15th International Congress of Phonetic Sciences (ICPhS-15)
We investigated whether individual formant frequencies or distances between spectrally adjacent formant frequencies are more suitable for predicting perceived vowel articulation. The relationship between perceptual and acoustic representations of a set of vowel tokens was modeled. The acoustic representation consisted of measurements of F0 and the first three formants. The perceptual representation consisted of coordinates representing each vowel token's perceived height, advancement or rounding. The comparison between the acoustic and perceptual representations was carried out through linear regression analysis. We concluded that formant frequencies were more suitable for modeling perceived articulation when the speaker-related variation was eliminated.
Bibliographic reference. Adank, Patti / Smits, Roel / Hout, Roeland van (2003): "Modeling perceived vowel height, advancement, and rounding", In ICPhS-15, 647-650.