English
This is the third in a series of projects aimed at improving the intelligibility of speech that has been degraded by interference from a competing talker. It was found that the basic principle was applicable to natural speech, but that in the natural-speech environment, performance of all phases of the process fell off seriously. The principal goal of the second project was therefore to ruggedize all the key portions of the process. When this was done, the separator could handle natural speech, but the quality of the recovered speech, as subjectively judged by listeners, left a good deal to be desired. The speech was considered unpleasant to listen to and occasionally unintelligible. In the current project, therefore, we have principally addressed the problem of overcoming these deficiencies. We have located the main problem area in the routine for resolving overlapping frequency components, and we have identified and explored several methods for improving the performance of this routine.
Sign in with your free account to read, download EPUBs or listen.
DTIC Archive · Parsons,Thomas W · POLYTECHNIC INST OF NEW YORK BROOKLYN DEPT OF ELECTRICAL ENGINEERING AND ELECTROPHYSICS · *SIGNAL PROCESSING · *SPEECH RECOGNITION · *SPEECH TRANSMISSION · *CROSSTALK · FOURIER TRANSFORMATION · FREQUENCY MODULATION · HARMONICS