Maxium likelihood non-linear transformation for environment adaptation in speech recognition systems

Abstract

In this paper, we describe an adaptation method for speech recognition systems that is based on a piecewise-linear approximation to a non-linear transformation of the feature space. The method extends a previously proposed non-linear transformation (NLT) technique by making the transformation function more sophisticated (piecewise-linear instead of piecewiseconstant), and by computing the transformation to maximize the likelihood of the adaptation data given its transcription (instead of just matching the global statistics of the test and training data). This method also differs from other linear techniques (such as MLLR, linear feature space transforms, etc.) in two ways - first, the computed transformation is non-linear, second, the tying structure of the transformation depends not on the phonetic class but rather on the location in the feature space. Experimental results show that the method performs well for the case of limited adaptation data, and the performance gains appear to be additive to those provided by MLLR - yielding upto 3.4% relative improvement over MLLR.