Speech Recognition Project

Modern speech recognition systems are very complex. A good model must account for the context-sensitive, time-dependent, and speaker-dependent nature of "phones", the basic phonological units of speech. Typically, this variation is manually encoded in the model using domain knowledge, or not modeled at all.

We want to leverage the techniques used in our parsing work to automatically induce latent models which learn this variation from the data. We also hope to simplify the decoding phase of speech recognition by adapting the decoding techniques used in our parser to the speech domain. We believe we can vastly simplify both the learning and decoding phases without losing performance.

Publications

!IS_A_LIST Learning Structured Models for Phone Recognition, Slav Petrov, Adam Pauls, and Dan Klein, In proceedings of EMNLP-CoNLL 2007. [pdf] [slides] [bib]