ABSTRACT The aim of the project is to collect descriptive quantitative data on the acoustic characteristics of American English vowels and consonants. The speech materials selected will be suitable for studying vowel reduction and vowel-dependent consonant coarticulation. The phonetic segments to be analyzed will come from three operationally defined styles of speech providing samples that range from strongly context-dependent (reduced, coarticulated) forms to neutral and less context-dependent ("clearly" spoken) tokens. The goal is to be able to specify how the acoustic parameters change across such forms, to evaluate the perceptual significance of the changes and to propose numerical models describing and predicting the observed patterns of variation. Key questions are: Do speakers change their speech when they try to "speak more clearly", that is when they adapt to listener needs? If so, what acoustic modifications do they invoke to disambiguate reduced, coarticulated and perceptually similar forms? What is the nature of the acoustic transforms that are communicatively successful? The project will take steps towards discovering the systematic structure of intra-speaker phonetic variation and should thus produce results relevant to attempts to resolve the invariance issue (Perkell & Klatt 1986), a classical topic for general theories of spoken language and one of the unresolved problems still facing the design of automatic systems for text-to-speech conversion and speech recognition/understanding.
