Trial-to-trial variability in talkers’ fundamental frequencies restrains spectral context effects in vowel categorization

Abstract

Perception of a given sound is influenced by spectral properties of surrounding sounds. For example, listeners perceive /ɪ/ (low F1) more often when following sentences filtered to emphasize high-F1 frequencies, and perceive /ɛ/ (high F1) more often following sentences filtered to emphasize low-F1 frequencies. These biases in vowel categorization are known as spectral contrast effects (SCEs). When preceding sentences were spoken by acoustically similar talkers (low variability in mean f0), SCEs biased vowel categorization, but sentences spoken by acoustically different talkers (high variability in mean f0) biased vowel categorization significantly less (Assgari et al., 2016 ASA). However, it was unclear whether these effects varied due to local (trial-to-trial) or global (across entire block) variability in mean f0. Here, the same sentences were arranged to increase/decrease monotonically in mean f0 across trials (low local variability) or vary substantially from trial-to-trial (high local variability) with equal global variability. On each trial, listeners heard a sentence filtered to add a low-F1 or high-F1 spectral peak to bias categorization of a subsequent vowel (/ɪ/-/ɛ/ continuum). Sentences with low local variability in mean f0 biased vowel categorization significantly more than sentences with high local variability. Relevance to studies of talker normalization will be discussed.