Enhancing/Intelligibility of speech in noisy or multi-talker environments

Two of the most serious types of interference in speech communications are wideband noise and the speech of a competing talker. This report describes research on techniques for coping with these types of interference. INTEL, a technique for enhancing speech obscured by wideband noise, has now been improved to the point that it will be realistic to implement it as a real-time system for use in practical applications. A technique has also been developed for dealing with interference from a competing talker. (The current technique is restricted to vocalic utterances). Separation is done by selecting the components of the desired voice in the Fourier transform of the input. In implementing this process, techniques have been developed for resolving overlapping spectrum components, for determining pitches of both talkers, and for assuring consistent separation. This report describes the improvements in INTEL and the implementation of the two-talker process, and discusses the prospects for basing a general two-talker process on these techniques.