JP3489587B2 - Adaptive noise reduction device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a noise reduction device using adaptive processing.

[0002]

2. Description of the Related Art Conventionally, as an adaptive noise reduction apparatus, FIG.
A circuit as shown in is known. In FIG. 6, 1 is a main input terminal, 2 is a reference input terminal, and the signal input through the main input terminal 1 is supplied to the combining circuit 4 via the delay circuit 3. Further, the signal input through the reference input terminal 2 is supplied to the combining circuit 4 through the adaptive filter circuit 5 and subtracted from the signal from the delay circuit 3.
The output of the synthesis circuit 4 is fed back to the adaptive filter circuit 5 and is also led to the output terminal 6.

In this noise reduction apparatus, a main input terminal 1 is supplied with a desired signal s and an unnecessary voice signal (noise) n0 uncorrelated with the desired signal s. On the other hand, noise n1 is input to the reference input terminal 2. The noise n1 of the reference input has no correlation with the desired signal s, but has a correlation with the noise n0.

The adaptive filter circuit 5 has a reference input noise n1.
To output a signal y that approximates noise n0. As the output signal y of the adaptive filter circuit 5, a signal having a phase opposite to that of the noise n0 and an equal amplitude can be obtained. The delay circuit 3 is for adjusting the time of the signals to be subtracted in consideration of the processing time in the adaptive filter circuit 5. The delay circuit 3 may not be provided in some cases.

The adaptive algorithm in the adaptive filter circuit 5 works so as to minimize the subtraction output (residual output) e which is the output of the synthesis circuit 4. That is, now s, n
Assuming that 0, n1 and y are statistically stationary and the average value is 0, the residual output e is e = s + n0-y. The expected value of the square of this is that s is n0,
Also, because it is y ^{uncorrelated, E [e 2] = E} [s 2] + E [(n0 -y) 2] + 2E [s (n0 -y)] = E [s 2] + E [(n0 - y) ² ]. Assuming that the adaptive filter circuit 5 converges,
The adaptive filter circuit 5 is adjusted so that E [e ² ] is minimized. At this time, since E [s ² ] is not affected, Emin [e ² ] = E [s ² ] + Emin [(n 0 −
y) ² ].

^{Namely, E [(n0 -y) 2} ] is minimized by E [e ^2] is minimized, the output y of the adaptive filter circuit 5 will estimate the noise n0. The expected value of the output from the synthesis circuit 4 is only the desired signal s. That is, adjusting the adaptive filter circuit 5 to minimize the total output power is equivalent to the subtraction output e becoming the least-squares estimated value of the desired speech signal s.

The output e is generally the signal s with some noise remaining, but since the output noise is given by n−y, minimizing E [(n−y) ² ] is Is equivalent to maximizing the signal-to-noise ratio of.

In some cases, the synthesis circuit 4 serves as a sound synthesis means. That is, the adaptive filter circuit 5 forms a noise canceling voice signal -y having a phase opposite to that of noise and an equal amplitude, and supplies this to a speaker or the like to acoustically add to the main voice to reduce noise. To do. The residual e in this case will be picked up by the residual detection microphone.

The adaptive filter circuit 5 can be realized by either an analog signal processing circuit or a digital signal processing circuit. FIG. 7 shows an example in which the adaptive filter circuit 5 is realized by using a digital filter. In this example, the so-called LMS (Least Mean Squares) method is used as an adaptive algorithm.

[0010] As shown in FIG. 7, in this example, adaptive Fi
The filter circuit 5 uses the FIR filter 300 as a digital filter . That is, the adaptive filter circuit 5
A plurality of delay circuits DL1, DL2, ... DLm each having a delay time τ (Z ⁻¹ ) of a unit sampling time
(M is a positive integer), input noise n1 and each delay circuit DL
1, DL2, ... Weighting circuits (coefficient multipliers) MX0, MX1, MX for multiplying the output signal of DLm and the weighting coefficient
2, ... MXm, and an adder circuit 310 for adding the outputs of the weighting circuits MX0 to MXm. The output of the adder circuit 310 is y.

The weighting coefficients supplied to the weighting circuits MX0 to MXm are formed by the LMS arithmetic circuit 320, which is, for example, a microcomputer, based on the residual signal e from the synthesizing circuit 4 and the reference input. The algorithm executed by the LMS arithmetic circuit 320 is as follows.

