Single and multiple channel approaches to speech enhancement

A description of two recently developed techniques for enhancing speech is presented. The first is a single channel approach where an optimum filter is found given a priori knowledge of the noise and the noisy speech models. The filter computation involves the solution of a single diophantine equation. The second is a two channel approach which can be extended to multiple channels. The microphones are placed 10 cm apart and the acoustic path transfer function between the two microphones is identified during a noise only period. The resulting output is shown to be a noise reduced but slightly filtered speech signal. Results are given to compare the two approaches in terms of SNR improvements and quality of the resulting processed signal