Defining essential characteristics of reliable spectral properties that elicit spectral contrast effects in vowel identification

Abstract

Auditory perception excels at extracting reliable spectral properties from the listening environment. Preceding acoustic contexts filtered to emphasize a narrow frequency region (Kiefte and Kluender, 2008 JASA), broad differences between two vowel spectra (Watkins, 1991 JASA), or resynthesized to shift wide ranges of frequencies (Ladefoged and Broadbent, 1957 JASA) all influence identification of a subsequent vowel sound. Spectral differences between filtered contexts and subsequent vowel targets were perceptually enhanced, resulting in contrast effects (e.g., emphasizing spectral properties of [I] in the context produced more [ɛ] responses). Historically, this phenomenon has been studied using filters whose gain was broadband and/or high-amplitude, providing very strong evidence for these reliable spectral properties. Essential characteristics of these filters that are necessary and/or sufficient to elicit spectral contrast effects are unknown. A series of experiments examined relative contributions of filter frequency, amplitude, and bandwidth to reliable spectral properties that elicit contrast effects in vowel identification. Preceding sentence contexts were processed by narrowband, broadband, or spectral envelope difference filters derived from endpoints of a vowel series differing in one frequency region (e.g., F1 in [I] and [ɛ]). Preliminary results suggest complex dependencies on filter amplitude and bandwidth for vowel identification; further results will be discussed.