Performance Comparison of HMM and VQ Based Single Channel Speech Separation

摘要

In this paper, single channel speech separation (SCSS) tech-niques based on hidden Markov models (HMM) and vectorquantization (VQ) are described and compared in terms of(a) signal-to-noise ratio (SNR) between separated and origi-nal speech signals, (b) preference of listeners, and (c) com-putational complexity. The SNR results show that the HMM-based technique marginally outperforms the VQ-based tech-nique by 0.85 dB in experiments conducted on mixtures offemale-female, male-male, and male-female speakers. Subjec-tive tests show that listeners prefer HMM over VQ for 86.70 %of test speech files. This improvement, however, is at the ex-pense of a drastic increase in computational complexity whencompared with the VQ-based technique.