Notches in sentence spectra bias subsequent phoneme categorization

Abstract

Spectral differences across earlier (context) and later (target) sounds are perceptually magnified, resulting in spectral contrast effects (SCEs) that bias categorization of later sounds. Most studies added spectral peaks to context sounds in order to produce SCEs, but noise contexts with spectral notches at speech frequencies biased phoneme categorization in a complementary manner (Coady et al., 2003 JASA). We tested whether this approach generalized to speech contexts with spectral notches. On each trial, a context sentence preceded the target phoneme (Experiment 1: /ɪ/-/ɛ/; Experiment 2: /d/-/g/). Sentences were processed by notch filters that attenuated energy in the same frequency regions that, when amplified to add spectral peaks, produced SCEs (Experiment 1: 100–400/550–850 Hz, as in Stilp et al., 2015 JASA; Experiment 2: 1700–2700/2700–3700 Hz, as in Stilp & Assgari, 2017 JASA). Notch depths ranged from −5 to −20 dB in 5-dB steps. In both experiments, notch-filtered sentences biased phoneme categorization in complementary directions to SCEs, with bias magnitudes increasing at larger notch depths. Whether earlier sounds have spectral peaks or notches, speech categorization is highly sensitive to the magnitudes of spectral differences across context and target sounds.Spectral differences across earlier (context) and later (target) sounds are perceptually magnified, resulting in spectral contrast effects (SCEs) that bias categorization of later sounds. Most studies added spectral peaks to context sounds in order to produce SCEs, but noise contexts with spectral notches at speech frequencies biased phoneme categorization in a complementary manner (Coady et al., 2003 JASA). We tested whether this approach generalized to speech contexts with spectral notches. On each trial, a context sentence preceded the target phoneme (Experiment 1: /ɪ/-/ɛ/; Experiment 2: /d/-/g/). Sentences were processed by notch filters that attenuated energy in the same frequency regions that, when amplified to add spectral peaks, produced SCEs (Experiment 1: 100–400/550–850 Hz, as in Stilp et al., 2015 JASA; Experiment 2: 1700–2700/2700–3700 Hz, as in Stilp & Assgari, 2017 JASA). Notch depths ranged from −5 to −20 dB in 5-dB steps. In both experiments, notch-filtered sentences biased phoneme catego...