JPH11231900A - Method and device for noise reduction - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method and a device for noise reduction capable of preventing the decrease in the level of an output signal and in an S/N ratio caused by cross talk from a signal source. SOLUTION: In a noise reduction method, where a noise signal based on a first input signal X2, which is inputted through a noise input part 3, to a first adaptive filter 7 is inputted, a differential signal E1 of a second input signal X1, which is inputted through a main input part 2 and includes at least a target signal S1, and an output signal Y1 from the adaptive filter 7 is generated, and a filter coefficient of the first adaptive filter 7 based on the differential signal E1 is set to output a target signal S1 as the differential signal E1, a target signal S2 included in the first input signal X2 is removed to output a signal Y1 after the removal to the first adaptive filter 7 as a noise signal.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a noise reduction method and a noise reduction device for reducing noise input together with voice to voice input means such as a microphone of a talker in a noise environment.

[0002]

2. Description of the Related Art Generally, when a pilot of a jet aircraft or the like talks through a microphone mounted inside a helmet, harsh noise such as engine sound is input to the microphone together with voice. Under such circumstances, the engine sound and the like are transmitted to the conversation partner,
Smooth conversation is hindered. For this reason, a countermeasure is taken by attaching a microphone to a position where the engine sound or the like is difficult to be input, or by using a bone conduction microphone. However, in reality, a sufficient effect cannot be obtained.

On the other hand, if the noise signal can be removed from the output signal of the microphone, the above-mentioned disadvantages can be solved. As a device for removing a noise signal from an output signal of a microphone of this type, a noise reduction device 21 shown in FIG. 2 is conventionally known. The noise reduction device 21 is a kind of an adaptive noise canceller, and includes a main input unit 2, a noise input unit 3, an adaptive filter 7, and a difference circuit 8, as shown in FIG. Here, the main input unit 2 includes, for example, a microphone having directivity in the direction of the signal source 11, and an A / D converter for converting an output signal of the microphone into a digital signal. The main input unit 2 simultaneously inputs a target signal S1 which is an audio signal emitted from the signal source 11 and a noise signal N1 from the noise source 12 mixed into the target signal S1, and converts the input signal into a signal X1. Analog-digital conversion and output. on the other hand,
The noise input unit 3 includes, for example, a microphone having directivity in the direction of the noise source 12, and an A / D converter for converting an output signal of the microphone into a digital signal. The noise input unit 3 receives the noise signal N2 output from the noise source 12 and converts the input signal into a signal X2.
To analog-to-digital conversion and output. The adaptive filter 7 is a digital filter having a fixed number of taps and a variable internal filter coefficient, and outputs a signal Y1 corresponding to a noise component by filtering the signal X2. Further, the difference circuit 8 converts the signal X1 to the signal Y1.
Is subtracted to obtain a difference signal E between the two signals X1 and Y1.
1 is output as the output signal of the device.

In this case, in the noise reduction device 21, the signal components of the signals X1 and X2 are represented by the following equations, respectively, and the signal component included in the difference signal E1 is represented by the following equation. The noise is reduced on the premise of this. X1 = S1 + N1 Expression X2 = N2 Expression E1 = X1-Y1 = S1 + N1-Y1 Expression

More specifically, in this noise reduction device 21,
The signal X2 is output from the noise input unit 3 and the main input unit 2
Output the signal X1 from the adaptive filter 7
Follows the so-called LMS (Least Mean Square) method,
The internal filter coefficients are sequentially corrected so that the square value of the difference signal E1 is minimized. In this case, since the correlation between the target signal S1 and the noise signal N1 and the noise signal N2 is low, if the signal component represented by the term (N1-Y1) in the above equation is minimized, The square value of the difference signal E1 becomes minimum. On the other hand, the signal component of the signal X2 is represented by the above equation. Therefore, the adaptive filter 7 outputs a signal Y1 of a component whose phase and signal level are close to the noise component of the noise signal N1. As a result, the above (N1-Y1)
Since the signal component represented by the term (4) is minimized, the difference circuit 8
Outputs the target signal S1 as the difference signal E1. As described above, in the noise reduction device 21, the internal filter coefficient of the adaptive filter 7 is appropriately corrected so that the square value of the difference signal E1 is minimized, and the noise signal N
By removing 1, the target signal S1, which is an audio signal having a good S / N ratio, is output as an output signal.

