A system has been developed that successfully separates a target voice from an overlapping voice. The algorithm is based mainly on monaural processing, exploiting the harmonicity of voiced speech, but also utilises some binaural information to initially allocate each separated sound to the appropriate speaker. Emphasis has been put on monaural processing because of the sensitivity of binaural cues to room reverberation.
Bibliographic reference. Shamsoddini, A. / Denbigh, P. N. (1995): "A system for speech separation", In EUROSPEECH-1995, 1583-1586.