Perceptual Continuous Speech Recognition.
Final rept. Oct 72-Jan 73,
PERCEPTION TECHNOLOGY CORP WINCHESTER MASS
Pagination or Media Count:
A speech recognition system based on a perceptual respresentation of speech is presented. The objective is to study and investigate the recognition of connected speech composed of a context-free limited vocabulary. The generalization will permit vocabularies of multi-syllable words to be recognized with a high degree of speaker independence. A new method of segmentation is presented based on the recognition of vowels and vowel-like phonetic segments. This is coupled with a speaker transformation that maps the vowels of each speaker into a standard space thus reducing inter-speaker variations. At the present the transformation and segmentation procedures are based on the vowels only. However, a method of extending these principles to the recognition of consonants is presented. Modified author abstract