A.R. Kizhanatham, N. Chandra, R.E. Yantorno, and S.J. Wenndt (USA)
Co-channel speech occurs when one speaker’s speech is corrupted by another speaker’s speech. A co-channel detection system could provide information to suspend the operation of any speech processing system whose operation would be degraded if it were processing co channel speech. In this paper we present two new methods of co-channel speech detection, one based on cyclostationarity and the other based on wavelet transform. Detection of co-channel speech in this paper refers to the detection of co-channel voiced speech, as it is not yet possible to detect unvoiced co-channel speech. Cyclostationary-based co-channel speech detection reveals that at least 65% of co-channel speech is correctly detected for different combinations of speech, e.g., male-male, female-female etc., with false alarms of approximately 24%. Investigation of the wavelet transform based co-channel speech detection reveals that at least 94% of co-channel speech is correctly detected with false alarms of approximately 28% making both methods tools for detecting co channel speech.
Important Links:
Go Back