[0006]

However, the conventional noise reduction device 21 has the following problems. That is, the conventional noise reduction device 21 attempts to improve the S / N ratio of the audio signal on the premise that the above equation is satisfied. However, in practice, as shown by the broken line in FIG.
The target signal S2 from the signal source 11 is mixed into the noise input unit 3 as crosstalk. Therefore, the signal component of the signal X2, which is the output signal of the noise input unit 3, is actually represented by the following equation.
This signal X2 is input. In this case, since the audio signal output from the signal source 11 has periodicity, a high correlation exists between the target signal S1 included in the signal X1 and the target signal S2 included in the signal X2. Have a relationship. Therefore, the adaptive filter 7 corrects the internal filter coefficient so as to minimize the square value of the difference signal E1, so that the internal filter coefficient becomes the phase and signal level of the signal X1 as a whole and the signal Y1 as a whole. In some cases, the phase and the signal level are corrected in a direction to match each other.
In such a case, the phase and the signal level of the target signal S2 included in the signal Y1 approximate to the phase and the signal level of the target signal S1 included in the signal X1, respectively. S1 will be removed from signal X1. Therefore, when crosstalk occurs from the signal source 11 to the noise input unit 3, the conventional noise reduction device 21 causes a reduction in the output signal level of the target signal itself and a reduction in the S / N ratio. There is a problem that sometimes. X2 = N2 + S2... Formula

SUMMARY OF THE INVENTION The present invention has been made in view of the above-mentioned problems, and a noise reduction method and a noise reduction method capable of preventing a decrease in output signal level and a decrease in S / N ratio due to crosstalk from a signal source. A main object is to provide a noise reduction device.

[0008]

According to a first aspect of the present invention, there is provided a noise reduction method comprising: inputting a noise signal based on a first input signal input via a noise input unit to a first adaptive filter; By generating a difference signal between the second input signal including at least the target signal and input through the main input unit and the output signal of the first adaptive filter, and setting a filter coefficient of the first adaptive filter based on the difference signal In a noise reduction method for outputting a target signal as a differential signal, the target signal included in the first input signal is removed, and the signal after the removal is output to the first adaptive filter as a noise signal.

In this noise reduction method, first, a target signal included in a first input signal input via a noise input unit is removed. Thus, the signal after the removal mainly includes a pure noise component. Next, the signal after the removal is output as a noise signal to the first adaptive filter, and the first adaptive filter sets its filter coefficient based on a difference signal between its own output signal and the second input signal. in this case,
A signal obtained by removing the target signal from the first input signal;
Since the correlation with the target signal included in the input signal is extremely low, the first adaptive filter outputs, for example, an output signal of a component whose phase and signal level are similar to the noise signal N1 included in the second input signal. Output. Next, a noise signal included in the second input signal is removed by generating a difference signal between the output signal of the first adaptive filter and the second input signal. In this case, since the target signal is not removed when the difference signal is generated, the S / N ratio of the target signal can be improved without lowering the signal level of the target signal.

The noise reduction method according to the second aspect is the first aspect.
In the noise reduction method described above, a first input signal is delayed to generate a delayed input signal, the delayed input signal is input to a second adaptive filter, and a difference signal between the first input signal and an output signal of the second adaptive filter is calculated. A noise signal is generated, and a filter coefficient of the second adaptive filter is set based on the difference signal.

For example, if the frequency of the target signal is known, the target signal contained in the first input signal input via the noise input unit can be removed by using a notch filter or the like. However, in this method, when the frequency of the target signal fluctuates, the target signal may not be removed from the first input signal, and when the band of the notch filter is wider than the frequency band of the target signal, In some cases, not only the target signal but also noise is removed. If the frequency of the target signal is not known, it is difficult to use this method itself.

