JP2013141118A - Howling canceller - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a howling canceller which can suppress howling stably even if acoustic impulse response changes abruptly or continuously, while reducing the circuit scale and the throughput.SOLUTION: In an adaptive howling canceller incorporating a delay sum microphone array 110 and a Griffith-Jim type adaptive beam former 120, the filter coefficient update operation of the adaptive filters 122-0 through 122-(N-1) of Griffith-Jim type adaptive beam former 120 and the adaptive filter 140 of howling canceller is carried out simultaneously in parallel by using a common error signal e(n).

Description

  The present invention relates to a howling canceller that suppresses howling of an audio signal using an adaptive filter.

  When a voice signal input from a microphone is output from a speaker, howling may occur, and various howling cancellers are known to avoid this.

Here, first, a mechanism for expanding the howling margin will be described with reference to FIG. 1 using a model obtained by simplifying a general loudspeaker system as an example. In FIG. 1, when the gain of the amplifier AMP is G _AMP , and the gain between the speaker SP and the microphone MIC is G _SM , the loudspeaker system can maintain stability if G _AMP × G _SM <1, and howling Does not occur. Hereinafter, G _AMP × G _SM will be referred to as “system gain”.

The size of the speaker's voice to be reproduced from the speaker is determined by the product _{G TM} × _{G AMP} of gain _{G AMP} gain _{G TM} and amplifier AMP between speakers Talker and microphone MIC. Hereinafter, G _TM × G _AMP is referred to as “amplification gain”.

In order to expand the howling margin, the system gain G _AMP × G _SM should be made as small as possible, and the sound gain G _TM × G _AMP should be made as large as possible. If _GTM is maintained at a constant value, _GSM may be made as small as possible.

A howling canceller using an adaptive filter attempts to expand the howling margin of a loudspeaker system by equivalently reducing only _GSM by arithmetic processing using the adaptive filter. However, it is to reduce the equivalent to G _SM merely by the operation of the internal howling canceller.

On the other hand, adaptation rather than filter, it is possible even by using a method of directional control of the microphone to reduce the G _SM while maintaining the loudspeaker gain. That may be used for a microphone having a directivity such as the gain G _SM speaker direction becomes smaller with respect to the gain G _TM speaker direction. As an example, there is a method in which the directivity characteristics of a microphone having a sharp main lobe in the speaker Talker direction is formed, the gain G _TM in the speaker direction is increased, and the howling margin of the loudspeaker system is increased.

However, it is difficult to realize a microphone system having a sharp main lobe in a wide frequency band. Therefore, sharp instead of forming the main lobe, to achieve a directional pattern having a sharp null (null) to a speaker SP direction, it is conceivable to enlarge the howling margin to reduce the gain G _SM speaker direction. Further, both main lobe formation in the speaker direction and null formation in the speaker direction may be performed.

  In general, in a sensor array system, it is easier to form a sharp null than to form a sharp main lobe. An addition type array system that forms a main lobe with a sharp directivity requires a large number of sensor elements, but a subtraction type array can form a null with only two sensor elements depending on conditions.

  Here, in a loudspeaker system installed in a room with reverberation, an array microphone having a plurality of nulls can provide a better howling margin expansion effect than an array microphone system having a single sharp main lobe. 2 and FIG.

  FIG. 2 is a schematic diagram showing an example of the positional relationship between the microphone MIC, the speaker SP, the speaker Talker, and the audience listener of the loudspeaker system installed in a room with reverberation. FIG. 3 is a diagram showing an impulse response of an acoustic system between a speaker and a microphone of the loudspeaker system installed in the room with reverberation shown in FIG.

  In a room with reverberation, the howling sound reproduced from the speaker SP and input to the microphone MIC is not only the direct sound component S0 but also the multiple reflections of the initial reflected sound components S1 and S2 from the side wall and the wall surface many times. Reverberation component. The actual initial reflected sound includes components reflected by the ceiling and floor, but for the sake of simplicity, the initial reflected sound components from the ceiling and floor are omitted in FIGS.

  As can be seen from FIG. 3, compared to the direct sound component and the initial reflected sound component, the reverberation component that has been subjected to multiple reflections on the wall surface many times attenuates the amplitude and decreases its energy. Therefore, in order to expand the howling margin of the loudspeaker system, it is possible to obtain a sufficient effect by simply reducing the direct sound component and the initial reflected sound component input from the speaker to the microphone.

