Segmentation ART: A Neural Network for Word Recognition from Continuous Speech
OA Version
Citation
Abstract
The Segmentation ATIT (Adaptive Resonance Theory) network for word recognition from a continuous speech stream is introduced. An input sequeuce represents phonemes detected at a preproccesing stage. Segmentation ATIT is trained rapidly, and uses a fast-learning fuzzy ART modules, top-down expectation, and a spatial representation of temporal order. The network performs on-line identification of word boundaries, correcting an initial hypothesis if subsequent phonemes are incompatible with a previous partition. Simulations show that the system's segmentation perfonnance is comparable to that of TRACE, and the ability to segment a number of difficult phrases is also demonstrated.
Description
License
Copyright 1998 Boston University. Permission to copy without fee all or part of this material is granted provided that: 1. The copies are not made or distributed for direct commercial advantage; 2. the report title, author, document number, and release date appear, and notice is given that copying is by permission of BOSTON UNIVERSITY TRUSTEES. To copy otherwise, or to republish, requires a fee and / or special permission.