3). These intact and delayed signals are supplied to an adaptive processing circuit whose one embodiment is shown in FIG. 2, and the adaptive processing circuit performs adaptive processing as described below (step S4).

As shown in FIG. 2, the adaptive processing circuit includes an adaptive filter (constituted by a transversal filter in the illustrated example) 1 with the number of taps k specified at one sample interval, and a coefficient of the adaptive filter. It comprises an adaptive filter coefficient successively updating unit 2 for successively updating and a subtractor 3.

In the above configuration, the signal x _j delayed in step S3 is applied to the adaptive filter 1 (see FIG. 1).
Is supplied and its filter coefficients (coefficient vector

[Outside 1] ) Are successively updated by the output from the adaptive filter coefficient successively updating unit 2. Note that the suffix j indicates that it is at time j. The output signal z _j of the adaptive filter 1 is applied to the subtraction terminal of the subtracter 3 and the above-described signal (hereinafter, referred to as “the following”) applied to the minuend terminal of the subtracter 3.
A reference signal) is subtracted from y _j (see FIG. 1), and the output of the subtracter 3 is the output signal e _{j of the} adaptive processing circuit shown in FIG.
(See FIG. 1).

The adaptive filter coefficient successively updating unit 2 outputs the output signal e _j of the adaptive processing circuit and each tap output signal (input vector) of the adaptive filter 1.

[Outside 2] ) Is supplied, and a filter coefficient (coefficient vector [1]) is obtained based on an algorithm based on a learning identification method described below.
Are sequentially updated and output.

In the following, in mathematical expressions and the like, it is assumed that uppercase letters of the alphabet represent vectors, and lowercase letters represent scalar quantities. An algorithm for successively updating filter coefficients by the learning identification method is represented by the following equation (1).

(Equation 1) Here, [External 2]: input signal vector to the adaptive filter [External 1]: filter coefficient vector of the adaptive filter mu: convergence coefficient z _j: output signal y _j of the adaptive filter: reference signal e _j: the adaptive processing unit Output signal

[Outside 3] : Dot product of coefficient vector and input signal vector

[Outside 4] : Norm of input signal vector

The meaning of the equation (1) is explained as follows. In the subtracter 3 subtracts the output signal z _j of the adaptive filter 1 from the input signal (reference signal) y _j to the minuend terminal of the subtracter 3, the result of the subtraction e _j, convergence coefficient μ and

[Outside 5] And a filter coefficient vector at the next time point (j + 1) is obtained by adding the filter coefficient vector

[Outside 6] And In addition,

[Outside 7] Is 0, [outside 5] is forcibly set to 0. The initial value of the filter coefficient vector

[Outside 8] May be any value, but is generally a zero vector.

Further, the difference e _j between successive updating the output signal z _j and the reference signal y _j of the adaptive filter of the filter coefficients
Is updated so that the difference e _j becomes small because there is a time lag of the number N of samples between the input signal x _j of the adaptive filter and the reference signal y _j. Determining a filter coefficient is equivalent to performing a filtering process for enhancing a signal component having a high temporal correlation. The degree of the temporal correlation can be controlled by the delay amount N and the number of taps k as parameters.

Further, it is necessary that the convergence of the filter coefficient be fast in order to enable the real-time processing of the sequential update. The convergence coefficient μ ( 0 ≦ μ ≦
Control is performed according to 1). These parameters including the delay amount N and the number of taps k can be set in the parameter control (step S5) of FIG.

The above parameters (N, k, μ) are set as follows:
While listening to the sound obtained from the device of the present invention, the sound is adjusted manually so that it is most audible. At this time, the output signal z _j of the adaptive filter 1 (see FIG. 2) has a shape close to the spectrum of the musical tone.

In FIG. 1, D / A conversion (step S
6) and each step of audio output (step S7)
This is a procedure for converting a digital signal in which a signal component (voice) having a low temporal correlation is emphasized by the above-described adaptive processing (step S4) into an analog signal so that the user can listen to the digital signal.

As described above, in the present invention, the first invention has been described as the invention of the method with reference to FIG. 1. An adaptive processing circuit for performing the adaptive processing (step S4) is shown in FIG. From the detailed description, it is clear that the first invention can be configured as an apparatus invention.

[Second Invention] FIG. 3 shows a digital input signal of one channel in which a desired signal (eg, voice) having a high temporal correlation and an undesired signal (eg, noise) having a low temporal correlation are mixed. 4 shows a flowchart of a signal processing procedure according to the signal extraction method of the present invention for extracting a desired signal having a high temporal correlation from the above.

