Maximum Likelihood Model Adaptation Using Piecewise Linear Transformation for Robust Speech Recognition

This paper presents a new model adaptation algorithm using piecewise linear transformation (PLT) for robust speech recognition. In this algorithm, the nonlinear relationship between training and testing mean vectors is approximated by a set of piecewise linear transformations. The PLT coefficients are estimated from adaptation data by the expectation-maximization (EM) algorithm and maximum likelihood (ML) criterion. The proposed algorithm could overcome the limitation of linear assumption in traditional transform-based adaptation algorithm. The experimental results show that the proposed approach is efficient and outperforms the linear model adaptation algorithm.