DYNAMO An Algorithm for Dynamic Acoustic Modeling.pdfVIP

DYNAMO: An Algorithm for Dynamic Acoustic Modeling Franc¸oise Beaufays, Mitch Weintraub, Yochai Konig Speech Technology and Research Laboratory SRI International Menlo Park, CA 94025 ABSTRACT in nature, as opposed to binary. For these reasons, we chose instead to base our models on neural networks. This paper summarizes part of SRI’s effort to improve acoustic mod- eling in the context of the Large Vocabulary Continuous Speech + More recently, Ostendorf et al. [OBB 97] showed that a combina- Recognition (LVCSR) project. It concentrates on two problems tion of acoustic and prosodic features could greatly help identifying that are believed to contribute to the large error rates observed with speechsegments that were erroneously recognized (32% predictabil- LVCSR databases: (1) the lack of discriminative power of the speech ity improvement for a 10-hour training subset of Switchboard). Sim- models in the acoustic space, and (2) the discrepancy between the ilar results were reported by various researchers working o n confi- criterion used to train the models (typically frame-level maximum