In FIG. 3, a signal (speech) having a high temporal correlation and a signal (noise) having a low temporal correlation are mixed.
First, a channel signal (analog signal) is input (step S8). This signal is, for example, an audio signal containing noise picked up by a microphone. This signal is digitized for an adaptive process described below, as in the case shown in FIG. 1 (step S9).
The digitized signal is branched as shown in the figure, and is subjected to high-frequency emphasis (step S10), delay 1 by the number of samples (delay amount) M (step S11), and delay 2 by the number of samples (delay amount) L ( Step S12) is performed.

FIG. 4 shows an example of an equivalent circuit of a high-frequency emphasis circuit for high-frequency emphasis performed in step S10, in which a difference signal between one sample of an input signal is generated, and the difference signal is generated with a predetermined gain A. In this configuration, the output is multiplied by the gain. The same is true even if the difference signal is generated after multiplying the input signal by the gain A.

The signal y _j emphasized in the high frequency range in step S10
Number of samples (delay amount) M delayed signal x _j in step S11, described above, FIG. 2 of the same circuit configuration (although, since the output signal is not a signal of interest, the output terminal is not required) of The reference signal of the adaptive processing circuit is supplied to the terminal to which the reference signal is supplied and to the adaptive filter 1, and the adaptive processing shown in step S13 is executed.

Also in this case, the difference e _j between the high-frequency emphasized signal y _j and the output signal of the adaptive filter 1, ie, the signal z _j supplied to the subtracted terminal of the subtractor 3, and the adaptive filter 1 based on the respective tap output signals, as described for the FIG. 2, sequentially updated by the learning identification method the filter coefficient of the adaptive filter 1 as the difference e _j becomes smaller. At this time, the input signal x of the adaptive filter 1
Between the _j and the reference signal y _j, since there is a time lag of just sample number M, to determine the filter coefficients so that the difference e _j decreases emphasizes the high signal component temporally correlated This is equivalent to performing a filtering process.

The control of the parameters (gain A, delay amounts M and L, number of taps k, convergence coefficient μ) shown in step S14 in FIG. 3 is the same as that described in step S5 in FIG. Here, the description is omitted. However, the gain A controls the amplitude level of the high-frequency emphasizing circuit shown in FIG. 4, which is not shown in FIG.

In FIG. 3, the digital signal delayed by the number of samples (delay amount) L in step S12 is subjected to matched filter processing in step S15 using the matched filter shown in FIG. The output of the matched filter, the filter coefficients when the filter coefficients of the adaptive filter obtained by the adaptive processing in step S13, thereby obtained is emphasized high signal component (audio) q _j temporally correlated. In FIG. 3, the white arrow indicates that the filter coefficient [1] of the adaptive filter obtained in step S13 is copied to the filter coefficient of the matched filter in step S15.

The reason why the digital input signal is supplied to the matched filter by the number of samples L in step S12 is as follows. That is, since the phonemes expressed over a certain period of time are successively formed in phonemes, extracting a desired voice with a matched filter as described later is determined by an adaptive filter at the end of the phonemes. Since it is necessary to control the matched filter from the start of the phoneme using the filter coefficient, the time difference is controlled by the delay time L. These control parameters such as the delay times M and L of the two delay circuits, the high-frequency emphasized signal gain A, the number of taps k of the adaptive filter, and the convergence coefficient μ used in the adaptive signal processing algorithm are determined by parameter control (step S14). It can be set. These parameters (M,
The setting of (L, A, k, μ) is to set each control parameter while listening to the sound obtained from the apparatus of the present invention so that the sound is most audible in terms of audibility.

The matched filter (see FIG. 5) used in step S15 has the same configuration as the adaptive filter in the adaptive processing circuit, and has the same number of tap coefficients k as the adaptive filter. Then, the filter coefficient of the adaptive filter used in step S13 is filtered by the filter coefficients are copied, and outputs the filtered digital audio output signal q _j. That is, the number of taps of the matched filter is k, and the input signal vector of the matched filter is

[Outside 9] The output signal q _j of the matched filter in which the coefficient vector of the copied matched filter is expressed by the following equation (2).

(Equation 2)

In this case, the delay of the number of samples L is for adjusting the time axis of the output signal of the adaptive processing circuit and the digital audio signal input to the matched filter, as described above. The digital audio output signal q _j from the matched filter is converted into an analog signal in step S16 and is listened (step S17).

As described above, the second invention of the present invention also includes a high-frequency emphasizing circuit for performing high-frequency emphasis (step S10), an adaptive processing circuit for performing adaptive processing (step S13), and a matched circuit for performing matched filter processing (S15). The filters are described with reference to the drawings and the like (the adaptive processing circuit is the same as that used in the first invention, and the drawings are omitted), and their parameter control is also described. Obviously, it can be configured as a device invention.