On the other hand, in this noise reduction method, first, the first
An input signal is delayed to generate a delayed input signal, and the delayed input signal is input to a second adaptive filter. Then the second
An adaptive filter sets its filter coefficient based on a difference signal between the first input signal and its own output signal. In this case, since the noise signal included in the delayed input signal is delayed, the correlation between the noise signal and the noise signal included in the first input signal is extremely low. on the other hand,
Since the target signal such as an audio signal has a certain degree of periodicity even if delayed, the target signal included in the delayed input signal mixed in the first input signal due to crosstalk and the first input signal Between the target signal contained in
There is a high correlation. Therefore, the second adaptive filter sequentially corrects its filter coefficients to thereby obtain the first adaptive filter.
A signal having a component whose phase and signal level are similar to the target signal included in the input signal is output. For this reason,
The difference signal between the output signal of the second adaptive filter and the first input signal mainly contains a pure noise signal. As a result, the first adaptive filter receives the difference signal and, based on the difference signal and the second input signal, a noise signal whose phase and signal level are close to the noise signal included in the second input signal. Can be output. Therefore, by generating a difference signal between the output signal of the first adaptive filter and the second input signal, it is possible to generate a target signal having a high S / N ratio without lowering the signal level.

According to a third aspect of the present invention, there is provided a noise reduction apparatus comprising: a first adaptive filter for inputting a noise signal based on a first input signal input via a noise input unit; And a first difference circuit that outputs a difference signal between the second input signal and the output signal of the first adaptive filter, and sets a filter coefficient of the first adaptive filter based on the difference signal, thereby obtaining an objective signal as the difference signal. In a noise reduction device configured to output a signal,
The signal processing apparatus further includes a target signal removing unit that removes the target signal included in the first input signal and outputs the signal after the removal as a noise signal to the first adaptive filter.

According to a fourth aspect of the present invention, there is provided a noise reduction apparatus.
In the noise reduction device described above, the target signal removing unit includes a delay circuit that outputs a delayed input signal obtained by delaying the first input signal, a second adaptive filter that inputs the delayed input signal, and a first input signal and a second input signal. A second difference circuit that outputs a difference signal of the output signal of the adaptive filter, wherein a filter coefficient of the second adaptive filter is set based on the difference signal output from the second difference circuit.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of a noise reduction method and a noise reduction device according to the present invention will be described below with reference to the accompanying drawings. The conventional noise reduction device 2
The same components as those of 1 are denoted by the same reference numerals, and redundant description will be omitted.

FIG. 1 shows a block diagram of the noise reduction device 1. As shown in the figure, a noise reduction device 1 includes a main input unit 2, a noise input unit 3, an adaptive filter 7, and a difference circuit 8, which are almost the same as the configuration of a conventional noise reduction device 21, a delay circuit 4, Adaptive filter 5 and difference circuit 6
Is further provided.

The delay circuit 4, together with the adaptive filter 5 and the difference circuit 6, constitutes a target signal removing means in the present invention, and a signal X2 corresponding to the first input signal in the present invention.
Is delayed by a predetermined time to generate a delayed input signal X3, and outputs the generated delayed input signal X3 to the adaptive filter 5. The adaptive filter 5 corresponds to a second adaptive filter in the present invention, and is configured by a digital filter in which the number of taps is smaller than the number of taps of the adaptive filter 7 and the internal filter coefficient is variable. In this case, the adaptive filter 5 sequentially corrects the internal filter coefficients so as to minimize the root mean square of the difference signal E2 according to the LMS method, and filters the delayed input signal X3 to thereby obtain the target signal as described later. A signal Y2 corresponding to S2
Is output. The difference circuit 6 corresponds to a second difference circuit in the present invention, and outputs a difference signal E2 corresponding to a noise signal in the present invention by subtracting the signal Y2 from the signal X2. The signal component included in the difference signal E2 is:
It is represented by the following equation. The adaptive filter 7 corresponds to a first adaptive filter in the present invention, and outputs a signal Y1.
The difference circuit 8 corresponds to the first difference circuit in the present invention, and outputs the difference signal E1 by subtracting the signal Y1 from the signal X1 corresponding to the second input signal in the present invention. E2 = X2-Y2 ... Equation

Next, a noise reduction method in the noise reduction device 1 will be described.

