Bayesian phonotactic language model for Acoustic Unit Discovery

Recent work on Acoustic Unit Discovery (AUD) has led to the development of a non-parametric Bayesian phone-loop model where the prior over the probability of the phone-like units is assumed to be sampled from a Dirichlet Process (DP). In this work, we propose to improve this model by incorporating a Hierarchical Pitman-Yor based bigram Language Model on top of the units transitions. This new model makes use of the phonotactic context information but assumes a fixed number of units. To remedy this limitation we first train a DP phoneloop model to infer the number of units, then, the bigram phone-loop is initialized from the DP phone-loop and trained until convergence of its parameters. Results show an absolute improvement of 1-2%on the Normalized Mutual Information (NMI) metric. Furthermore, we show that, combined with Multilingual Bottleneck (MBN) features the model yields a same or higher NMI as an English phone recogniser trained on TIMIT.