An iterative least-squares technique for dereverberation

Kshitiz Kumar, Bhiksha Raj, Rita Singh, Richard M. Stern
2011 2011 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)  
Some recent dereverberation approaches that have been effective for automatic speech recognition (ASR) applications, model reverberation as a linear convolution operation in the spectral domain, and derive a factorization to decompose spectra of reverberated speech in to those of clean speech and room-response filter. Typically, a general non-negative matrix factorization (NMF) framework is employed for this. In this work 1 we present an alternative to NMF and propose an iterative least-squares
more » ... ative least-squares deconvolution technique for spectral factorization. We propose an efficient algorithm for this and experimentally demonstrate it's effectiveness in improving ASR performance. The new method results in 40-50% relative reduction in word error rates over standard baselines on artificially reverberated speech.
doi:10.1109/icassp.2011.5947601 dblp:conf/icassp/KumarRSS11 fatcat:3cy4q4hlfra27hxqrdgz3ulzdq