Co-Channel Interference Reduction

摘要

Co-channel speaker interference occurs when the voice of one speaker is corruptedby the superposition of another speaker's voice on the same communications channel. The presence of co-channel interference in a communications scenario results in decreased intelligibility (for human listeners), both of the target speaker's speech as well as that of the interfering speaker. While humans can partially compensate for such interference, the performance of automatic speech processing systems, such as speech recognizers and speaker recognition systems, deteriorates drastically in the presence of such interference. This report describes a new methodology for speaker separation in the audio domain. This methodology is based on a frame-by-frame technique which is called the pole-partitioning separation algorithm.