Generative factor analyzed HMM for automatic speech recognition

Abstract

We present a generative factor analyzed hidden Markov model (GFA-HMM) for automatic speech recognition. In a standard HMM, observation vectors are represented by mixture of Gaussians (MoG) that are dependent on discrete-valued hidden state sequence. The GFA-HMM introduces a hierarchy of continuous-valued latent representation of observation vectors, where latent vectors in one level are acoustic-unit dependent and latent vectors in a higher level are acoustic-unit independent. An expectation maximization (EM) algorithm is derived for maximum likelihood estimation of the model. We show through a set of experiments to verify the potential of the GFA-HMM as an alternative acoustic modeling technique. In one experiment, by varying the latent dimension and the number of mixture components in the latent spaces, the GFA-HMM attained more compact representation than the standard HMM. In other experiments with varies noise types and speaking styles, the GFA-HMM was able to have (statistically significant) improvement with respect to the standard HMM.