[Another Embodiment of the First Invention] When the first invention is to be realized by software processing by reading a signal extraction program from a recording medium into a computer, one channel of a digital input signal is sampled. If the processing is performed one by one, the calculation processing takes time depending on the CPU used, and real-time processing may be difficult. In order to solve this, a method of processing a digital input signal for each block will be described below.

FIG. 6 shows the configuration of data buffering in this case. When the present embodiment is configured as an apparatus, a block dividing unit, a first buffer, and a second buffer are required.

Hereinafter, the sizes of the first and second buffers (memory) and the way of storing (writing and reading) the digital input signal will be described in an itemized manner. Thereby, the entire process of signal extraction can be computerized. The digital input signal is cut out into blocks (block data) every F samples from the start of capturing. In the present embodiment, F = 256 samples. Blocks can be cut anywhere. Let k be the number of taps in the adaptive filter. In the present embodiment, k = 128 is adopted as an initial value. A buffer 1 provided on the input side of the adaptive processing circuit, and N
The sample is stored in the buffer 2 which also serves as a delay (see FIG. 1). At this time, the buffer 1 has a (k + F) buffer length (the number of samples), and the buffer 2 has a buffer length of (N + 2 × k + F) including a delay of the number of samples N.

At the position of F samples (indicated by * in FIG. 6) from the end of each buffer (buffer 1 and buffer 2), the temporally latest (newest) sample in the block is the end of each buffer. The extracted block data is stored so as to be stored in the. The address of the data stored in each buffer is controlled and read out by shifting one sample at a time from the second sample at the head of each buffer, and the processing of the first invention described above is performed. As described above, according to the present embodiment, by performing address control on the memory, the above processing can be performed without shifting the physical memory storage location, so that the calculation time can be reduced, which is extremely advantageous. It is.

Writing and reading of digital signals are performed as follows. First, the second sample from the head of the buffer 1 is used as a reference signal y _j (see FIGS. 1 and 2) of the adaptive processing circuit.
X _j (output signal of the step S3) input signal of the adaptive processing circuit corresponding to the reference signal is a sample delayed by N samples from the second samples of the beginning of the buffer 1, the tap output of the adaptive filter Since the signal is k samples from the sample to the sample delayed by k samples, the signal corresponds to k samples from the second sample at the beginning of the buffer 2. Therefore, k samples are read from the second sample at the head of the buffer 2 and filtering is performed using the filter coefficients of the adaptive filter 1 (see FIG. 2). The filtering process, the successive updating of the filter coefficients, and the process of subtracting the output signal of the adaptive filter from the reference signal and outputting the result are repeated while shifting one sample from the second sample at the head of each buffer in the time direction. That is, using data of F samples from the second sample at the head of the buffer 1 and data of (F + k-1) samples from the second sample at the head of the buffer 2, the vector data corresponding to the time j and the time j + F-1 While updating the filter coefficient of the adaptive filter F times until reaching the vector data corresponding to the following equation (3),

(Equation 3) The process of obtaining the output z _{j of the} adaptive filter and subtracting it from the reference signal y _j to obtain the output signal e _j is repeated.

Subsequently, the buffer 1 stores k from the end.
Shift the sample forward by F samples. Buffer 2
Shifts (k × 2 + N) samples from the end by F samples forward. Next, the next block data is stored in the last F samples of each buffer, and the subsequent processing is repeated.

In this embodiment, for convenience, the buffer length is (k + F) for buffer 1 and (N + 2 × k + F) for buffer 2, but as can be seen from FIG. 6, each buffer is controlled at the same timing. The buffer length of each buffer is at least F for buffer 1 and at least (N + k + F) for buffer 2.
I just need.

[Another Embodiment of the Second Invention] As in the other embodiment of the first invention, the digital input signal is processed for each block, so that a signal extraction program is read from a recording medium into a computer. The processing speed at the time of software processing is increased to enable real-time processing.

FIG. 7 shows the configuration of data buffering in this case. When this embodiment is configured as an apparatus, a third buffer is required in addition to the block dividing means, the first buffer, and the second buffer.

Hereinafter, the sizes of the first, second and third buffers (memory) and the manner of storing (writing and reading) the digital input signal will be described in an itemized manner. Thereby, the entire process of signal extraction can be computerized. The digital input signal is cut out into blocks (block data) every F samples from the start of capturing. In the present embodiment, F = 256 samples. Blocks can be cut anywhere. Let k be the number of taps in the adaptive filter. In the present embodiment, k = 128 is adopted as an initial value. The cut-out block data is stored in a buffer 1 provided on the input side of the high-frequency emphasizing circuit, a buffer 2 also serving as a delay 1 input to the adaptive filter, and a buffer 3 serving as a delay 2 input to the matched filter (see FIG. 3). ). At this time, the buffer 1 has a buffer length of (k + F) pieces (the number of samples), the buffer 2 has a buffer length of (M + 2 × k + F) including a delay of the number of samples M, and the buffer 3 has a delay of the number of samples L. (L + 2 × k +
F) is the buffer length.

