JP2845476B2 - Noise canceller - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial application field) The present invention relates to a noise canceller.

(Prior Art) In recent years, systems using advanced signal processing techniques such as an acoustic echo canceller, a speech codec, and a speech recognition device have been widely used.

Conventionally, the above-described system has been used in a relatively favorable environment such as inputting voice by bringing the mouth close to a microphone or inputting voice in a quiet place such as a conference room.
In recent years, with the diversification of the purpose of use, for example, in an environment that is not necessarily good in offices, factories, automobiles, etc.,
In addition, the microphone may be used in a situation where the mouth is away from the microphone.

For this reason, attempts to reduce the influence of noise have been made actively by installing a noise canceller that removes noise mixed into the microphone based on the signal from the noise source before the input of each of the above-described systems. ing.

Such a noise canceller is as follows.

FIG. 7 is a diagram for explaining a conventional noise canceller.

The noise canceller shown in FIG. 1 converts a microphone 1, a noise canceling microphone 2, and an analog signal from each of the microphones 1 and 2 into a digital signal.
It comprises A / D converters 3, 4, a subtractor 5, and an adaptive filter 6.

A noise source signal N (Z) that can be handled as a discrete signal by the A / D converters 3 and 4 (where Z = EXP (2πj f / f
s) fs; sampling frequency) is the indoor acoustic characteristic HB
(Z) is input to the microphone 2 and
The desired signal S (Z) is input to the microphone 1 via the acoustic characteristic HA (Z).

The adaptive filter 6 uses an adaptive algorithm such as a learning identification method to calculate a noise transfer function from the output signals of the microphones 1 and 2 when there is no desired signal S (Z). And a pseudo noise transmission path (Z) inside To generate pseudo noise (Z).

Then, the noise component is canceled by the subtractor 5 from the output signal YA (Z) of the microphone 1 as in the following equation. That is, Becomes

As described above, in the conventional system, it is possible to substantially cancel the noise component mixed in the microphone 1.

However, the above-described conventional noise canceller has not been widely adopted because of the following problems.

(1) Since the characteristics of the acoustic characteristics HA (Z) and HB (Z) are complicated, the time range of the impulse response estimated by the adaptive filter 6 is very long, 100 msec or more, and the sampling frequency of the system is fs = 8 KHz. At this time, an adaptive filter having a long tap length of 1000 taps or more is required, which causes an increase in system cost.

(2) In the above-described configuration, it is assumed that no sound that becomes the desired signal S (Z) is mixed into the microphone 2 at all.
In this case, the adaptive filter 6 cancels the desired signal itself, which causes distortion of the desired signal.

(Problems to be Solved by the Invention) As described above, in the above-described conventional noise canceller, an adaptive filter having a long tap length is required, so that there is a problem that the cost increases in the system configuration. In such a case, since sound is mixed into the noise canceling microphone, there is a problem that the sound signal is eventually canceled by the adaptive filter and the sound signal is distorted.

The present invention has been made to solve the above-described conventional problems, and by using a filter with high efficiency, a filter having a long tap length can be equivalently realized, thereby reducing the cost of the system configuration. It is an object of the present invention to provide a noise canceller that can effectively prevent distortion of a desired signal.

[Means for Solving the Problems] The present invention provides a first input unit for inputting a desired signal, a second input unit for noise cancellation, and the first and second input units. And a part for estimating a transfer function of a noise component included in the partial band extracted by the partial band extracting means and generating a pseudo noise component of the partial band. A band component generating unit, an entire band component generating unit that returns the pseudo noise component of the partial band generated by the partial band component generating unit to the entire band of the output signal to generate a pseudo noise component of the entire band, Noise removing means for removing a noise component in the output signal by subtracting a pseudo-noise component in the entire band generated by the full-band component generating means from an output signal of the first input means. Eteiru.

(Operation) In the present invention, a transfer function of the noise is estimated from an extracted signal obtained by extracting a predetermined partial band from the output signals of the first and second input means, and a pseudo noise component is generated. By returning to the entire band, a pseudo noise component of the entire band is generated.
Then, the noise component in the output signal is removed by subtracting the pseudo noise component of the entire band from the output signal from the first input means.

Accordingly, by using a filter of a predetermined partial band with high efficiency, a filter having a long tap length can be equivalently realized, thereby reducing the cost of the system configuration and effectively preventing distortion of a desired signal. can do.

(Example) Hereinafter, an example of the present invention is described using a drawing.

FIG. 1 is a block diagram for explaining a configuration of a noise canceller according to one embodiment of the present invention.

As shown in the figure, this noise canceller mainly includes a first microphone 11 for inputting an audible sound,
Microphone 12 for noise cancellation, each microphone
Convert analog signal from 11 and 12 to digital signal
A / D converters 13, 14, a delay circuit 15 for delaying and outputting an output signal from the A / D converter 13 for a predetermined period, and each of the A / D converters 13, 1
Phase linear filters 16 and 17 for extracting a predetermined partial band in the output signal from 4, sampling processors 18 and 1 for sampling the signals from the respective filters 16 and 17
9, an adaptive filter 20, a subtractor 21, a sampling processor 22 for sampling a signal from the adaptive filter 20, a phase linear filter 23, and subtraction for subtracting a signal from the phase linear filter 23 from a signal from the delay circuit 15 described above. Container 24.

In FIG. 1, a signal HA (Z) of a noise source N (Z),
HB (Z) is 0-fs / 16 as shown by the hatched portion in FIG.
(For example, 500 Hz).

