This paper describes two algorithms for separating two overlapping speech signals. Both rely heavily on accurate measurements of the pitch of the target voice. The first uses a single microphone, and an important feature is the exploitation of the onset of a voiced sound as an aid to the extraction of its pitch in the presence of interference. The second uses two microphones, and an important feature is the utilisation of the direction of the target voice.
Bibliographic reference. Denbigh, P. N. / Zhao, J. (1991): "Pitch extraction and separation of overlapping speech", In EUROSPEECH-1991, 545-548.