MAP Estimation of Whole-Word Acoustic Models with Dictionary Priors

The intrinsic advantages of whole-word acoustic modeling are offset by the problem of data sparsity. To address this, we present several parametric approaches to estimating intra-word phonetic timing models under the assumption that relative timing is independent of word duration. We show evidence that the timing of phonetic events is well described by the Gaussian distribution. We explore the construction of models in the absence of keyword examples (dictionary-based), when keyword examples are abundant (Gaussian mixture models), and also present a Bayesian approach which unifies the two. Applying these techniques in a point process model keyword spotting framework, we demonstrate a 55% relative improvement in performance for models constructed from few examples.