Each buffer (buffer 1, buffer 2
And at the location of F samples (indicated by * in FIG. 7) from the end of the buffer 3) so that the temporally latest (newest) sample in the block is stored at the end of each buffer. Stores the block data. The address of the data stored in each buffer is controlled so as to be shifted one sample at a time from the second sample at the head of each buffer (the first sample for buffer 1) and read out. Perform processing. As described above, according to the present embodiment, by performing address control on the memory, the above processing can be performed without shifting the physical memory storage location, so that the calculation time can be reduced, which is extremely advantageous. It is.

The writing and reading of digital signals are performed as follows. First, the difference between the first sample from the head of the buffer 1 and the sample after it is taken, and this difference is used as a reference signal for the adaptive processing circuit. X _j (output signal of the step S11) the input signal of the adaptive processing circuit corresponding to the reference signal,
Number of samples M from the second sample at the beginning of buffer 1
And the tap output signal of the adaptive filter is k samples from the sample to the sample further delayed by k samples. Therefore, k tap samples from the second sample at the beginning of the buffer 2 Sample. The matched filter input signal (output signal of step S12) at that timing is a sample delayed by the number of samples L from the second sample at the head of the buffer 1, and each tap output signal of the matched filter is Is the k samples up to the sample further delayed by k samples, and thus corresponds to k samples from the second sample at the head of the buffer 3.

Therefore, k samples are read from the second sample at the head of each of the buffers 2 and 3, and
While filtering using a filter coefficient of the adaptive filter 1 (see FIG. 2) to obtain an output signal q _j performs filtering by copying the filter coefficients of the adaptive filter to the filter coefficients of the matched filter. The filtering process and the successive update process of the adaptive filter coefficients are repeated while shifting one sample at a time from the second sample at the head of each buffer (the first sample for buffer 1). That is, F + 1 samples from the first sample at the beginning of buffer 1 and (F (F) samples from the second sample at the beginning of buffers 2 and 3)
+ K-1) The filter coefficient of the adaptive filter is calculated F times from the vector data corresponding to the time j to the vector data corresponding to the time j + F−1 using the high-frequency emphasis signal y _j using the sample data. While updating and copying it to the filter coefficient of the matched filter,

(Equation 4) The process of obtaining the output q _{j of the} matched filter is repeated.

Subsequently, the buffer 1 stores k from the end.
Shift the sample forward by F samples. Buffer 2
Shifts (k × 2 + M) samples from the end by F samples forward. Buffer 3 stores (k
× 2 + L) Shift the sample forward by F samples.
Next, the next block data is stored in the last F samples of each buffer, and the subsequent processing is repeated.

In this embodiment, for convenience, the buffer length is (k + F) for buffer 1, (M + 2 × k + F) for buffer 2, and (L + b) for buffer 3.
+ 2 × k + F). However, as can be seen from FIG. 7, the buffer length of each buffer for controlling each buffer at the same timing is at least (F +
1) For buffer 2, at least (M + k +
F), at least (L + k + F) for buffer 3
I just need.

In the signal extraction method and apparatus of the present invention described above, the first invention and another embodiment thereof (the invention capable of extracting a sound from a signal of one channel in which a sound and a musical sound are mixed) can be used. The fact that the components are reduced and the sound components are obtained with emphasis (effects of the present invention) are shown by signal waveforms (a) and (b) in FIG. 8 showing the input signal and the output signal of the device of the present invention, respectively. ing.

Also, in the signal extraction method and apparatus of the present invention, the second invention and another embodiment thereof (an invention capable of extracting voice from a signal of one channel in which voice and noise are mixed) can be used. FIG. 9 is a diagram showing an input signal and an output signal of the device of the present invention, in which the component is reduced and the voice component is obtained with emphasis (effect of the present invention)
10 (a) and (b), and the running spectrum waveforms (a) and (b) of FIG.

[0068]

According to the present invention, it is possible to extract only a desired audio signal in real time from a one-channel signal in which musical sounds and noises are mixed, in a state where it is most audible. Further, the present invention can be applied to various fields such as preprocessing for voice recognition and hearing aids used by elderly people, hearing-impaired persons, and the like.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating a signal extraction method of the present invention (particularly, a first invention).

FIG. 2 shows the configuration of an adaptive processing circuit as an equivalent circuit.

FIG. 3 is a flowchart illustrating a signal extraction method of the present invention (particularly, a second invention).

