Abstract:
To study the role of formant frequency movements in vowel perception, 300 utterances were selected from a database consisting of 1668 /hVd/ syllables spoken by 45 men, 48 women, and 46 children [Hillenbrand et al., J. Acoust. Soc. Am. 97, 3099--3111 (1995)]. Measurements from these signals were used to synthesize two versions of each utterance: (1) an ``original formant'' version that followed the measured contours of F1-F3, and (2) a ``flat formant'' version with F1-F3 fixed at steady-state values. Results indicated: (1) the original, naturally produced signals were identified with greater accuracy than either of the synthetic versions and (2) the original-formant synthetic signals were identified with much greater accuracy than the flat-formant signals. Discriminant analysis methods of Nearey and Assmann [J. Acoust. Soc. Am. 80, 1297--1308 (1986)] show that differences between listeners' responses in the flat formant and either the natural or the original formant stimuli are significantly correlated (p<0.01, by randomization tests) with predictions from a simple nucleus +offglide model of vowel trajectories. Logistic regression analysis of this data is reported in a companion paper, 3aSC13. [Work supported by NIH and SSHRC.]