  However, as shown in FIG. 4, in the method of designing the microphone array so as to form a sharp main lobe in the speaker direction T, the null is sharp in the direction of the direct sound component S0 from the speaker and the initial reflected sound components S1 and S2. It is not guaranteed that you can. On the other hand, as shown in FIG. 5, in the method of designing the microphone array so as to have a plurality of nulls in the directional characteristics, it is relatively easy to form sharp nulls in the directions of S0, S1, and S2.

  In principle, if K microphones are used, an array having nulls in the K-1 direction can be formed. If a sufficient number of microphones are used, the direct sound component from the speaker can be obtained. The initial reflected sound component can be sufficiently suppressed.

  Therefore, even in a room with reverberation, a microphone array that forms a sharp null can effectively expand the howling margin of the loudspeaker system, compared to a microphone array having a sharp main lobe.

  If null formation by subtraction-type processing and main lobe formation in the speaker direction by addition-type processing are performed simultaneously, an even better howling margin expansion result can be obtained.

  Patent Document 1 discloses a howling canceller using such a microphone array in a loudspeaker system. The howling canceller disclosed in Patent Document 1 uses a signal collected by microphones arranged at equal intervals on a straight line, and adaptively obtains a filter coefficient that lowers the sensitivity in the speaker direction. The direction sensitivity is greatly attenuated to prevent howling.

  Specifically, the microphone array disclosed in Patent Document 1 performs the following processing. Out of M microphones (M is an integer of 2 or more) arranged in a straight line at equal intervals, M−1 units obtained by removing the incoming sound component from the target sound direction from the collected sound signal for each adjacent microphone pair Get the signal.

  Next, the pseudo target sound is filtered by a filter coefficient that gives a difference between the arrival time from the target sound direction to the m-th microphone and the arrival time from the target sound direction to the first microphone. One signal is generated.

  Subsequently, for each microphone pair, M−1 signals obtained by removing the incoming sound component from the collected sound signal and the pseudo target sound are added, and the addition result for each microphone pair is filtered with a different filter coefficient, and fixed. The filter coefficient is learned as a desired signal obtained by adding the M−1 filtered signals to the pseudo target sound to which the delay is given.

  Finally, signals obtained by filtering the collected sound signals picked up by the first to (M-1) -th microphones with the same filter coefficients as the adaptive filter are added and output.

Japanese Patent Laid-Open No. 2003-87891

  However, the above-described howling canceller disclosed in Patent Document 1 deteriorates the convergence characteristics of the adaptive filter due to a large cross-correlation between the direct sound from the speaker input to the microphone and the loud sound output from the speaker. In order to prevent this, a complicated circuit configuration must be used, and learning of the adaptive filter must be turned on / off by generating a virtual target sound. That is, in the conventional howling canceller, the circuit scale and the processing amount must be increased in order to prevent howling.

  In addition, the howling canceller disclosed in Patent Document 1 can move the microphone while the speaker moves while holding the microphone, or block between the microphone and the speaker when a person other than the speaker walks. When this occurs, there is a problem that an abrupt or continuous change in the impulse response of the acoustic system occurs, making it impossible to suppress howling or make the operation unstable.

  An object of the present invention is to provide a howling canceller capable of stably suppressing howling even when a sudden change or a continuous change of an acoustic impulse response occurs without increasing the circuit scale and the processing amount. It is to be.

  A howling canceller according to the present invention is a howling canceller incorporating a microphone array and a Griffith-Jim type adaptive beamformer, which is reproduced from a speaker and input to a microphone included in a desired signal obtained by the microphone array. A first adaptive filter that generates a signal approximating a sound and updates a filter coefficient so that an energy of an error signal obtained using the desired signal and a signal approximating the howling sound is minimized, and the Griffith The Jim type adaptive beamformer generates a signal that approximates a loudspeaker signal component from a speaker included in the output signal of the microphone array, and an error obtained by using the desired signal and the signal that approximates the loudspeaker signal component Minimize signal energy Comprising a second adaptive filter for performing a filter coefficient updating operation, the error signal is obtained by subtracting a signal that approximates a signal and the loudspeaker signal component which approximates the howling sound from the desired signal, a configuration.

