Research on Narrowband Communications
Abstract
This document reports on work toward a very low rate vocoder. We model speech as a Markov Chain of spectral templates for the unsupervised learning approach to very low rate vocoding. This quarter investigated some variations in the spectral clustering algorithms. We also decided to use the speech of many speakers for the clustering and sequential modeling. We also began work on synthesizing the speech of many speakers from a diphone data base recorded from a single speaker. In phonetic recognition, we compared two methods for 'training' the diphone network, and concluded that the distance metric needs to be improved.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1981
- Accession Number
- ADA102417
Entities
People
- John Makhoul
- John Sorensen
- Micheal Krasner
- Richard Schwartz
- Salim Roukos
Organizations
- BBN Technologies