hide
Free keywords:
Natural regularity, pitch, timbre, scat singing, scene analysis, Shepard tones
Abstract:
Speech phoneme perception can be biased to follow production expectations. We propose some vowel-phonemes (such as /i/ as in “deed” and /Ʌ/ as in “dud”) sound, respectively, higher and lower in pitch in part because they are typically vocalized at higher and lower f0s, and listener pitch perception is biased toward those production tendencies. Resonant sounds, like vowels, possess two principal spectral pitch qualities: musical note/chroma (reflecting f0), and tone height (reflecting spectral centroid). Participants could exhibit perceptual biases to experience vowel phonemes at differing tone heights related to their typically voiced f0s in speech and song, even when they are f0-matched. Experiment 1 measures the similarity of 12 f0-matched vowel sounds and reconfirms two principal spectral timbre dimensions, one which is related to pitch: tone height/brightness (spectral centroid), and the other, harmonicity/consonance (harmonic overtone alignment). Experiment 2 finds participants vocally mimic high- and low-frequency sinusoidal sounds using the phonemes at height extremes: 8 kHz elicits /i/ and 60 hz elicits /Ʌ/. Participants also mimic consonant and dissonant sounds using the phonemes at harmonicity extremes: 256 hz sinusoid elicits /u/ (“dude”) and an inharmonic cicada call elicits /æ/ (“dad”). Experiment 3 confirms that, in freeform speech and scat singing, /i/, /I/ (“did”), and /Ʌ/ vowel phonemes exhibit characteristic f0s that correlate with height ratings in Experiment 1. The overall findings confirm a natural regularity for vowel sounds to have systematically different f0s in speech and freeform song. Listeners incorporate this pattern as a perceptual bias when rating tone height of f0-matched vowel phonemes and in production when mimicking non-speech sounds, consistent with the scene-parsing principle of perceptual biases matching natural acoustic patterns.
