Formant-harmonic interactions and vowel perception

Abstract

The technique of “turning over” in singing which causes perceivable vowel quality shifts has been attributed to the crossing over of H2 over F1. The present work seeks to validate these perceptual claims. A modified AXB discrimination task was completed by 63 speakers, consisting of Klatt synthesized vowels. Versions of the vowels (/i, u, ɑ, ɔ/) varying in the F1 (+/50 Hz) dimension, the F2 (+/−100 Hz) dimension or both F1 and F2 dimensions were used as reference “A/B” tokens, while the “X” token varied along 9 semitones in an F0 continuum. F1 of the vowel aligned to H2 at step 5 of the F0 continuum. Listeners judged whether the vowel quality of token ‘X’ (F0 modified) was closer to “A” or “B” (F1/F2 modified). Results find that listeners perceive the vowel to have a lower F1 during H2-F1 crossover, but the vowel height is perceived to be lower before and after the alignment. This pattern was more robust in the vowel height than the vowel backness dimension, with similar results across vowels. These findings were confirmed with logistic regression models. We conclude that formant harmonic crossovers have psychoacoustic effects on vowel perception, holding implications in vowel modification in singing.