Audio signals are digitized and an FFT is conducted on samples of the digitized signals to produce corresponding frequency spectrums. These spectrums are analyzed, such as by determining one or more peak frequency magnitudes which are 33 dB greater than harmonics or subharmonics of the frequency in a plurality of several successive spectrums, to detect resonating feedback frequencies. The offending frequency is then filtered in the time domain, either in the digitized form or analog form, to eliminate the feedback.