Now, the reference input vector X _{k at} time _k
As shown in FIG. 7, X _k = [x _0k x _1k x _2k ... x _mk ] ^T , the output is y _k , and the weighting coefficient is w _jk (j = 0,1,2, ... m). ), The input / output relationship is as shown in the following Equation 1.

[0013]

[Equation 1] Becomes

Then, the weight vector W _{k at} time _k
The, if defined as _{W k = [w 0k w 1k} w 2k ··· w mk] T, the input-output relationship is given by ^{_{y k = X k T · W}} k ... (1). If the desired response is d _k , the error e _{k from the} output is expressed as follows. e _k = d _k −y _k = d _k −X _k ^T · W _k (2) In the LMS method, the weight vector is updated by W _{k + 1} = W _k +2 μ · e _k · X _k (3) I will go by the formula. The initial value of the weighting coefficient is set to a constant value or a random value. μ is a gain factor (step gain) that determines the speed and stability of adaptation.

In the above formula (3), a vector for modifying the coefficient vector W _k at a certain time k, is a second term of the right side of the equation (3), a scalar gain factor μ and instantaneous error e _k Both directly affect the modified value. Similarly, since the reference input vector X _k also works in the form of a product, this also affects the correction value. The average convergence time constant τ _a is represented by τ _a = (n + 1) / 4 μ · trE [X _i X _j ^T ]. Here, n + 1 is the order of the reference input vector ( n corresponds to the number of taps of the FIR filter), trE [X _i
X _j ^T ] is the average power of the reference input. That is, FI
The larger the number of taps of the R filter, the slower the convergence speed, and the larger the gain factor μ, the faster the convergence speed.

In the case of a stationary signal, the final residual noise level is large when the convergence speed is fast, and conversely the final noise level is small when the convergence is slow. However, when the target signal fluctuates like a voice, its properties change before it completely converges. Therefore, the amount of cancellation increases as the convergence speed increases to some extent.

The value of the gain factor μ approaches the one where the output y of the adaptive filter circuit 5 cancels the noise n0 and converges so that the output of the device becomes equivalent to the desired signal s. Need to be satisfied. When 0 <μ <(power of signal) / (number of taps of FIR filter + 1) (4) μ is larger than the range of Expression (4), the output y of the adaptive filter circuit 5 diverges, and It makes a lot of noise as an output.

Conventionally, the value of the gain factor μ takes into consideration the number of taps of the adaptive filter, which directly affects the quality of the processed signal, and the magnitude (power) of the reference input signal, and is given by the above equation (4). ) Is satisfied so that the adaptive filter circuit 5 operates normally (converges), and is set to a constant value.

[0019]

In the adaptive noise reduction system described above, when the signal of interest is a stationary signal and a signal such as white noise,
The desired system output can be obtained. However, in the case of a non-stationary and colored audio signal such as a human voice, the desired system output may not be obtained in some cases.

For example, when the adaptive operation is performed by using the adaptive algorithm such as the LMS method or the learning identification method described above so as to minimize the residual output power of the noise with respect to the human voice, the human voice has the power. It was found that the adaptive filter sometimes constitutes a moving average type filter having the characteristics of a low-pass filter due to the adaptive operation because it is biased to the low frequency band.

That is, for example, assuming that both the main input and the reference input are voices from the microphones, a signal from a sound source is incident on the main input microphone and the reference input microphone. A component equivalent to the noise component n in the input is assumed to be incident on the reference input at time t + D.

Then, at this time, the number of taps and the delay with respect to the main input are appropriately set to operate the system adaptively. For example, when the human voice that becomes noise has a level twice that of the reference input and is incident on the main input at the same time as the reference input (D = 0), the number of taps of the adaptive filter is 2 (the order is 3), FIG. 8 shows a state in which the delay amount of the delay circuit 3 is set to 1 times τ and the initial values of the weighting coefficients are all set to 0 and the adaptive operation is performed. 8A, 8B, and 8C show weighting factors w _0k , w _1k , and w.
Equivalent to _2k . In this case, at the input terminal of the subtraction circuit 4, since the delay amount of the reduction target signal during the main input and the reference input is τ, the adaptive filter circuit 5 is essentially the output of the delay circuit DL1. The weighting coefficient w _1k for the second tap should be in a larger state than the other weighting coefficients.

