Discrimination of Computer-Synthesized Speech

Abstract

The effects of learning on the discrimination of computer-synthesized speech was assessed by presenting 100 computer-produced monosyllabic words to 2 groups of 15 adult subjects. One group's errors were corrected while the other group's errors were uncorrected. A comparison of errors on the first 50 vs the second 50 presentations showed significant improvement for corrected group only. It was concluded that the discrimination of computer-synthesized monosyllables can be improved with correction of errors but is still about 14% poorer than the discrimination of human speech.