An effective speech separation system which requires no a priori information

The separation of the speech signals of two male talkers communicating simultaneously on the same monophonic channel is addressed. This is sometimes referred to as the cochannel interference problem. No a priori information is available, and the average power of the two signals may differ by as much as 18 dB. Developments which led to the implementation of the improved speech separation system are described. These techniques include an improved model of voiced speech spectra and an algorithm for estimating the pitch of a speech signal which is being masked by a much louder speech signal.<>