Speech Recognition using SVMs

Abstract

An important issue in applying SVMs to speech recognition is the ability to classify variable length sequences. This paper presents extensions to a standard scheme for handling this variable length data, the Fisher score. A more discriminatory mapping is introduced based on the likelihood-ratio. The score-space de ned by this mapping avoids some of the problems with the Fisher score. It also allows the discriminative power of the generative models to be directly incorporated into the score-space. The mapping schemes, and appropriate normalisation schemes, are evaluated on a speaker-independent isolated letter task where the new mapping outperforms both the Fisher score and standard HMMs.