When an audio signal as the target signal S1 is output from the signal source 11, the target signal S1 is input to the main input unit 2.
1 and a noise signal N2 such as an engine sound emitted from a noise source 12 such as an engine, for example, and the main input unit 2 outputs a signal X composed of a signal component represented by the above equation.
Outputs 1. On the other hand, a noise signal N2 from a noise source 12 and a signal source 11
, And the noise input unit 3 outputs a signal X2 including a signal component represented by the above equation to the delay circuit 4 and the difference circuit 6. Next, the delay circuit 4 delays the signal X2 by a predetermined time and outputs a delayed input signal X3.
Is output to the adaptive filter 5.

The difference circuit 6 subtracts the signal Y2 output from the adaptive filter 5 from the signal X2 to generate a difference signal E2 having a signal component represented by the following equation, and generates the difference signal E2. Output to the adaptive filter 5. E2 = N2 + S2-Y2 ... Equation

On the other hand, the adaptive filter 5 corrects the internal filter coefficient so that the mean square value of the difference signal E2 is minimized. In this case, since the noise signal included in the delayed input signal X3 is delayed, the noise signal and the signal X3 are delayed.
2 has a very low correlation with the noise signal. Conversely, since the target signal S2 such as an audio signal has a certain degree of periodicity even if delayed, the target signal S2 mixed into the noise input unit 3 due to crosstalk and included in the delayed input signal X3 , There is a high correlation with the target signal S2 included in the signal X2. Therefore, in the above equation, if the signal component represented by the term (S2-Y2) is minimized, the mean square value of the difference signal E1 is minimized. For this reason, the adaptive filter 5 outputs the signal Y2 of a component whose phase and signal level are similar to the target signal S2 included in the signal X2 by sequentially correcting the internal filter coefficients. Also, at this time, since the adaptive filter 5 is configured with a small number of taps, the target signal S mixed into the noise input unit 3 due to crosstalk.
Modification of the internal filter coefficients for 2 can be followed very quickly. As a result, the adaptive filter 5 reliably and quickly outputs a signal Y2 of a component whose phase and signal level are similar to the target signal S2 included in the signal X2.

As a result, the difference circuit 6 outputs a noise component corresponding to the noise signal N2 to the adaptive filter 7 as a difference signal E2 by subtracting according to the above equation.
Next, the adaptive filter 7 is connected to the noise reduction device 21 described above.
In the same manner as in the processing in, the noise signal N1 included in the signal X2 is based on the input difference signal E2 and signal X1.
And outputs a signal Y1 having a signal component whose phase and signal level are close to each other to the difference circuit 8. Next, the difference circuit 8 outputs the signal Y output from the adaptive filter 7.
The difference signal E1 is finally generated by subtracting 1 from the signal X1.

As described above, according to the noise reduction device 1, the adaptive filter 5 includes the signal source 1 mixed with the noise input unit 3.
The adaptive filter 7 removes the target signal S2 from the signal source 1 and outputs the signal to the adaptive filter 7, so that the adaptive filter 7 converts the noise signal N2 without being affected by the crosstalk input from the signal source 11 to the noise input unit 3. 8 can be output. As a result, the difference circuit 8 removes the signal component of the target signal S1 included in the signal X1 without removing the signal X1.
, It is possible to remove only the signal component of the noise signal N1 included in the above. Thereby, the noise reduction device 1 has a high S
The target signal S1 having the / N ratio can be generated. Also,
In the noise reduction apparatus 1, even if the frequency of the target signal S1 fluctuates with time, the adaptive filters 5 and 7 sequentially correct the internal filter coefficients in accordance with the frequency fluctuation, so that the influence of crosstalk is reliably removed. can do. Further, since the adaptive filters 5 and 7 modify the internal filter coefficients according to the LMS method, the processing can be performed with a small amount of calculation.
1 can be output quickly.