  The howling canceller of the present invention is a howling canceller incorporating a microphone array and a Griffith-Jim type adaptive beamformer, and is reproduced from a speaker and input to a microphone included in a first desired signal obtained by the microphone array. A first adaptive filter that generates a signal approximating the howling sound and updates the filter coefficient so that the energy of the first error signal obtained by subtracting the signal approximating the howling sound from the first desired signal is minimized. The Griffith-Jim type adaptive beamformer generates a signal approximating a loudspeaker signal component included in an output signal of the microphone array, and uses the first error signal as a second desired signal. 2. A signal approximating the loudspeaker signal component from the desired signal A configuration in which the energy of the second error signal obtained by subtracting to mount an second adaptive filter for performing a filter coefficient updating operation so as to minimize.

  The howling canceller of the present invention is a howling canceller incorporating a microphone array and a Griffith-Jim type adaptive beamformer, wherein the Griffith-Jim type adaptive beamformer is included in an output signal of the microphone array from a speaker. A filter coefficient update operation that generates a signal approximating the loud signal component and subtracting the signal approximating the loud signal component from the first desired signal obtained by the microphone array so that the energy of the first error signal is minimized. A first adaptive filter that performs the above operation, and generates a signal that approximates a howling sound that is included in the second desired signal and that is reproduced from a speaker and input to a microphone, using the first error signal as a second desired signal, Approximating the howling sound from the second desired signal A configuration in which the energy of the second error signal obtained by subtracting the issue of comprising a second adaptive filter for performing a filter coefficient update so as to minimize.

  According to the present invention, the circuit scale and the processing amount can be reduced, and howling can be stably suppressed even when an abrupt change or a continuous change in the acoustic impulse response occurs.

Diagram showing a simplified model of a general loudspeaker system Schematic diagram showing an example of the positional relationship between the microphone, speaker, speaker, and audience of a loudspeaker system installed in a room with reverberation The figure which showed the impulse response of the acoustic system between the speaker and microphone of the loudspeaker system installed in the room with the reverberation shown in FIG. Schematic diagram showing the relationship between the main lobe and null for the direction of sound arrival of the directional characteristics of the microphone array Schematic diagram showing the relationship between the main lobe and null for the direction of sound arrival of the directional characteristics of the microphone array The block diagram which shows the structure of the loudspeaker which concerns on Embodiment 1 of this invention. The figure which shows the constitution of the filter coefficient update circuit Schematic diagram showing the arrangement of the microphone array Schematic diagram showing the arrangement of the microphone array The block diagram which shows the structure of the loudspeaker which concerns on Embodiment 2 of this invention. The block diagram which shows the structure of the loudspeaker based on Embodiment 3 of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. However, in the embodiment, components having the same function are denoted by the same reference numerals, and redundant description is omitted.

(Embodiment 1)
FIG. 6 is a block diagram showing the configuration of the loudspeaker 100 according to Embodiment 1 of the present invention. Hereinafter, the configuration of the loudspeaker 100 will be described with reference to FIG.

  A loudspeaker 100 shown in FIG. 6 includes a delay-and-sum microphone array 110, a Griffith-Jim type adaptive beamformer 120, a decorrelation circuit 130, an adaptive filter 140, an amplitude limiting circuit 150, a digital analog (D / A). ) A converter 160, a power amplifier 170, and a speaker 180.

  The delay sum microphone array 110 includes microphones 111-0 to 111-N, microphone amplifiers 112-0 to 112-N, analog / digital (A / D) converters 113-0 to 113-N, and FIR filters 114-0 to 114. -N and adder 115 are provided, and a main lobe is formed in the direction of the speaker, thereby forming a directivity that maximizes the gain in the direction of the speaker.

  The microphones 111-0 to 111-N convert the sound including the loud sound output from the speaker 180 into an analog sound signal. Analog audio signals output from the microphones 111-0 to 111-N are amplified by the microphone amplifiers 112-0 to 112-N and input to the A / D converters 113-0 to 113-N. In describing the operation of the loudspeaker 100, the audio signal may be considered in the same manner as the noise inside the amplifier (power amplifier, microphone amplifier) or the background noise in the room, which becomes a howling excitation signal during silence. Does not clearly indicate a human voice signal and speaker input from the microphones 111-0 to 111-N.

  The A / D converters 113-0 to 113-N convert analog audio signals output from the microphone amplifiers 112-0 to 112-N into digital audio signals. The digital audio signal is input to the FIR filters 114-0 to 114-N.

