9th Annual Conference of the International Speech Communication Association

Brisbane, Australia
September 22-26, 2008

HMM-Based Estimation of Unreliable Spectral Components for Noise Robust Speech Recognition

Bengt J. Borgström, Abeer Alwan

University of California at Los Angeles, USA

This paper presents a novel approach for reconstructing unreliable spectral components, which utilizes HMM-based missing feature algorithms, and applies them to noise robust speech recognition. The proposed technique uses the forward-backward algorithm to estimate corrupt spectrographic data based on nearby reliable features, noisy observations, and on an underlying statistical model. The estimation process can be applied based on intra-channel information, intra-feature information, or a combination of both. The overall system is shown to provide vast improvements for the Consonant Challenge Database [1], for both MFCCs and PLP features, when using an oracle mask. Moreover, through downsampling of statistical models [2], the required complexity of the system is greatly reduced with negligible effects on results.


  2. B. J. Borgström and A. Alwan, An Efficient Approximation of the Forward-Backward Algorithm to Deal With Packet Loss, With Applications to Remote Speech Recognition, Proc. of ICASSP 2008.

Full Paper

Bibliographic reference.  Borgström, Bengt J. / Alwan, Abeer (2008): "HMM-based estimation of unreliable spectral components for noise robust speech recognition", In INTERSPEECH-2008, 1769-1772.