Automatic separation of simultaneous speech of two talkers
Thomas W. Parsons · The Journal of the Acoustical Society of America · 1974
A common type of interference in speech transmission is that caused by the speech of a competing talker. While the brain is good at discerning individual voices in such a situation (the “cocktail party effect”, it relies heavily on the availability of binaural data. When voices compete over a single channel, separation is much more difficult and intelligbility is degraded. The present paper describes methods of attenuating this interference by selecting the components of the desired signal in the Fourier transform of the input. Implementation of these methods is described, and problem areas and limitations are discussed. Preliminary results are presented, with examples of separation achieved under certain simplifying constraints.