  The FIR filters 114-0 to 114-N are designed so that the value obtained by adding the propagation delay proportional to the distance between the speaker and each microphone and the group delay of each FIR filter is constant. The FIR filters 114-0 to 114-N function as delay circuits that adjust the timing for aligning the phases of the digital audio signals output from the A / D converters 113-0 to 113-N and adding them by the adder 115. To do. The digital audio signal whose phase and timing are adjusted is input to the adder 115 and the adders 121-0 to 121- (N-1). There are various methods for realizing a delay circuit having an arbitrary group delay using an FIR filter, but the simplest methods are Toshiro Ohga, Yoshio Yamazaki, Yutaka Kaneda, "Acoustic System and Digital Processing", edited by IEICE. Corona, p. The method described in 215-216 may be used.

  The adder 115 adds the digital audio signals output from the FIR filters 114-0 to 114 -N, and inputs the addition result to the delay circuit 124.

  The Griffith-Jim type adaptive beamformer 120 includes adders 121-0 to 121- (N-1), adaptive filters 122-0 to 122- (N-1), an adder 123, a delay circuit 124, and an adder 125. The delay sum microphone array 110 forms a null in the directivity characteristic in the speaker direction while maintaining the directivity having the main lobe in the speaker direction.

  The adders 121-0 to 121- (N-1) cancel the speech component of the speaker by obtaining the difference between the output signals of adjacent FIR filters, and the obtained difference is applied to the adaptive filters 122-0 to 122- ( N-1).

The adaptive filters 122-0 to 122-(N−1) add a signal approximating the loudspeaker signal component from the speaker 180 included in the output signal of the delay sum microphone array 110 delayed by the delay circuit 124. It is generated from the signal output from −0 to 121- (N−1). The adaptive filters 122-0 to 122-(N−1) are signals that approximate the loudspeaker signal component from the desired signal d [n] output from the delay circuit 124 in the adder 125 and signals that approximate the howling sound described later. The filter coefficient update operation (adaptive operation) is performed so that the energy of the error signal e [n] obtained by subtracting is minimized. As a result, the microphone array forms a directional characteristic having a null in the speaker direction. The adaptive filters 122-0 to 122- (N-1) are composed of a filter coefficient variable FIR filter circuit and a filter coefficient update circuit. The filter coefficient update circuit may be realized by the filter coefficient update circuit shown in FIG. 7 using the most common LMS (Least Mean Square) algorithm. The filter coefficient update circuit shown in FIG. 7 performs the filter coefficient update calculation of the following equation (1).

Where x [n−i] is the input signal of the filter coefficient updating circuit, e [n] is the error signal of the adaptive filter, μ is the step size parameter of the LMS algorithm, W _{n + 1} [i] is the updated filter coefficient, reg Is a register holding the filter coefficient Wn [i], variable n is time, and variable i is the tap position of the FIR filter.

  As an adaptive algorithm for the adaptive filter, an NLMS (Normalized LMS) algorithm, a projection algorithm, or the like may be used in addition to the above LMS algorithm.

  FIG. 6 will be referred to again. The decorrelation circuit 130 outputs an output signal of the adaptive system fed back to the input side, that is, a residual signal e [n] that is an output signal from the adder 125 and a digital audio signal x [that is an input signal of the adaptive system. n] (hereinafter referred to as the input signal x [n]), the residual signal e [n] is transformed so as to reduce the correlation with the input signal x [n] and is output to the amplitude limiting circuit 150. Thereby, the convergence characteristic of the adaptive filter 140 can be improved. The decorrelation circuit 130 is generally a delay circuit or a frequency shift circuit that can be easily mounted on the circuit, but may be a pitch shift circuit, an echo circuit, a reverb circuit, or the like.

  The adaptive filter 140 forms an adaptive howling canceller together with the adder 125. The adaptive howling canceller is reproduced from the speaker 180 and included in the desired signal d [n] output from the delay circuit 124, and the microphones 110-0 to 111. A signal approximating the howling sound input to -N is generated from the input signal x [n] output from the amplitude limiting circuit 150, and the adder 125 approximates the howling sound from the desired signal d [n] and a loudspeaker. The adaptive operation is performed so that the energy of the error signal e [n] obtained by subtracting the signal component is minimized. The adaptive filter 140 includes an FIR filter circuit having a variable filter coefficient and a filter coefficient update circuit. The filter coefficient update circuit is realized by the filter coefficient update circuit shown in FIG. 7 using the most common LMS algorithm. do it. As an adaptive algorithm for the adaptive filter, an NLMS (Normalized LMS) algorithm, a projection algorithm, or the like may be used in addition to the above LMS algorithm.

