13th Annual Conference of the International Speech Communication Association

Portland, OR, USA
September 9-13, 2012

Speech Recognition by Denoising and Dereverberation Based on Spectral Subtraction in a Real Noisy Reverberant Environment

Kyohei Odani (1), Longbiao Wang (2), Atsuhiko Kai (2)

(1) Graduate School of Engineering; (2) Faculty of Engineering;
Shizuoka University, Hamamatsu, Japan

A blind dereverberation method based on spectral subtraction (SS) using a multi-channel least mean squares (MCLMS) algorithm was previously proposed. The results of a large vocabulary continuous speech recognition (LVCSR) task showed that this method achieved significant improvements over the conventional method based on cepstral mean normalization (CMN) and beamforming in a simulated reverberant environment without additive noise. In this paper, we evaluate the blind dereverberation method in a real noisy reverberant environment. We present a denoising and dereverberation method based on power SS or generalized SS (GSS), and evaluate our proposed method using speech in a real environment.The GSS-based method achieves an average relative word error reduction rate of 39.1% and 11.5% compared to the conventional CMN and power-SS based methods, respectively.

Index Terms: hands-free speech recognition, blind dere- verberation, noise reduction, spectral subtraction, real en- vironment

Full Paper

Bibliographic reference.  Odani, Kyohei / Wang, Longbiao / Kai, Atsuhiko (2012): "Speech recognition by denoising and dereverberation based on spectral subtraction in a real noisy reverberant environment", In INTERSPEECH-2012, 1251-1254.