Poster E62, Thursday, August 22, 2019, 3:45 – 5:30 pm, Restaurant Hall
Tracking the building blocks of domain-general processes supporting vowel perception
Ellie Abrams1,2, Laura Gwilliams1,2, Alec Marantz1,2;1New York University, 2NYUAD Institute
INTRODUCTION Correctly identifying meaningful categories from auditory input involves processing relevant components of the acoustic signal. For vowel perception, at least two features drive the eventual percept: i) the relevant frequency components in the signal (e.g. F1, F2, F3) and ii) the ratios between those components. In order to fully orthogonalise these two features, and therefore independently track their contribution to neural responses, and ultimately the perceptual output, we used sinusoidal tones as a test case. METHOD Two types of tones were created – pure and complex tones – with a perceived pitch ranging from 220-624Hz (mapping onto musical notes within three diatonic scales). The acoustic signal either consisted of a sinusoid at a single frequency (F0), mapping directly to the perceived pitch (pure tones); or they consisted of five harmonic frequencies: integer multiples of, but not including, F0 (complex tones). Critically, the ratio between the harmonic sinusoids generates the same pitch percept as the pure tones, even though the fundamental frequency is absent. Fourteen participants listened to sequences of these tones while magnetoencephalography (MEG) was recorded. RESULTS Multivariate analyses were used to decode pitch and tone-type from activity across MEG sensors. The goal was to assess when each feature becomes present in the neural signal. At 100ms, we could decode pitch from the pure F0 tones and the complex tones; responses to the different tone-types was indistinguishable. This suggests that the ratio between frequencies of the complex tones has been utilised at this latency to map onto a shared pitch category. At 200-300ms, we could again decode the pitch of the tone, but the spatial pattern differed depending on tone-type. This later response likely reflects processing spectral content that further characterises the acoustic percept. DISCUSSION In line with previous work on the temporal unfolding of vowel categorisation, our results confirm that an early neural response is involved in encoding the ratio between frequency components. Spectral information, namely the precise collection of frequencies present in the signal, may later inform more complex categorisation, such as source identification. Given the importance of formant ratios for accurate vowel perception, and the modulation of F0 for speaker identification, we can relate this domain-general process to the processing of vowels, which may unfold along a similar time-course.
Themes: Perception: Auditory, Perception: Speech Perception and Audiovisual Integration
Method: Electrophysiology (MEG/EEG/ECOG)