However, as shown in FIG. 8 , in this case, the coefficients w _{0k to} w _2k all have substantially the same coefficient value, and the FIR filter in this case temporally adds and averages (moving average). Has the same effect as a low pass filter.

In such a case, in the adaptive noise reduction system, the reduction processing is carried out in the low frequency range, so that the system output cannot completely cancel the unnecessary signal and the residual output has a high frequency range. Remains, which adversely affects the timbre of the human voice.

Another example will be given in which the desired reduced output cannot be obtained as the system output. That is, when the signal to be reduced is supplied to the main input and the reference input through various paths, respectively, the filter configured by the adaptive operation is
As mentioned above, it is configured to minimize error power. Therefore, the reduction amount of a signal with large power is large,
The reduction amount of a signal having a small power is small.

An object of the present invention is to provide an adaptive noise reduction device that can overcome the above drawbacks.

[0027]

In order to solve the above-mentioned problems, a noise reduction device according to the present invention has a correlation between a main input terminal to which a main input audio signal is input and a signal to be reduced in the main input audio signal. Reference input terminal to which a certain reference input audio signal is input, and multiple delay circuits with unit delay time
Delay means connected in series, a plurality of coefficient multipliers for multiplying a signal from each tap of the delay means by a weighting coefficient, and an addition means for adding the outputs of the coefficient multipliers And an adaptive filter unit that is supplied with a reference input audio signal from the reference input terminal and outputs a signal approximate to a reduction target signal in the main input audio signal, and an adaptive filter unit of the adaptive filter unit. A synthesis means for subtracting the output signal from the main input speech signal and a means for adjusting the adaptive filter means so that the output power of the synthesis means is minimized, and at least the reduction in the main input speech signal. said delay of said adaptive filter means corresponding to the delay time difference between the target signal, the reduction target signal in the reference input speech signal
An adaptive noise reduction device configured to set a weighting coefficient at a tap of the means relatively large with respect to a weighting coefficient at another tap.

[0028]

According to the above configuration of the present invention, the adaptive filter means is initially set to the weighting coefficient with which the speech to be reduced is originally reduced. Therefore, even if the signal is a non-stationary signal such as a human voice and is a colored voice signal, it is possible to avoid the adaptive filter from having a moving average filter configuration, and the system output is not affected by the magnitude of the signal power. Is obtained.

[0029]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the adaptive noise reduction device according to the present invention will be described below with reference to FIGS.

In FIG. 1, 11 is a main input microphone for picking up a desired voice, and 21 is a reference input microphone for picking up unwanted voice or ambient noise in a direction to be removed as noise. This example, the arrival direction of desired sound is mainly, as indicated by an arrow AR in FIG. 3, the diagram is a direction from the top to the bottom (hereinafter referred to a front direction), the rear direction the direction opposite to the direction (hereinafter It is an example of realizing a device that does not pick up the voice from () as noise.

[0031] In this example, the main input microphone 11 is composed of a nondirectional microphone, as shown in FIG. On the other hand, the reference input microphone 21 is
As shown in FIG. 3 , the microphone is a unidirectional microphone having low sensitivity in the desired voice arrival direction and high sensitivity in the rear direction.

Further, in the case of this example, these microphones 11 and 21 are arranged in the direction of the arrow AR, and the distance d between them is such that the sound propagates during the unit delay time τ of the adaptive filter circuit described later. Distance (d = cτ)
Equal to.

The voice signal picked up by the main input microphone 11 and converted into an electric signal is supplied to the AD converter 13 via the amplifier 12 and converted into a digital signal. It is supplied to the subtraction circuit 14. Further, the sound signal picked up by the reference input microphone 21 and converted into an electric signal is supplied to the A / D converter 23 via the amplifier 22, is converted into a digital signal, and is adaptively filtered. It is supplied to the circuit 24.

In this embodiment, the adaptive filter circuit 24
Is it consists as shown in FIG. That is, the adaptive filter circuit 24 includes two delay circuits 241, 242 having a unit delay time τ and three weighting circuits 243, 244, 2
45 and a third-order FIR filter 240 including an adding circuit 246, and a weighting circuit 24 of the FIR filter 240.
LM which outputs the weighting coefficient supplied to 3,244,245
It is composed of an S arithmetic circuit 247. The adaptive filter circuit 24 can be configured by a DSP (digital signal processor) including a microcomputer. Then, the digital signal from the A-D converter 23 is
The FIR filter 2 is supplied to the arithmetic circuit 247 and
It is supplied to the subtraction circuit 14 via 40.