It should be noted that the present invention is not limited to the above-described embodiment of the present invention, and the configuration thereof can be appropriately changed. For example, in the embodiment of the present invention, adaptive filters 5 and 7 modify the internal filter coefficients according to the LMS method, but may modify the internal filter coefficients according to another method such as the steepest descent method. Also, the adaptive filters 5, 7
May be configured to be variable in the number of taps of the internal filter. Further, the algorithm of the present invention can be implemented by either hardware processing or software processing. In the embodiment of the present invention,
Although the delay circuit 4 and the adaptive filter 5 are provided separately and independently, the present invention is not limited to this, and the delay circuit 4 and the adaptive filter 5 may have an integrated hardware configuration. Further, in the embodiment of the present invention, the noise input unit 3
Although a case has been described where the noise output from 2 is directly input, the present invention is not limited to this, and it is needless to say that the present invention can be applied even in a noise environment where a noise source cannot be specified. The present invention can be applied to a signal separation method and a signal separation device for separating a periodic signal and an aperiodic signal, a speech recognition method and a speech recognition device, and the like.

[0025]

As described above, according to the noise reduction method according to the first aspect and the noise reduction apparatus according to the third aspect, the target signal included in the first input signal input through the noise input unit is removed. By doing so, it is possible to operate the first adaptive filter without being affected by crosstalk, and thereby to reliably reduce noise contained in the second input signal. A signal can be generated.

According to the noise reduction method of the second aspect and the noise reduction apparatus of the fourth aspect, the delayed input signal is input to the second adaptive filter, and the first input signal and the output signal of the second adaptive filter. Based on the differential signal of
By setting the filter coefficient of the adaptive filter, even if the frequency of the target signal fluctuates or is unknown, only the noise signal included in the second input signal is removed without lowering the signal level Therefore, a target signal having a high S / N ratio can be generated.

[Brief description of the drawings]

FIG. 1 is a block diagram of a noise reduction device according to an embodiment of the present invention.

FIG. 2 is a block diagram of a conventional noise reduction device.

[Explanation of symbols]

 Reference Signs List 1 noise reduction device 2 main input unit 3 noise input unit 4 delay circuit 5 adaptive filter 6 difference circuit 7 adaptive filter 8 difference circuit 11 signal source 12 noise source E1, E2 difference signal N1, N2 noise signal S1, S2 target signal X1, X2 signal X3 Delayed input signal Y1, Y2 signal

Claims (4)

[Claims] A noise signal based on a first input signal input via a noise input unit is input to a first adaptive filter, and a second input signal including at least a target signal and input via a main input unit and the second input signal are input to the first adaptive filter. A noise reduction method of generating a difference signal of an output signal of the first adaptive filter and setting a filter coefficient of the first adaptive filter based on the difference signal to output the target signal as the difference signal; A noise reduction method comprising: removing the target signal included in an input signal; and outputting the signal after the removal as the noise signal to the first adaptive filter. 2. A delay input signal is generated by delaying the first input signal, the delayed input signal is input to a second adaptive filter, and a difference between the first input signal and an output signal of the second adaptive filter is generated. 2. The noise reduction method according to claim 1, wherein a signal is generated as the noise signal, and a filter coefficient of the second adaptive filter is set based on the difference signal. 3. A first adaptive filter for inputting a noise signal based on a first input signal input via a noise input unit, and a second adaptive filter including at least a target signal and input via a main input unit.
A first difference circuit that outputs a difference signal between an input signal and an output signal of the first adaptive filter, and sets a filter coefficient of the first adaptive filter based on the difference signal, thereby obtaining the difference signal as the difference signal. In a noise reduction device configured to output a signal, a target signal removing unit that removes the target signal included in the first input signal and outputs the signal after the removal as the noise signal to the first adaptive filter. A noise reduction device comprising: 4. A delay circuit for outputting a delayed input signal obtained by delaying the first input signal, wherein the target signal removing means comprises:
A second adaptive filter that inputs the delayed input signal; and a second differential circuit that outputs a differential signal between the first input signal and the output signal of the second adaptive filter, the second differential filter being output from the second differential circuit. Based on the difference signal, the second
4. The noise reduction device according to claim 3, wherein a filter coefficient of the adaptive filter is set.