FIG. 4 shows an equivalent circuit of a high-frequency emphasis circuit.

FIG. 5 shows a configuration of a matched filter by an equivalent circuit.

FIG. 6 shows a configuration of data buffering in the signal extraction device of the present invention (in particular, another embodiment of the first invention).

FIG. 7 shows a configuration of data buffering in the signal extraction device of the present invention (in particular, another embodiment of the second invention).

FIG. 8 shows the effects of the present invention (especially, the first invention and the first invention).
Another embodiment of the present invention is shown by comparison of signal waveforms.

FIG. 9 shows the effects of the present invention (especially, the second invention and the second invention)
Another embodiment of the present invention is shown by comparison of signal waveforms.

FIG. 10 shows an effect of the present invention (particularly, the second invention and another embodiment of the second invention) by comparing waveforms of a running spectrum.

[Explanation of symbols]

 Reference Signs List 1 adaptive filter 2 adaptive filter coefficient successively updating unit 3 subtractor D 1 sample delay Σ summer A gain

 ────────────────────────────────────────────────── ─── Continuing from the front page (72) Inventor Tetsuo Umeda F-term in the Japan Broadcasting Corporation, Broadcasting Research Institute, 1-10-11 Kinuta, Setagaya-ku, Tokyo 5J023 DB03 DB05 DD03 DD05

Claims (13)

