Parallel architecture for labeling, segmentation, and lexical processing in speech understanding
Speech understanding is a complex task which requires extensive computation. To increase the processing speed, a speech understanding system is decomposed into tasks which can be performed by a series of distributed processing subsystems. An architecture to perform labeling, segmentation, and lexical processing is described. Using a parametric characterization of the speech signal, this system divides an utterance into labeled homogeneous regions. The system then performs dictionary lookups based on all probable labelings and segmentations in order to generate a complete set of word hypotheses. Using realistic assumptions from existing speech understanding systems, a statistical model of speech input, and simulations of the speech processing algorithms, the attributes of the parallel system to perform labeling, segmentation, and lexical processing for real-time speech understanding are derived. 36 references.
- OSTI ID:
- 5258289
- Resource Relation:
- Conference: Sponsored by Ohio State University, Bellaire, MI, USA, 23 Aug 1983
- Country of Publication:
- United States
- Language:
- English
Similar Records
Overview of a distributed parallel architecture for speech understanding
Single-chip self-contained speech recognizer