The desired signal S (Z), which is an audio signal, is an audio signal of fs / 2 (for example, 4 KHz) or less as shown by the solid line in FIG.

These signals HA (Z), HB (Z) and S (Z)
Are input to microphones 11 and 12 and then A / D
The converters 13 and 14 use discrete signals YA (Z) and YB shown below.
(Z).

YA (Z) = S (z) + HA (Z) N (Z) YB (z) = HB (Z) N (Z) (3) These signals have phases having characteristics as shown in FIG. A phase linear filter 16, which is a linear band limiting filter,
After being band-limited by 17, sampling processors 18, 19
(For example, sampling at 2 KHz), as shown in FIG. 4, the sampled band division signal Y'A
(Z ') and Y'B (Z') are as follows: Y'A (Z ') = L (Z') Z- ^{d / 2} YA (Z) Y'B (Z ') = L (Z') Z ^{−d / 2} YB (Z) (4) where Z ′ = EXP (2πj f / fs2) d / 2; fs The amount of delay of the band-limiting filter depending on the number of Hz samples L (Z ′); (Real number).

In addition, the adaptive filter (AF) 20 and the subtractor 21 use a learning algorithm such as a learning identification method to generate the signal (Y ′).
A (Z ′), Y′B (Z ′)), a transfer function between the microphones 11 and 12 regarding the noise signal, However, H'A (Z ') and H'B (Z') are HA (Z),
The transfer function in the subband of HB (Z) is estimated, and a pseudo noise '(Z') is generated from '(Z') = '(Z') Y'B (Z ') (6).

This pseudo noise is re-sampled by the sampling processor 22 to become a signal having a folded component as shown in FIG. 5, and then a linear phase filter 23 having a linear phase for band limitation as shown in the following equation. , 0-fs
A pseudo noise signal in a band of / 16 (for example, 500 Hz).

(Z) = L (Z) Z− ^{d / 2} · N ′ (Z ′) (7) Here, as shown in FIG. 3 at ffs / 8, L (Z) =
If L (Z) = 1 at 0 ffs / 16, then N
(Z) is given by: (Z) = Z− ^{d / 2} N ′ (z ′) 0 ffs / 16 (Z) = L (Z) · Z− ^{d / 2} · ′ (Z ′) fs / 16 <f <fs / 8 (8) (Z) = 0 fs / 8f.

Therefore, substituting equation (6) into equation (8) at 0 ffs / 16 gives And from equations (3) and (4) Get.

Here, the estimation of the adaptive filter 20 is correct and the band 0f
fs / 16, H'A (Z ') = HA (Z) H'B (Z') = HB (Z) (11) Becomes

On the other hand, the input signal to the microphone 11 is a signal YA (Z) composed of a desired signal and a noise component.
When the following subtraction is performed in consideration of (Z ′) being delayed by d samples of the delay of the band division filter, the residual signal E (Z) becomes Thus, as shown in FIG. 6, the noise component can be canceled.

Therefore, in this embodiment, it is possible to realize a noise canceller using band division processing without introducing a band division filter into a desired signal, and to provide a good noise canceller that does not give distortion to the desired signal. be able to.

In this embodiment, when the frequency fs2 to be resampled by the band division processing is 1 / M of the original sample frequency fs, the adaptive filter in the band division processing has N taps, and when the band division is not performed, N / fs The impulse response system that can be processed is MN
/ fs, which is M times that of the conventional noise canceller.

Further, the interval of one sample in the band division processing is M times that in the case where band division is not performed, and the processing amount of the adaptive filter per sample (fS) becomes 1 / M of the conventional one.

Therefore, efficiency of ^twice the conventional M is realized by the introduction of the band division process to the noise canceller.

Further, according to the present invention, since a desired signal does not include a band division filter or the like, distortion due to band division processing such as a band gap does not occur.

Furthermore, in the present invention, since the noise canceller is applied only to a limited band such as 0 to 500 Hz, for example, a desired signal is input to the noise canceling microphone 12, so that the generated distortion does not cover the entire band. The effect of distortion is greatly reduced as compared with the related art.

In this description, the band to which the band division is applied is applied to the reduction, but any band may be used as long as it is a narrow band noise.

[Effects of the Invention] As described above, the noise canceller of the present invention reduces the cost of the system configuration by implementing a filter having a long tap length equivalently by using a filter of a predetermined partial band with high efficiency. And distortion of the desired signal can be effectively prevented.

[Brief description of the drawings]

FIG. 1 is a block diagram for explaining a noise canceller according to an embodiment of the present invention, FIGS. 2 to 6 are diagrams for explaining each signal in the noise canceller of FIG. 1, and FIG.
FIG. 1 is a block diagram showing a conventional noise canceller. 11, 12 …… Microphone, 13, 14 …… A / D converter, 15…
… Delay circuit, 16, 17, 23 …… Phase linear filter,
18, 19, 22 ... sampling processor, 20 ... adaptive filter, 21, 24 ... subtractor.

Claims (1)

(57) [Claims] A first input means for inputting a desired signal; a second input means for noise cancellation; and a partial band for extracting a predetermined partial band from output signals of the first and second input means. Extracting means, a partial band component generating means for estimating a transfer function of a noise component included in the partial band extracted by the partial band extracting means and generating a pseudo noise component of the partial band, Full-band component generating means for returning the generated pseudo-noise component of the partial band to the entire band of the output signal to generate a pseudo-noise component of the entire band; and outputting the full-band component from the output signal of the first input means. Noise canceling means for removing noise components in the output signal by subtracting pseudo-noise components in all bands generated by the generating means. .