[Claims] 1. A signal extracting method for extracting a desired signal having a low temporal correlation from a digital input signal of one channel in which a desired signal having a low temporal correlation and an undesired signal having a high temporal correlation are mixed. Dividing the digital input signal into two, delaying one of the two branched digital input signals by a specified number N of samples, and converting the digital input signal delayed by the number N of samples into one.
Input to an adaptive filter consisting of the number of taps k specified at the sample interval, and output the output signal of the adaptive filter by using a set of tap output signals of the adaptive filter and a set of filter coefficients that are sequentially updated. An operation is performed based on an operation expression, the operation result is subtracted from the other of the two branched digital input signals, the subtraction result is output as the desired signal having a low temporal correlation, and the digital input signal In synchronization with the sequential input, based on the subtraction result, each set of tap output signals of the adaptive filter, the set of filter coefficients, and the designated convergence coefficient μ, the filter coefficients are reduced so that the subtraction result becomes smaller. A step of sequentially updating the set of filter coefficients by newly generating a set, and repeatedly performing the calculation, the subtraction, and the output. No. extraction method. 2. A signal extraction method for extracting a desired signal having a high temporal correlation from a digital input signal of one channel in which a desired signal having a high temporal correlation and an undesired signal having a low temporal correlation are mixed. Branching the digital input signal into three, subjecting the first signal of the three-branched digital input signal to high-frequency emphasis processing and a designated gain multiplication, and outputting the result as a high-frequency emphasis signal; Delaying a second signal of the branched digital input signal by a specified number of samples M, delaying a third signal of the three branched digital input signals by a specified number of samples L, A digital input signal delayed by several M is input to an adaptive filter consisting of a specified number of taps k at one sample interval, and an output signal of the adaptive filter is output to each of the adaptive filters. Calculating using a set of top output signals and a set of sequentially updated filter coefficients based on a first calculation expression, subtracting a calculation result based on the first calculation expression from the high-frequency emphasized signal; Based on the subtraction result, the set of tap output signals of the adaptive filter, the set of filter coefficients, and the designated convergence coefficient μ, the filter coefficient set is newly generated so that the subtraction result is reduced. A set of filter coefficients is sequentially updated, a digital input signal delayed by the number of samples L is input to a matched filter having the same configuration as the adaptive filter, and an output signal of the matched filter is output to each tap output signal of the matched filter. And a set of the filter coefficients is used to perform a calculation based on a second calculation expression, and outputs a calculation result based on the second calculation expression as the desired signal having a high temporal correlation, The set of filter coefficients of the adaptive filter that is sequentially updated is copied to the set of filter coefficients of the matched filter, and in synchronization with the sequential input of the digital input signal, an operation based on the first operation expression; The subtraction, the successive update of a set of filter coefficients of the adaptive filter, the second
A signal extraction method comprising repeatedly performing an operation based on the operation formula (1) and copying the set of the output and the filter coefficient. 3. A signal extracting apparatus for extracting a desired signal having a low temporal correlation from a digital input signal of one channel in which a desired signal having a low temporal correlation and an undesired signal having a high temporal correlation are mixed. Means for splitting the digital input signal into two, and one delaying means for receiving one of the digital input signals which has been split into two by the splitting means and delaying the signal by a designated number of samples N and outputting the delayed signal. An adaptive filter having an output signal of the delay means as an input signal and having a tap number k specified at one sample interval, wherein an output signal of the adaptive filter is a set of tap output signals of the adaptive filter; An adaptive filter that performs an operation based on a predetermined operation expression using a set of filter coefficients that are sequentially updated, and outputs an operation result of the operation; 2 by 2 branching means
Subtraction means for subtracting from the other signal of the branched digital input signal, and outputting the subtraction result as a desired signal having a low temporal correlation; and the subtraction means in synchronization with the sequential input of the digital input signal. Based on the output signal, the set of tap output signals of the adaptive filter, the set of filter coefficients, and the specified convergence coefficient μ, the set of filter coefficients is newly generated so that the subtraction result is reduced. A signal extraction apparatus comprising at least adaptive filter coefficient successively updating means for sequentially updating the set of filter coefficients. 4. The signal extraction device according to claim 3, wherein
Block splitting means for splitting the digital input signal into blocks for every F samples is further provided at an input end of the apparatus, and the two-branching means and the subtracting means of the other signal system of the bifurcated digital input signal are further provided. First buffer means having a storage capacity of at least F samples interposed between the first buffer means and the first buffer means for sequentially storing output signals of the block dividing means in units of the blocks, and after the storing, A storage sample before (F-1) samples in time counted from a first predetermined sample predetermined according to the storage capacity of the first buffer means stored in the first buffer means; The first signal is input to the subtraction means as the other signal of the input signal, and is stored in the first block during storage of the next block unit after storage in the block unit. A first method of repeatedly inputting a total of F samples to the subtraction means while shifting one sample at a time from the storage sample before (F-1) samples counted from a fixed sample in the first predetermined sample direction; Buffer means, and second buffer means having a storage capacity of at least (N + k + F) samples instead of the delay means, wherein the output signals of the block dividing means are sequentially stored in the block unit, and after the storage, At the same timing as the first predetermined sample, counting from a second predetermined sample predetermined according to the storage capacity of the second buffer means stored in the second buffer means at a time (N
+ K + F-1) from the stored sample before the sample to the second
