Psychoacoustic model compensation for robust continuous speech recognition in additive noise

Abstract

This paper addresses the problem of speech recognition in the presence of additive noise. It focuses on Psychoacoustic Model Compensation (Psy-Comp) scheme, which has been shown to be a powerful technique for noise robustness. It has further implemented model domain mean and variance normalization along with Psy-Comp to alleviate channel noise for robust continuous speech recognition in noisy conditions. The proposed algorithms are validated through experiments on noise corrupted TIMIT speech recognition database. We show that the Psy-Comp scheme along with model domain mean and variance normalization provide 9.5% performance gain compared to the Vector Taylor Series (VTS) scheme. Moreover, the computational cost of the proposed method is significantly less than the VTS scheme.