Perceptual sensitivity to spectral properties of earlier sounds during speech categorization

Abstract

Recent sounds can change what speech sounds we hear later. This can occur when the average frequency composition of earlier sounds differs from that of later sounds, biasing how they are perceived. These "spectral contrast effects" are widely observed when sounds' frequency compositions differ substantially. We reveal the lower limit of these effects, as +3 dB amplification of key frequency regions in earlier sounds was enough to bias categorization of the following vowel or consonant sound. Speech categorization being biased by very small spectral differences across sounds suggests that spectral contrast effects occur frequently in everyday speech perception.

Natural speech statistics shift phoneme categorization

“…Analogous to the present study, sentences with skewed distributions of higherfrequency energy would be expected to bias categorization of phonemes distinguished by higher-frequency energy. Research has confirmed this expectation, as sentence MSDs in low-F 3 (1,700-2,700 Hz) or high-F 3 (2,700-3,700 Hz) frequency regions influence categorization of /d/-/g/ varying in F 3 transition onset frequency (Stilp & Assgari, 2018b). Statistics beyond MSDs also influence spectral context effects.…”

Musical instrument categorization is highly sensitive to spectral properties of earlier sounds

“…It bears mentioning that musicians were not explicitly recruited to participate in the present study. But, if these null results are representative, it would suggest that musical training does not modulate the proposed low-level mechanisms that are responsible for SCEs (adaptation or adaptation-like mechanisms in the early auditory system; Delgutte, 1996;Delgutte, Hammond, Kalluri, Litvak, & Cariani, 1996;Holt, Lotto, & Kluender, 2000;Holt & Lotto, 2002;Stilp & Assgari, 2018). Future research with a highly musically trained participant sample would provide a stronger test of this suggestion.…”

Spectral contrast effects are modulated by selective attention in “cocktail party” settings

“…Spectral contrast effects affect a wide range of spectrally cued phonemic contrasts, including vowels (F2 contrast between /ɑ/ vs. /a:/; Bosker, Reinisch, & Sjerps, 2017;Reinisch & Sjerps, 2013; F1 contrast between /ɪ/ vs. /ɛ/; Sjerps, McQueen, & Mitterer, 2013;Stilp & Assgari, 2018), consonants (/b/ vs. /g/; Lotto & Kluender, 1998; /s/ vs. /f/; Sjerps & Reinisch, 2015), lexical tones (Huang & Holt, 2009;Sjerps, Zhang, & Peng, 2018), and even whole words ("Laurel" vs. "Yanny"; Bosker, 2018). Empirical evidence suggests that SCEs are not specific to speech or language, as they are also induced by filtered noise (Watkins & Makin, 1994) and pure tones (Holt, 2005(Holt, , 2006.…”

“…Others have challenged this view, suggesting that SCEs involve general auditory processes that compute a representation of the average energy across frequencies, like a long-term average spectrum (LTAS). This average spectral representation of a context then serves as a referent for representation for subsequent sounds (Feng & Oxenham, 2018a;Holt & Lotto, 2002;Huang & Holt, 2009;Laing, Liu, Lotto, & Holt, 2012;Lotto & Holt, 2006;Stilp & Assgari, 2018;Watkins, 1991), independent of talker knowledge. That is, exposure to contexts with greater energy below 500 Hz results in contrastive enhancement of the frequencies above 500 Hz in following ambiguous target vowels, biasing perception of ambiguous /ɪɛ/ vowels towards /ɛ/.…”