Variational Approximation of Long-span Language Models for LVCSR

Recurrent Neural Network, Language Model, Variational Inference

We have presented experimental evidence that (n-gram) variational approximations of long-span LMs yield greater accuracy in LVCSR than standard n-gram models estimated from the same training text.

Long-span language models that capture syntax and semantics are seldom used in the first pass of large vocabulary continuous speech recognition systems due to the prohibitive search-space of sentencehypotheses. Instead, an N-best list of hypotheses is created using tractable n-gram models, and rescored using the long-span models. It is shown in this paper that computationally tractable variational approximations of the long-span models are a better choice than standard n-gram models for first pass decoding. They not only result in a better first pass output, but also produce a lattice with a lower oracle word error rate, and rescoring the N-best list from such lattices with the long-span models requires a smaller N to attain the same accuracy. Empirical results on the WSJ, MIT Lectures, NIST 2007 Meeting Recognition and NIST 2001 Conversational Telephone Recognition data sets are presented to support these claims.