Understanding speech corrupted by noise

Intelligibility enhancement of speech signals in high levels of additive, wideband noise is a problem of significant interest and application. Consequently, a number of speech processing methods have been developed. Although a large number of methods have been proposed, an implementation that significantly increases the intelligibility of speech corrupted by high levels of wideband noise has not been demonstrated. It is the purpose of this paper to review the processing strategies and philosophies used in these methods and to make some observations based on this review that suggest a processing framework for speech intelligibility enhancement in noise. The paper concludes with a description of one implementation of this suggested framework.