A set of k consecutive samples toward the predetermined sample is input to the adaptive filter as each tap output signal of the adaptive filter, and after the block unit storage, the next block unit storage is performed. A total of F sets of the k consecutive samples are shifted from the storage sample before (N + k + F−1) samples one sample at a time in the second predetermined sample direction counting from the two predetermined samples. A second buffer means for repeatedly inputting each tap output signal of the adaptive filter to the adaptive filter in synchronization with the first buffer means. 5. The signal extraction device according to claim 3, wherein the signal extraction device further includes parameter control means for designating the number N of samples of the delay amount, the number k of taps, and the convergence coefficient μ, respectively. A signal extraction device characterized by the following. 6. A signal extracting apparatus for extracting a desired signal having a high temporal correlation from a digital input signal of one channel in which a desired signal having a high temporal correlation and an undesired signal having a low temporal correlation are mixed. A third branching means for the digital input signal, a first signal of the digital input signal branched into three by the three branching means, and high-frequency emphasizing processing and a designated gain multiplication are performed on the signal; High-frequency emphasis means for outputting as a high-frequency emphasis signal; and a second signal of a digital input signal branched into three by the three-branch means, which is delayed by a designated number of samples M and output. A second delay means for receiving a third signal of the digital input signal branched into three by the first branch means and delaying the signal by a designated number L of samples. And an adaptive filter having a tap number k specified at one sample interval and having an output signal of the first delay means as an input signal, wherein the output signal of the adaptive filter is a tap output signal of each of the adaptive filters. And an adaptive filter that operates based on a first operation expression using a set of filter coefficients and a set of filter coefficients that are sequentially updated; and a subtraction unit that subtracts an output signal of the adaptive filter from an output signal of the high-frequency emphasis unit; As a result of the subtraction by the subtraction means, based on the set of each tap output signal of the adaptive filter, the set of the filter coefficients, and the designated convergence coefficient μ, the set of the filter coefficients is renewed so that the subtraction result is reduced. And a matched filter to which an output signal of the second delay means is inputted, wherein the adaptive filter coefficient successively updating means for sequentially updating the set of the filter coefficients is generated. Calculating the output signal of the matched filter using a set of tap output signals of the matched filter and a set of filter coefficients based on a second calculation expression, and calculating a calculation result based on the second calculation expression with the time. A matched filter that outputs a highly correlated desired signal, and means for copying the successively updated set of filter coefficients of the adaptive filter to a set of filter coefficients of the matched filter. In synchronization with the sequential input of the signal, the operation based on the first operation expression, the subtraction, the sequential update of the set of adaptive filter coefficients, the operation based on the second operation expression,
A signal extracting apparatus, wherein the output and the copy of the set of filter coefficients are repeatedly performed. 7. The signal extraction device according to claim 6, wherein
A third delay means for delaying the input signal by one sample; a difference means for generating and outputting a difference signal between input and output signals of the third delay means; A signal extraction device comprising: a multiplication means for multiplying an input signal connected to a stage before the means or a stage after the difference means by the gain. 8. The signal extracting device according to claim 7, further comprising a block dividing means for dividing the digital input signal into blocks for every F samples, at an input end of the device, and further comprising: First buffer means having an alternative storage capacity of at least (F + 1) samples,
The output signal of the block dividing unit or the output signal of the block dividing unit multiplied by the gain is sequentially stored in units of the blocks, and the first buffer unit stored in the first buffer unit after the storage is stored. A set of two consecutive samples counted from a first predetermined sample predetermined in accordance with the storage capacity of the storage device and stored in time from the storage sample before F samples toward the first predetermined sample. Of the first predetermined sample counted from the first predetermined sample until the storage of the next block unit following the storage in the block unit, and the first sample is stored one sample at a time from the storage sample before F samples. A first buffer means for repeatedly outputting a total of F sets of two consecutive samples to the difference means while shifting in a predetermined sample direction; At least (M + k + F) instead of the first delay means
A second buffer unit having a storage capacity of samples, wherein the output signals of the block dividing unit are sequentially stored in units of the blocks, and after the storage, the second signal is output at the same timing as the first predetermined sample. From the storage sample before (M + k + F-1) samples in time counting from a second predetermined sample predetermined according to the storage capacity of the second buffer unit stored in the second buffer unit. A set of k continuous samples toward a predetermined sample is input to the adaptive filter as each tap output signal of the adaptive filter, and the second block is stored until the next block is stored after the block is stored. In the direction of the second predetermined sample one by one from the storage sample before (M + k + F-1) samples from the predetermined sample in time. Second buffer means for repeatedly inputting the set of k consecutive samples of the total F sets in synchronization with the first buffer means as each tap output signal of the adaptive filter to the adaptive filter; At least (L + k + F) instead of the delay means of 2.
A third buffer unit having a sample storage capacity, wherein the third buffer unit sequentially stores output signals of the block dividing unit in units of the blocks, and stores the output signals at the same timing as the first predetermined sample after the storage. The third predetermined time is counted from a storage sample before (L + k + F-1) samples counted from a third predetermined sample predetermined according to the storage capacity of the third buffer means stored in the buffer means. A set of k consecutive samples toward the sample of the above is input to the matched filter as each tap output signal of the matched filter, and the third block is stored until the next block unit is stored after the block unit is stored. (L + k + F-1)
The third sample is stored one sample at a time from the stored sample before the sample.
A total of F sets of the k
