Linear Versus Nonlinear Multi-scale Decomposition for Co-channel Speaker Identification System
- Resource Type
- Authors
- Wajdi Ghezaiel; Amel Ben Slimane; Ezzedine Ben Braiek
- Source
- Recent Advances in Nonlinear Speech Processing ISBN: 9783319281070
Recent Advances in Nonlinear Speech Processing
- Subject
- Nonlinear system
Computer science
Speech recognition
Wavelet transform
False alarm
USable
Mixture model
Speech processing
Hilbert–Huang transform
Communication channel
- Language
Co-channel speech is a combination of speech utterances over a single communication channel. Traditional approach to co-channel speech processing is to attempt to extract the speech of the speaker of interest (target speech) from other (interfering) speech. Usable speech criteria are proposed to extract minimally corrupted speech for speaker identification in co-channel speech. In this paper, we present usable speech extraction method based on pitch information obtained from linear multi-scale decomposition by dyadic wavelet transform and nonlinear multi-scale decomposition by empirical mode decomposition. Detected usable speech are organized into speaker stream, and applied to speaker identification system. The proposed methods are evaluated and compared across various Target to Interferer Ratio (TIR) for speaker identification system.