Adaptive noise suppressor

A method of noise suppression for providing a noise filtered signal comprising the steps of:

• (a) delaying an input signal X by a fixed delay .DELTA.;
• (b) transforming the delayed signal into the frequency domain;
• (c) adaptably filtering the delayed, transformed signal using a plurality of weights and a mean square algorithm in the frequency domain to provide a filtered output signal;
• (d) transforming the filtered output signals into a time domain signal comprising the noise filtered signal Y;
• (e) subtracting the signal Y from X for obtaining a noise error signal .epsilon.;
• (f) transforming the noise error signal .epsilon. into the frequency domain to produce a transformed error signals;
• (g) multiplying the transformed noise error signals by a vector gain .mu.;
• (h) removing the effects of circular convolution from the multiplied noise error signal;
• and
• (i) feeding back the results of step (h) to adjust the weights in the mean square algorithm in a manner to minimize the difference between the signal X and the signal Y, thereby minimizing the error signal .epsilon. and producing noise suppressed signals in the noise filtered signal Y.

A feedback suppression device for providing feedback suppressed signals comprising:

• (a) means for delaying an input speech signal X by a fixed time delay .DELTA.;
• (b) first means for transforming the delayed signal into the frequency domain;
• (c) weighted means, having weights and having inputs connected to the first transforming means for adaptably filtering the delayed, transformed signal using a mean square algorithm in the frequency domain;
• (d) second means for transforming the output of the weighted means into a time domain signal comprising a signal Y;
• (e) means having as inputs the signals X and Y for subtracting the signal Y from X, the output of the subtracting means being a feedback suppressed signal .epsilon.;
• (f) third means for transforming the signal .epsilon. into the frequency domain to produce a transformed feedback suppressed signals;
• (g) means for multiplying the transformed feedback suppressed signal by a vector gain .mu.;
• and
• (h) projection operator means receiving the output of the multiplying means for removing the effects of circular convolution, the output of the projection operator means being fed back to the weighted means to cause the weights of the weighted means to be readjusted in a manner to minimize the difference between the signals X and Y, thereby minimizing the feedback suppressed signal .epsilon..

A method of feedback suppression to provide feedback suppressed signals comprising the steps of:

• (a) delaying an input signal X by a fixed time delay .DELTA.;
• (b) transforming the delayed signal into the frequency domain;
• (c) adaptably filtering the delayed, transformed signal using a plurality of weights and a mean square algorithm in the frequency domain to provide filtered output signals;
• (d) transforming the filtered output signals of the weighted means into a time domain signal comprising a signal Y;
• (e) subtracting the signal Y from X for obtaining a feedback suppressed signal .epsilon.;
• (f) transforming the signal .epsilon. into the frequency domain to produce a transformed feedback suppressed signals;
• (g) multiplying the transformed feedback suppressed signal by a vector gain .mu.;
• (h) removing the effects of circular convolution from the multiplied noise error signal;
• and
• (i) feeding back the results of step (h) to adjust the weights in the mean square algorithm in a manner to minimize the difference between the signals X and Y, thereby minimizing the feedback suppressed signal .epsilon..

A time domain feedback suppression device comprising:

• means for delaying a speech input signal X by a fixed delay .DELTA., .DELTA. being selected to be on the order of at least greater than a phonem;
• weighted means having weights and having inputs connected to the delaying means for adaptably filtering the delayed signal using a least-mean-square algorithm, the output of the filtering means comprising a signal Y;
• means, having as inputs the signals X and Y for subtracting the signal Y from X, the output of the subtracting means being a feedback suppressed speech signal .epsilon.;
• means having as an input signal .epsilon. for multiplying the signal by a gain .mu., the output of the multiplying means being fed back to the weighted means to cause the weights of the weighted means to be readjusted in a manner to minimize the difference between the signal X and the signal Y, thereby minimizing the feedback suppressed speech signal .epsilon..

A method of feedback suppression comprising the steps of:

• delaying a speech input signal X by a fixed delay .DELTA., .DELTA. being selected to be on the order of at least a phonem;
• adaptably filtering the delayed signal using a plurality of weights and a least-mean-square algorithm, the output of the filtering means comprising a signal Y;
• subtracting the signal Y from X, for providing a feedback suppressed speech signal .epsilon.;
• multiplying the feedback suppressed speech signal .epsilon. by a gain .mu.;
• and
• feeding back to the weighted means to cause the weights of the weighted means to be readjusted in a manner to minimize the difference between the signal X and the signal Y, thereby minimizing the feedback suppressed speech signal .epsilon..