基于子带粒子滤波一种语音增强方法89210.pdfVIP

27 4 Vol.27 No.4 2006 4 Journal on Communications April 2006 , , , 116024 AR R-B AR R-B R-B TN911 A 1000-436X(2006)04-0023-06 Subband particle filtering for speech enhancement JIN Nai-gao, YIN Fu-liang, WANG Dong-xia, CHEN Zhe (School of Electronic and Information Engineering, Dalian University of Technology, Dalian 116024, China) Abstract: Rao-Blackwellised particle filtering provided an effective solution for speech enhancement. A novel method was proposed for speech enhancement based on rao-blackwellised particle filters and subband decomposition. In rao-blackwellised particle filters, the dimension of sampling space was reduced by subband decomposition. Subband speech signals were obtained by subband decomposition, then modeled as low-order time-varying AR processes. Rao-blackwellised particle filters were applied to estimate the parameters of AR model and filter the subband speech sig- nals. The enhanced fullband speech signals were finally reconstructed by synthesizing the enhanced subband speech sig- nals. Simulation results show that the proposed method not only has lower computational complexity, but also gives bet- ter speech enhancement performance. Key words: speech enhancement; rao-blackwellised particle filter; subband decomposition; cosine modulated filter banks 1