The output signal of the subtraction circuit 14 is the arithmetic circuit 24.
The analog signal is returned to the analog signal by the D / A converter 15 and is led to the output terminal 16. In addition, D
The -A converter 15 may be omitted, and the output signal of the subtraction circuit 14 may be output to the output terminal 16 as a digital signal.

In this case, the LMS of the adaptive filter circuit 24
The weighting factors w _0k , w _1k , w _2k from the arithmetic circuit 247 are obtained from the following arithmetic expressions. w _{0k + 1} = w _0k + 2μ ₀ · e _k · x _0k (5) w _{1k + 1} = w _1k +2 μ ₁ · e _k · x _1k (6) w _{2k + 1} = w _2k +2 μ ₂ · e _k X _2k (7) Then, in this example, among the initial values w ₀₀ , w ₁₀ and w ₂₀ of the weighting coefficients in the equations (5), (6) and (7), the input terminal of the subtraction circuit 14 At the tap position corresponding to the delay time difference between the reduction target signal in the main input and the reduction target signal in the reference input (in this example, the reduction target signal from the direction opposite to the arrow AR is the main input microphone). 11
, The sound is delayed by the distance d, that is, τ, than the reference input microphone 21. Therefore, the initial value w ₁₀ of the weighting coefficient w _1k ) is set larger than the initial values of the other weighting coefficients. To be done. This is because the adaptive filter is
The subtraction circuit 14 works so as not to cause a delay time difference between the reduction target signal in the main input and the reduction target signal in the reference input at the input end of the subtraction circuit 14, and a coefficient of the tap position of the delay corresponding to the delay time difference is set. It was decided based on the idea that it should be treated with priority over other coefficients.

The time- _dependent changes of the three weighting factors w _0k , w _1k and w _2k when the adaptive system having the above-mentioned configuration is operated are as shown in FIGS. 4A, 4B and 4C.

In this case, the signal to be reduced from the direction opposite to the arrow AR is picked up by the main input microphone 11 with a delay of τ from the reference input microphone 21, and therefore the adaptive filter circuit. as the second tap, that is, the operating state weighting coefficient w _1k is preferably larger state than the other weighting factors for the output x _1k of the delay circuit 241, FIG. 4, that has in its operating state I understand.

Thus, according to the present invention, the target reduction target signal can be sufficiently reduced without the adaptive filter exhibiting the characteristics of the conventional low-pass filter. Incidentally, the reduction degree in the conventional example shown in FIG. 8 was −19 dB, whereas the reduction degree in this example was −22 dB to −23 dB.

In the above example, the initial value of the weighting coefficient w _1k is set to be larger than the other weighting coefficients, but the point is that this weighting coefficient w _1k is a relative problem with other weighting coefficients. ,
Instead of increasing the weighting factor w _1k, other weighting factors, the same effect can be obtained so as to be smaller than the weighting factor w _1k.

Further, instead of controlling the initial value of the weighting coefficient, the value of each step gain μ0, μ1, μ2 in the updating equations (5), (6), (7) of each weighting coefficient is controlled. Good.

That is, from the idea that the coefficient of the tap position of the delay corresponding to the delay time difference between the reduction target signal in the main input and the reduction target signal in the reference input needs to be treated with priority over other coefficients. The value of the step gain μ1 of w _1k is set to be larger than the other step gains μ0 and μ2.

FIG. 5 shows the changes over time of the three weighting factors when the adaptive system having such a configuration is operated. Also in the case of this example, in the adaptive filter circuit, the weighting factor w _1k for the second tap, that is, the output x _1k of the delay circuit 241 is always larger than the other weighting factors, and the desired operating state is obtained. Has become. Therefore, in the subtraction circuit 14, the error after the reduction target signal is subtracted from the main input is reduced.

[0044]

As described above, according to the present invention, the adaptive filter means according to the delay time difference between the reduction target signal in the main input signal and the reduction target signal in the reference input is provided.
By setting the initial value of the weighting coefficient for the delayed output or the step gain to be larger than those of other taps, it is possible to obtain a non-stationary signal such as a human voice and when it is colored. Also, the target reduction target signal in the main input can be appropriately reduced.

