Listener training for speech-output applications

Abstract

The specificity of the adaptation to synthetic speech known to occur with practice was examined by giving listeners selective exposure to a subset of English phonemes (a control group was “trained” on analogous materials produced by a human speaker), and then testing their ability to identify words created from both the previously heard and novel phonemes. The results indicated that while synthetic voice training was generally facilitative, it was most helpful in the identification of the sounds heard before. However, this specific learning effect occured for only certain phonemes. The findings imply that one way to maximize early adaptation to synthetic speech is to identify the “learnable” sounds, and to increase users' exposure to them during introductory or training dialogs.