Abstract:
Speech recognition systems work reasonably well in laboratory conditions, but their performance deteriorates drastically when they are deployed in practical situations where the speech is corrupted by additive noise distortion. One way to improve the performance of a speech recognition system in the presence of noise is to enhance the speech signal (and remove noise) prior to its recognition. Recently, a singular value decomposition based technique has been proposed for speech enhancement [Dendrinos et al., Speech Commun. 10, 45--57 (1991)]. In this technique, singular value decomposition was applied to an overdetermined, overextended data matrix formed from the noisy speech signal and a noise-free, low-rank approximation was obtained by retaining a specific number of singular values. This technique was applied here as a preprocessor for recognizing speech in the presence of noise and was found to improve the recognition performance significantly, especially at lower signal-to-noise ratios. [Work supported by the Australian Research Council.]