CLASSIFICATION AND RECOGNITIONWITH DIRECT SEGMENT MODELS

Segment based direct models have recently been used to im-prove the output of existing state-of-the-art speech recogniz-ers. To date, however, they have relied on an existing HMM system to provide segment boundaries. This paper takes ini-tial steps at using these models on their own, first by develop-ing a segment-based maximum entropy phone classifier, and then by utilizing the features in a segmental conditional ran-dom field for recognition. To produce a feature representa-tion that is independent of segment length, we utilize a set of ngram features based on vector-quantized representations of the acoustic input. We find that the models are able to integrate information at different granularities and from dif-ferent streams. Contextual information from around the seg-ment boundaries is particularly important. We obtain compet-itive results for TIMIT phone classification, and present initial recognition results