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 , for both MFCCs and PLP features, when using an oracle mask. Moreover, through downsampling of statistical models , the required complexity of the system is greatly reduced with negligible effects on results.
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.