  The amplitude limiting circuit 150 includes a D / A converter 160, a power amplifier 170, a speaker 180, a microphone 111-0 to 111-N, a microphone amplifier 112-0 to 112-N, and an A / D converter 113-0 to 113-. The amplitude value of the input signal x [n] is limited to a certain value or less so that N always operates in the linear operation region. The limit values of the amplitude limiting circuit 150 are the D / A converter 160, the power amplifier 170, the speaker 180, the microphones 111-0 to 111-N, the microphone amplifiers 112-0 to 112-N, and the A / D converter 113-0. -113-N is set to the maximum value that operates in the linear operation region without being saturated. Specifically, if the absolute value of the amplitude of the output signal of the decorrelation circuit 130 is equal to or smaller than the threshold value K, the amplitude limiting circuit 150 operates in the linear region and outputs the input signal x [n] as it is, and the decorrelation is performed. If the absolute value of the amplitude of the output signal of the circuit 130 is larger than the threshold value K, the amplitude of the output signal of the decorrelation circuit 130 in a non-linear operation is limited to -K or K, and then the input signal x [n] is set. Output. The limit value setting method will be described later.

  The D / A converter 160 converts the input signal (digital audio signal) x [n] output from the amplitude limiting circuit 150 into an analog audio signal. The power amplifier 170 amplifies the analog audio signal output from the D / A converter 160.

  The speaker 180 reproduces the analog received audio signal output from the power amplifier 170 and outputs it as audio. The loud sound output from the speaker 180 is input to the microphones 111-0 to 111-N.

  In the above-mentioned Griffith-Jim type adaptive beamformer, the number of directional characteristic nulls that can be formed depends on the number of microphones, and N nulls can be realized with respect to the number N + 1 of microphones. Therefore, even when there are a plurality of speakers, if the number of speakers is N or less, it is possible to form a null in the arrival direction of the loud sound from each speaker.

  In addition, the Griffith-Jim type adaptive beamformer has not only the direct sound component from the speaker but also the arrival direction of the initial reflected sound if the number of speakers is N-1 or less with respect to the number N + 1 of the microphones. Can form a null.

  Further, the Griffth-Jim type adaptive beamformer can reduce the gain in the speaker direction while maintaining the gain in the speaker direction of the microphone, so that the howling suppression performance is improved accordingly. In addition, since the adaptive operation is continuously performed, even if the direction of the speaker seen from the microphone changes, the directivity null is formed following the change automatically, and howling suppression can be performed stably. it can.

  Since the Griffth-Jim type adaptive beamformer is incorporated in a closed loop to which feedback is applied, if the system gain of the loudspeaker system exceeds 0 dB, there is no input audio signal (silent state or In the silent state), the adaptive operation proceeds using the background noise in the room or the internal noise of the power amplifier and the microphone amplifier as excitation signals.

  As described above, the loudspeaker shown in FIG. 6 uses a common error signal for the adaptive filters 122-0 to 122- (N-1) of the Griffith-Jim type adaptive beamformer 120 and the adaptive filter 140 of the adaptive howling canceller. Operate simultaneously and in parallel using e [n]. Therefore, even when a change in the speaker direction seen from the microphone, a change in the distance between the microphone and the speaker, and even a change in the impulse response that occurs when a person blocks the microphone and the speaker, occur at the same time. The adaptive filter 122-0 to 122- (N-1) adaptively operates with respect to the direction change and the adaptive filter 140 with respect to the distance change and the impulse response change so that the howling is suppressed well. It operates adaptively at the same time.

  The effect of howling suppression can be obtained not only from the Griffith-Jim type adaptive beamformer but also from a howling canceller using an adaptive filter. Even when the convergence of the adaptive filters 122-0 to 122- (N-1) of the Griffith-Jim type adaptive beamformer is slow, the effect of howling suppression by the adaptive filter 140 of the howling canceller can be obtained.

  Then, while the howling is suppressed by the adaptive filter 140, the adaptive filters 122-0 to 122- (N-1) of the Griffith-Jim type adaptive beamformer converge to secure a larger howling margin. .

Here, a setting method of the limit value of the amplitude limiting circuit 150 shown in FIG. 6 will be described. In the following description, the resolution of the D / A converter 160 and the resolution of the A / D converters 113-0 to 113-N are the same, and the maximum output signal level of the D / A converter 160 and the A / D converter 113- It is assumed that the maximum input signal level of 0 to 113-N is also the same. Further, the amplitude limiting circuit 150 has input / output characteristics represented by the following equation (2).

  In equation (2), in [i] is an input signal, x [i] is an output signal, i is an argument representing time, and K is a limit value.

