Auditory Speech Features: The Sounds of English as Processed by a Model of the Ear.
Abstract
The model of the peripheral auditory system developed at the Neurophysiology Branch in conjunction with a PDP-11/20 computer is being used to process and recognize speech by identifying the phonetic features of each sound. This report describes the first stage of research aimed at identification of these features in terms of the auditory model. The English vowels were found to be identifiable with the auditory model by features corresponding quite closely with the well-known spectral 'formants'. Final temporal detail was preserved as well, and this is shown to be a significant advantage for certain sounds characterized by rapid changes in the signal. The model also proved capable of registering such secondary features of vowels as duration, pitch and diphthongalization, which are essential to identification where different speakers and different word contexts are involved. For those consonants studied, the auditory model produced patterns for particular features as specific as those reported for other processing systems, again with the preservation of fine temporal structure. For detection of some consonant features, special purpose circuitry was designed and tested. The report also discusses strategies, based on the phonetic composition of some sounds, for writing integrated phonemic recognition algorithms, and presents the results of some preliminary tests of such algorithms.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1974
- Accession Number
- ADA002604
Entities
People
- John J. Godfrey
Organizations
- University of Dayton Research Institute