Regularization, Adaptation, and Non-Independent Features Improve Hidden Conditional Random Fields for Phone Classification
Abstract
We show a number of improvements in the use of Hidden Conditional Random Fields (HCRFs) for phone classification on the TIMIT and Switchboard corpora. We first show that the use of regularization effectively prevents overfitting, im- proving over other methods such as early stopping. We then show that HCRFs are able to make use of non-independent features in phone classification, at least with small numbers of mixture components, while HMMs degrade due to their strong independence assumptions. Finally, we successfully apply Maximum a Posteriori adaptation to HCRFs, decreas- ing the phone classification error rate in the Switchboard cor- pus by around 1% – 5% given only small amounts of adapta- tion data.