  1. First, the open loop gain of the loudspeaker system is measured. The feedback path (FBP) is disconnected and a test signal such as a sine wave is input to the D / A converter 160, and the ratio between the amplitude of the input signal and the amplitude of the output signal of the A / D converters 113-0 to 113-N Is the open loop gain.

  2. When the open loop gain measured in the previous section is G and the allowable input signal level of the D / A converter 160 is −K to K, the limit value of the amplitude limiting circuit 150 is set to K / G, and the amplitude limiting circuit 150 Output signal level, that is, the input signal level of the D / A converter 160 is set to −K / G to K / G.

  By setting the limit value of the amplitude limiting circuit 150, the D / A converter 160 and the A / D converters 113-0 to 113-N operate without being saturated, and the maximum output signal level of the D / A converter 160 is increased. The maximum input signal level of the power amplifier 170 is obtained. Therefore, in order to prevent the power amplifier 170 from being saturated, the power amplifier 170 having a maximum input signal level higher than the maximum output signal level of the D / A converter 160 may be selected and used.

  The product of the maximum input signal level of power amplifier 170 and the gain of power amplifier 170 is the maximum output signal level of power amplifier 170. Therefore, in order for the speaker 180 to operate in a linear region without being saturated, the speaker 180 having a maximum input signal level higher than the maximum output signal level of the power amplifier 170 may be selected and used.

  Since it is difficult to actually measure the gain (attenuation amount) of the acoustic system between the speaker 180 and the microphones 111-0 to 111-N, in order for the microphones 111-0 to 111-N to operate without being saturated, the following is required. Thus, the microphones 111-0 to 111-N having appropriate characteristics may be selected.

  1. While the feedback path (FBP) is disconnected, a test signal having the same maximum amplitude as the previously set limiter value is input to the D / A converter 160. A triangular wave or a sine wave may be used as the test signal.

  2. By observing the output signal waveforms of the microphones 111-0 to 111-N, it can be determined whether or not the microphones 111-0 to 111-N are saturated. If saturated, the deficiency in the allowable input signal level can be found from the change in the output waveform. Therefore, a microphone having a maximum input signal level greater than the deficiency may be selected and used as the microphones 111-0 to 111-N. .

  In order to prevent the microphone amplifiers 112-0 to 112-N from being saturated, the maximum input signal level of the A / D converters 113-0 to 113-N is set to the gain of the microphone amplifiers 112-0 to 112-N. The microphone amplifiers 112-0 to 112-N having a maximum input signal level larger than the value divided by the above may be selected and used.

  In this way, the limit value of the amplitude limiting circuit 150 is set, and the D / A converter 160, the power amplifier 170, the speaker 180, the microphones 111-0 to 111-N, the microphone amplifiers 112-0 to 112-N, A / By determining characteristics that operate in the linear region without saturating all the components of the D converters 113-0 to 113-N, the impulse response of the acoustic system changes suddenly when the loudspeaker 100 is activated. Even in this case, it is possible to prevent each component from being saturated and causing nonlinear distortion, and thereby preventing the convergence of the adaptive filter from being hindered.

  Next, the arrangement of the microphone array will be described. Here, as shown in FIG. 8, it is assumed that speakers are arranged on both sides of the speaker. In such a case, the microphone array is arranged in a straight line in the direction that the speaker is facing, so that the distance between the speaker and each microphone is small relative to the size of the microphone array. Even when a sharp main lobe cannot be formed, a directional characteristic having a relatively good null can be formed in the speaker directions on both sides. Therefore, such a microphone array arrangement is suitable for a handheld microphone.

  As described above, according to the present embodiment, the filter coefficient update operation of the adaptive filter of the Griffith-Jim type adaptive beamformer and the adaptive filter of the howling canceller is simultaneously performed in parallel using the common error signal. As a result, even if any of the change in the speaker direction seen from the microphone, the change in the distance between the microphone and the speaker, the change in the impulse response between the microphone and the speaker occurs, or even when all the changes occur simultaneously, Since a plurality of adaptive filters converge in parallel operation, howling can be stably suppressed. That is, according to the present embodiment, it is possible to stably suppress howling even when a sudden change or a continuous change in the acoustic impulse response occurs without increasing the circuit scale and the processing amount. it can.

  Further, according to the present embodiment, it is possible to expand a better howling margin as compared with a case where either a Griffith-Jim type adaptive beamformer or a howling canceller using an adaptive filter is used alone.