[Brief description of drawings]

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

FIG. 2 is a circuit diagram of an example of an adaptive filter circuit used in an embodiment of the present invention.

FIG. 3 is a diagram showing an example of directivity of first and second microphones.

FIG. 4 is a diagram for explaining the operation of the embodiment of the present invention.

FIG. 5 is a diagram for explaining the operation of the embodiment of the present invention.

FIG. 6 is a block diagram showing an outline of an adaptive noise reduction device.

FIG. 7 is a diagram illustrating a configuration example of an adaptive filter circuit.

FIG. 8 is a diagram for explaining the operation of a conventional adaptive noise reduction device.

[Explanation of symbols] 11 Main input microphone 21 Reference input microphone 14 Subtraction circuit 24 Adaptive filter circuit 241,242 delay circuit 243, 244, 245 Coefficient multiplier (weighting circuit) 247 LMS arithmetic circuit

─────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Takayuki Mizuuchi 6-735 Kitashinagawa, Shinagawa-ku, Tokyo Sony Corporation (56) Reference JP-A-61-194914 (JP, A) JP 61-194913 (JP, A) JP-A-4-174500 (JP, A) JP-A-4-216599 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) G10L 21 / 02

Claims (3)

(57) [Claims] 1. A main input terminal to which a main input audio signal is input, a reference input terminal to which a reference input audio signal having a correlation with a reduction target signal in the main input audio signal is input, and a delay of a unit delay time. A delay means including a plurality of circuits connected in series; a plurality of coefficient multipliers for multiplying a signal from each tap of the delay means by a weighting coefficient; and an adding means for adding the outputs of the coefficient multipliers. An adaptive filter means that is supplied with a reference input audio signal from the reference input terminal and outputs a signal approximate to a reduction target signal in the main input audio signal; and an output signal of the adaptive filter means as a main input. And a means for adjusting the adaptive filter means so as to minimize the output power of the synthesizing means, and at least the main input voice signal. Signal to be reduced and the reduction target signal in the reference input audio signal, the weighting coefficient at the tap of the delay means of the adaptive filter means corresponding to the delay time difference is compared with the weighting coefficient at the other taps. The adaptive noise reduction device is designed to be relatively large. 2. A main input terminal to which a main input audio signal is input, a reference input terminal to which a reference input audio signal having a correlation with a reduction target signal in the main input audio signal is input, and a delay of a unit delay time. A delay means including a plurality of circuits connected in series; a plurality of coefficient multipliers for multiplying a signal from each tap of the delay means by a weighting coefficient; and an adding means for adding the outputs of the coefficient multipliers. An adaptive filter means that is supplied with a reference input audio signal from the reference input terminal and outputs a signal approximate to a reduction target signal in the main input audio signal; and an output signal of the adaptive filter means as a main input. And a means for adjusting the adaptive filter means so as to minimize the output power of the synthesizing means, and at least the main input voice signal. Signal to be reduced in the reference input audio signal and a gain factor used in updating the weighting factor at the tap of the delay means of the adaptive filter means corresponding to the delay time difference between the reduction target signal in the reference input audio signal. , An adaptive noise reduction device that is set to be relatively large with respect to a gain factor used when updating weighting coefficients in other taps. 3. A main input terminal to which a main input audio signal is input, a reference input terminal to which a reference input audio signal having a correlation with a reduction target signal in the main input audio signal is input, and a delay of a unit delay time. A delay means including a plurality of circuits connected in series; a plurality of coefficient multipliers for multiplying a signal from each tap of the delay means by a weighting coefficient; and an adding means for adding the outputs of the coefficient multipliers. An adaptive filter means that is supplied with a reference input audio signal from the reference input terminal and outputs a signal approximate to a reduction target signal in the main input audio signal; and an output signal of the adaptive filter means as a main input. And a means for adjusting the adaptive filter means so as to minimize the output power of the synthesizing means, and at least the main input voice signal. Signal to be reduced in the reference input speech signal and the reduction target signal in the reference input audio signal, the initial value of the weighting factor at the tap of the delay means of the adaptive filter means corresponding to the delay time difference is weighted at the other taps. An adaptive noise reduction device in which a coefficient is set relatively large with respect to an initial value.