A third buffer means for repeatedly inputting a set of consecutive samples in synchronization with the first buffer means and inputting each tap output signal of the matched filter to the matched filter. apparatus. 9. The signal extraction device according to claim 6, wherein the signal extraction device includes a sample number M and L of the delay amount, a tap number k, the convergence coefficient μ, and the gain. The signal extraction apparatus further comprises parameter control means for respectively specifying the following. 10. A computer for extracting a desired signal having a low temporal correlation from a digital input signal of one channel in which a desired signal having a low temporal correlation and an undesired signal having a high temporal correlation are mixed. A medium on which a program is recorded, wherein the signal extracting program is provided to a computer, wherein taps specified at one-sample intervals are configured in a computer using a digital input signal delayed by a specified number N of samples as an input signal. An adaptive filter consisting of several k is operated on the output signal of the adaptive filter based on a predetermined arithmetic expression using a set of tap output signals of the adaptive filter and a set of filter coefficients that are sequentially updated. Is subtracted from the digital input signal, and the result of the subtraction is output as the desired signal having a low temporal correlation. In synchronization with the sequential input of the digital input signal, the subtraction result is reduced based on the subtraction result, the set of tap output signals of the adaptive filter, the set of filter coefficients, and the designated convergence coefficient μ. A medium in which a signal extraction program is recorded, wherein the set of filter coefficients is sequentially updated by newly generating the set of filter coefficients, and the calculation, the subtraction, and the output are repeatedly performed. 11. A computer for extracting a desired signal having a low temporal correlation from a digital input signal of one channel in which a desired signal having a low temporal correlation and an undesired signal having a high temporal correlation are mixed. A medium on which a program is recorded, wherein the signal extracting program causes a computer to divide the digital input signal into blocks for each F sample, and to sequentially store the block-divided signals in block units; After storing the unit, the time is counted (N + k + F-1) counting from the sample at a predetermined position in the stored block based on the designated number of samples N and the number of taps k.
A set of k continuous samples in the time direction from the storage sample before the sample, and one storage sample before (F-1) sample in time counted from the sample at the predetermined position are read out. Using the set of samples and the set of filter coefficients that are sequentially updated, the output signal of the adaptive filter consisting of the number of taps specified at one sample interval k is calculated based on a predetermined calculation formula, and the calculation result is read out. Subtracting from one sample, outputting the subtraction result as the desired signal having a low temporal correlation, synchronizing with the sequential input of the digital input signal, synchronizing the subtraction result, the set of read samples, and the filter coefficient. Based on the set and the designated convergence coefficient μ, the filter coefficient set is newly generated so that the subtraction result is reduced, and The set of coefficients is successively updated, and the set of k consecutive samples and the one sample are sequentially sampled one by one in the time direction until storage of the next block unit after storage of the block unit. A signal extraction program for causing a total of F sets of samples and one sample to be read out while being shifted, and repeatedly performing the calculation, the subtraction, the output, and the successive update of the set of filter coefficients. Medium on which is recorded. 12. A computer for extracting a desired signal having a high temporal correlation from a digital input signal of one channel in which a desired signal having a high temporal correlation and an undesired signal having a low temporal correlation are mixed. A medium on which a program is recorded, wherein the signal extracting program is provided to a computer, wherein a tap designated at one sample interval is configured in a computer having a digital input signal delayed by a designated number M of samples as an input signal. An adaptive filter consisting of several k is made to perform an operation based on a first arithmetic expression using an output signal of the adaptive filter using a set of tap output signals of the adaptive filter and a set of filter coefficients that are sequentially updated. An arithmetic result based on the first arithmetic expression is obtained from a high-frequency emphasis signal obtained by subjecting the digital input signal to high-frequency emphasis processing and a designated gain multiplication. Based on the result of the subtraction, the set of each tap output signal of the adaptive filter, the set of the filter coefficients, and the designated convergence coefficient μ so that the set of the filter coefficients is reduced so that the result of the subtraction becomes small. And a set of the filter coefficients is sequentially updated, and a matched filter having the same configuration as the adaptive filter configured in the computer having a digital input signal delayed by a specified number of samples L as an input signal is provided. The output signal of the matched filter is converted to a second signal using the set of each tap output signal of the matched filter and the set of the filter coefficients.
And outputs a calculation result based on the second calculation expression as the desired signal having a high temporal correlation, and sets the set of filter coefficients of the adaptive filter, which is sequentially updated, to the matched value. A set of filter coefficients of the filter is copied in synchronization with the sequential input of the digital input signal, an operation based on the first operation expression, the subtraction, a sequential update of the set of filter coefficients of the adaptive filter, the second A medium for recording a signal extraction program, which repeatedly performs a calculation based on the calculation formula of (1) and a copy of the set of the output and the filter coefficient. 13. A computer for extracting a desired signal having a high temporal correlation from a digital input signal of one channel in which a desired signal having a high temporal correlation and an undesired signal having a low temporal correlation are mixed. A medium on which a program is recorded, wherein the signal extraction program causes a computer to divide the digital input signal into blocks for every F samples, and to sequentially store the block-divided signals in block units; After memorizing units,
Based on the designated number of samples M and the number of taps k, the first continuous k samples in the time direction from the stored sample (M + k + F-1) before the sample counted from the predetermined position in the stored block. A set of samples, a second set of k samples in time direction from the stored sample (L + k + F-1) samples in time direction counted from the sample at the predetermined position based on the designated number of samples L and k. And reading out a third set of two consecutive samples in the time direction from the stored sample that is temporally F samples before counting from the sample at the predetermined position, and sequentially reading out the read first set of samples. The output signal of the adaptive filter having the number of taps k specified at one sample interval is calculated based on the first calculation expression using the updated set of filter coefficients. Subtracting an operation result based on the first operation expression from a difference value between two samples constituting the set of the third sample multiplied by a designated gain; Based on the set of one sample, the set of filter coefficients, and the designated convergence coefficient μ, the set of filter coefficients is newly generated so that the subtraction result is reduced, and the set of filter coefficients is sequentially updated. Using the read second set of samples and the set of filter coefficients to calculate an output signal of a matched filter having the same configuration as that of the adaptive filter based on a second arithmetic expression;
An operation result based on the second operation expression is output as a desired signal having a high temporal correlation, and the first set of samples is stored until the next block unit is stored following the block unit storage. Reading out a set of a total of F samples while shifting the second set of samples and the third set of samples one by one sequentially in the time direction, the operation, the subtraction, the output, and the A medium on which a signal extraction program is recorded, wherein a sequential update of a set of filter coefficients of an adaptive filter is repeatedly performed.