Usable Speech Detection using the Modified Spectral Autocorrelation Peak to Valley Ratio using the LPC Residual

N. Chandra and R.E. Yantorno (USA)


Co-channel speech, usable speech,SAPVR, SAPVR-Residual, LPC residual.


Many applications of speech communication and speaker identification suffer from the problem of interfering or co-channel speech. The proposed method is to find the segments of co-channel speech, which can be identified as “usable” and could then be processed by a speech processing system. The Spectral Autocorrelation Peak to Valley Ratio of the Linear Predictive Coding Residual (SAPVR-Residual) is a modified version of the Spectral Autocorrelation Peak to Valley Ratio (SAPVR) used for detection of usable speech segments in co-channel speech. Results obtained using the SAPVR-Residual, were useful in spotting approximately 71% of usable segments, with a corresponding false alarm rate of 37%. This method of identification of usable speech represents the front-end process of a next generation co-channel speech processing system, involving an information fusion/decision system whose goal is speech separation.

Important Links:

Go Back