Computer Identification of Phonemes in Continuous Speech.
Abstract
An approach to computer recognition of continuous speech through phoneme identification is presented. Speech data is digitally processed through correlation, recognition, and location programs. Methods of phoneme prototype production were explored including single and multiple speaker discrete and averaged prototypes. The identification process presents a rank ordering of probable phonemic occurrances at each time period. The method is used to attain an average recognition rate of 72% on continuous speech spoken by dissimilar speakers. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1977
- Accession Number
- ADA053268
Entities
People
- Michael F. Guyote
- Patrick L. Sisson
Organizations
- Air Force Institute of Technology