  In the present embodiment, as an example of the arrangement of the microphone array, each microphone is linearly arranged in the direction facing the speaker. However, the present invention is not limited to this, and as shown in FIG. The microphones may be arranged in a straight line perpendicular to the direction in which the speaker is facing. In this case, if the distance between the speaker and the speaker is sufficiently long with respect to the size of the microphone array, the propagation delay time of the acoustic system from the speaker to the individual microphones can be regarded as constant. Therefore, the delay using the FIR filter The circuit can be omitted. Such a microphone array arrangement is suitable for a microphone whose position is fixed.

(Embodiment 2)
FIG. 10 is a block diagram showing the configuration of the loudspeaker 200 according to Embodiment 2 of the present invention. Hereinafter, the configuration of the loudspeaker 200 will be described with reference to FIG.

  The adaptive filter 210 constitutes an adaptive howling canceller together with the adder 220. The adaptive howling canceller is reproduced from the speaker 180 and included in the desired signal d1 [n] output from the adder 115, and the microphones 111-0 to 111. A signal approximating the howling sound input to −N is generated from the input signal x [n] output from the amplitude limiting circuit 150, and the adder 220 subtracts the signal approximating the howling sound from the desired signal d1 [n]. The adaptive operation is performed so that the energy of the error signal e1 [n] is minimized.

  The Griffith-Jim type adaptive beamformer 230 includes adders 121-0 to 121- (N-1), adaptive filters 122-0 to 122- (N-1), an adder 123, a delay circuit 124, and an adder 231. Have.

  The adaptive filters 122-0 to 122-(N−1) add a signal approximating the loudspeaker signal component from the speaker 180 included in the output signal of the delay sum microphone array 110 delayed by the delay circuit 124. The adder 231 uses the error signal e1 [n] generated from the signals output from −0 to 121- (N−1) and delayed by the delay circuit 124 as the desired signal d2 [n]. ], The filter coefficient update operation (adaptive operation) is performed so that the energy of the error signal e2 [n] obtained by subtracting the signal approximating the loudspeaker signal component is minimized. As a result, the microphone array forms a directional characteristic having a null in the speaker direction.

  Thus, according to the present embodiment, the filter coefficient updating operations of the adaptive filter of the Griffith-Jim type adaptive beamformer and the adaptive filter of the howling canceller are performed independently. As a result, even if any of the change in the speaker direction seen from the microphone, the change in the distance between the microphone and the speaker, the change in the impulse response between the microphone and the speaker occurs, or even when all the changes occur simultaneously, Since a plurality of adaptive filters operate and converge, howling can be stably suppressed. That is, according to the present embodiment, it is possible to stably suppress howling even when a sudden change or a continuous change in the acoustic impulse response occurs without increasing the circuit scale and the processing amount. it can.

(Embodiment 3)
FIG. 11 is a block diagram showing a configuration of a loudspeaker 300 according to Embodiment 3 of the present invention. Hereinafter, the configuration of the loudspeaker 300 will be described with reference to FIG.

  The Griffith-Jim type adaptive beamformer 310 includes adders 121-0 to 121- (N-1), adaptive filters 311-0 to 311- (N-1), an adder 123, a delay circuit 124, and an adder 312. Have.

  The adaptive filters 311-0 to 311-(N−1) add a signal approximating the loudspeaker signal component from the speaker 180 included in the output signal of the delay sum microphone array 110 delayed by the delay circuit 124. An error signal e1 that is generated from the signals output from −0 to 121- (N−1) and that is obtained by subtracting in the adder 312 a signal that approximates the loudspeaker signal component from the desired signal d1 [n] delayed by the delay circuit 124. The filter coefficient update operation (adaptive operation) is performed so that the energy of [n] is minimized. As a result, the microphone array forms a directional characteristic having a null in the speaker direction.

  The adaptive filter 320 forms an adaptive howling canceller together with the adder 330. The adaptive howling canceller sets the error signal e1 [n] as the desired signal d2 [n] and reproduces it from the speaker 180 included in the desired signal d2 [n]. Then, a signal approximating the howling sound input to the microphones 111-0 to 111-N is generated from the input signal x [n] output from the amplitude limiting circuit 150, and the adder 330 generates the desired signal d2 [n]. The adaptive operation is performed so that the energy of the error signal e2 [n] obtained by subtracting the signal approximating the howling sound is minimized.

