VOICE SOUND RECOGNITION.
Final rept. for Apr 64-Apr 65,
PHILCO-FORD CORP BLUE BELL PA COMMUNICATIONS AND ELECTRONICS DIV
Pagination or Media Count:
The report examines the merits of a new speech perception theory and its application to the voice sound recognition problem. Most conventional speech recognition systems require 7 important parameters to activate the recognition logic the frequency of the first three formants, the amplitude of the first three formants and a voice-unvoiced decision. The theory tested uses just three important parameters the frequency of a single equivalent formant SEF, the SEF amplitude, and a voicing decision. This decrease of more than two to one in input parameters should mean significantly more than a two to one reduction in the complexity of the recognition logic. Statistics were compiled on a set of 20 words uttered by an ensemble of 5 speakers 3 male and 2 female. Although some recognition confusions were encountered in some phonetically similar sounds they were not unexpected, since the statistics were compiled on segmented phonemes sans transient cues. However, other confusions were the result of imperfect parameter extractors, and hopefully will be corrected as improved circuits are developed. Recognition rates as high as 98 percent were measured in this initial phase of the program. Author