Ninth International Conference on Spoken Language Processing

Pittsburgh, PA, USA
September 17-21, 2006

Performance Evaluation of Three Features for Model-Based Single Channel Speech Separation Problem

M. H. Radfar (1), R. M. Dansereau (1), A. Sayadiyan (2)

(1) Carleton University, Canada; (2) Amirkabir University of Technology, Iran

This paper addresses the efficiency of three features for the modelbased single channel speech separation problem. The separability of three features: log spectrum, modulated lapped transform (MLT) coefficients, and a fusion of pitch and envelop information are evaluated using a VQ-based speech separation technique. At the core of this approach are two trained codebooks of the quantized feature vectors of speakers, whereby the main evaluation for separation is performed. The experiments are conducted in two different scenarios: speakerdependent and speaker independent. The results show that the log spectrum outperforms the other features for speaker-dependent scenario. However, for the speaker-independent scenario, the best results are obtained from applying the pitch-envelop feature.

Full Paper

Bibliographic reference.  Radfar, M. H. / Dansereau, R. M. / Sayadiyan, A. (2006): "Performance evaluation of three features for model-based single channel speech separation problem", In INTERSPEECH-2006, paper 2005-Thu2FoP.9.