Two Microphone Noise Reduction for Speech Using VLSI LMS Adaptive Filters

Abstract

ABSTIUCTNoise reduction in speech using LMS adaptive filters has been dealt with by Widrow, et al, and several papers have dealt with the simulation of this method on a large computer. Because of the computational requirements, most of the experiments were simulated on computer and have used short filters. The restriction on the filter length enabled only simple periodic interfering signals to be effectively cancelled.Implementation of a practical noise canceller in a more realistic acoustic environment is considerably complicated by such problems as the reverberation time of the room, multiple noise sources, and the computational requirements of implementing filters that are sufficiently long as to be effective.'Ihis paper reports the use of VLSI adaptive filters and a DSP processor for the practical implementation of an adaptive noise canceller using a filter of typically 256 taps or longer.'I'his system was built and several experiments in successively more coniplicated acoustic environments were carried out. ?lie experiments are reported and interpreted to provide an insight into factors affecting performance such as adaption speed of the filter, filter length, type and number of noise sources, reverberation time of the room, the effect of time varying acoustic conditions and the spatial significance of using only two microphones.