  Thus, according to the present embodiment, the filter coefficient updating operations of the adaptive filter of the Griffith-Jim type adaptive beamformer and the adaptive filter of the howling canceller are performed independently. As a result, even if any of the change in the speaker direction seen from the microphone, the change in the distance between the microphone and the speaker, the change in the impulse response between the microphone and the speaker occurs, or even when all the changes occur simultaneously, Since a plurality of adaptive filters operate and converge, howling can be stably suppressed. That is, according to the present embodiment, it is possible to stably suppress howling even when a sudden change or a continuous change in the acoustic impulse response occurs without increasing the circuit scale and the processing amount. it can.

  In each of the above embodiments, the delay-and-sum microphone array has been described as an example. However, the present invention is not limited to this, and a filter-and-sum type microphone array is used instead of the delay-and-sum microphone array. Also good.

  In addition, when the microphones, microphone amplifiers, and A / D converters that are used in plurality in the above-described embodiments all have the same characteristics, the limit value of the amplitude limiting circuit is set as follows. Work can be simplified. Limit values are set so that the microphone closest to the speaker, the subsequent microphone amplifier, and the A / D converter do not saturate. As a result, without checking whether all microphones, microphone amplifiers, and A / D converters operate linearly, other microphones at a position where the distance from the speaker is larger and microphone amplifiers in the subsequent stages, A / D The converter can also be avoided from being fully saturated.

  The present invention is suitable for use in a howling canceller for a loudspeaker, a howling canceller for a hearing aid, and the like.

100, 200, 300 Loudspeaker 110 Delay-sum microphone array 111-0 to 111-N Microphone 112-0 to 112-N Microphone amplifier 113-0 to 113-N A / D converter 114-0 to 114-N FIR filter 115, 121-0 to 121- (N-1), 123, 125, 220, 231, 312, 330 Adder 120, 230, 310 Griffith-Jim type adaptive beamformer 122-0 to 122- (N-1) , 140, 210, 311-0 to 311- (N-1), 320 Adaptive filter 124 Delay circuit 130 De-correlation circuit 150 Amplitude limiting circuit 160 D / A converter 170 Power amplifier 180 Speaker

Claims (4)

A howling canceller incorporating a microphone array and a Griffith-Jim type adaptive beamformer,
An error signal obtained by using the desired signal obtained by the microphone array to approximate a howling sound reproduced from a speaker and input to the microphone, and using the desired signal and a signal approximating the howling sound. A first adaptive filter that updates filter coefficients so that the energy of
The Griffith-Jim type adaptive beamformer is
A signal approximating the loudspeaker signal component from the speaker included in the output signal of the microphone array is generated, and the error signal energy obtained using the desired signal and the signal approximating the loudspeaker signal component is minimized. Comprises a second adaptive filter for performing a filter coefficient updating operation,
The error signal is obtained by subtracting a signal approximating the howling sound and a signal approximating the loudspeaker signal component from the desired signal.
Howling canceller. A howling canceller incorporating a microphone array and a Griffith-Jim type adaptive beamformer,
A signal approximated to a howling sound reproduced from a speaker and input to a microphone included in the first desired signal obtained by the microphone array is generated, and a signal approximating the howling sound is subtracted from the first desired signal. Comprising a first adaptive filter for updating filter coefficients so that the energy of the first error signal is minimized;
The Griffith-Jim type adaptive beamformer is
A signal that approximates the loudspeaker signal component from the speaker included in the output signal of the microphone array, the first error signal is the second desired signal, and the loudspeaker signal component is approximated from the second desired signal A second adaptive filter that performs a filter coefficient updating operation so that the energy of the second error signal obtained by subtracting is minimized.
Howling canceller. A howling canceller incorporating a microphone array and a Griffith-Jim type adaptive beamformer,
The Griffith-Jim type adaptive beamformer is
A first error obtained by generating a signal approximating the loudspeaker signal component from the speaker included in the output signal of the microphone array and subtracting the signal approximating the loudspeaker signal component from the first desired signal obtained by the microphone array A first adaptive filter that performs a filter coefficient update operation so that the energy of the signal is minimized;
Using the first error signal as a second desired signal, a signal that approximates a howling sound that is reproduced from a speaker and input to a microphone and is included in the second desired signal is generated, and the howling sound is generated from the second desired signal. A second adaptive filter that updates the filter coefficient so that the energy of the second error signal obtained by subtracting the approximate signal is minimized;
Howling canceller. The amplitude limiting circuit which limits the amplitude of an input signal below a predetermined threshold value which ensures that all the components which comprise the said howling canceller operate | move in a linear area | region is provided in any one of Claims 1-3. Howling canceller.

Images