TRAINING DATA SELECTION BASED ON CONTEXT-DEPENDENT STATE MATCHING

In this paper we construct a data set for semi-supervised acous-tic model training by selecting spoken utterances from a massive collection of anonymized Google Voice Search utterances. Semi-supervised training usually retains high-confidence utterances which are presumed to have an accurate hypothesized transcript, a neces-sary condition for successful training. Selecting high confidence ut-terances can however restrict the diversity of the resulting data set. We propose to introduce a constraint enforcing that the distribution of the context-dependent state symbols obtained by running forced alignment of the hypothesized transcript matches a reference distri-bution estimated from a curated development set. The quality of the obtained training set is illustrated on large scale Voice Search recognition experiments and outperforms random selection of high-confidence utterances. Index Terms — data selection, semi-supervised training 1