Monaural voiced speech segregation based on combined cues and energy distribution

Monaural speech segregation is important for speech signal processing, and it has been extensively studied on the basis of auditory scene analysis principles. However, current segregation algorithms can not achieve satisfactory performance in high frequency range. In this paper, we propose a system for monaural voiced speech segregation, in which two novel ideas are investigated. First, combined cues (including cross-channel correlation, temporal continuity, and onset/offset) are employed to generate segments in high frequency range. Second, the energy distribution of mixed signal is employed to indicate the reliabilities of cues in high frequency range, according to which, an alternative segmentation strategy is performed. Systematic evaluation and comparison show that the proposed system produces improvement on SNR gain.