Model Adaptation Based on Improved Variance Estimation for Robust Speech Recognition

This paper proposes a model adaptation algorithm based on improved variance estimation for noise robust speech recognition. In this algorithm, the approximate closed-form variance estimation is extended from the feature space to the model space and the dynamic parameters of the hidden Markov model (HMM) as well as the static parameters are converted to testing conditions. The experimental results show that the proposed model adaptation algorithm can converge quickly and outperforms the feature compensation method using the approximate closed-form variance estimation.