Abstract
Recent studies [Reilly and Port, J. Acoust. Soc. Am. Suppl. 1 78, S21 (1985)] show that timing measurements can be used to discriminate among items in a small vocabulary. Current efforts question the extent to which timing measurements can be used to augment basic segmental knowledge in a continuous, real‐time speech recognition situation. These experiments examine a single template of segments (vowel‐strong fricative‐stop‐vowel) each of which is embedded in different words in 112 different sentences for four English speakers. Syllable and word boundaries occur between each of the four segments. Discriminant analysis is used to predict the stress patterns, possible syllabifications, and word boundaries from timing measurements in the templates of a fourth male speaker after training on three others. Results indicate that within the template a linear combination of simple timing measurements can be used as predictors of lexical stress (95% correct). Word and syllable boundaries are much less reliably predicted (60%‐75%) and apparently bear no simple linear relationship to timing measurements. [Research supported by NSF.]
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call