Neural network enhancement for a two speaker separation system

A multilevel perceptron has been used to perform critical classification tasks within an automated two-speech separation system. The neural network provides decisions as to the number of simultaneous speakers and their voicing state for each time frame of speech data. A small database of actual mixed speech has been recorded and digitized for evaluating the performance of the mixed speech separator. Results of tests made using these samples indicate that neural networks can make accurate judgments as to the nature of a mixed-speech waveform.<<ETX>>

[1]  Brian A. Hanson,et al.  The harmonic magnitude suppression (EMS) technique for intelligibility enhancement in the presence of interfering speech , 1984, ICASSP.

[2]  T. W. Parsons Separation of speech from interfering speech by means of harmonic selection , 1976 .

[3]  T. F. Quatieri,et al.  An approach to co-channel talker interference suppression using a sinusoidal model for speech , 1988, ICASSP-88., International Conference on Acoustics, Speech, and Signal Processing.

[4]  K. Min,et al.  Automated two speaker separation system , 1988, ICASSP-88., International Conference on Acoustics, Speech, and Signal Processing.