On Usable Speech Detection by Linear Multi-Scale Decomposition for Speaker Identification

Abstract

Usable speech is a novel concept of processing co-channel speech data. It is proposed to extract minimally corrupted speech that is considered useful for various speech processing systems. In this paper, we are interested for co-channel speaker identification (SID). We employ a new proposed usable speech extraction method based on the pitch information obtained from linear multi-scale decomposition by discrete wavelet transform. The idea is to retain the speech segments that have only one pitch detected and remove the others. Detected Usable speech was used as input for speaker identification system. The system is evaluated on co-channel speech and results show a significant improvement across various Target to Interferer Ratio (TIR) for speaker identification system.