Vowel and formant representation in human auditory speech cortex

Abstract

Vowel sounds are a fundamental component of human speech across all languages. Vowels are cued acoustically by formants, the resonance frequencies determined by the shape of the vocal tract during speaking. An outstanding question in neurolinguistics is how the human brain processes vowel formants during speech perception. We used high-density intracranial recordings from the human speech cortex on the superior temporal gyrus (STG) while participants listened to natural continuous speech to address this question. We derived two-dimensional receptive fields based on the first and second formants to characterize tuning to vowel sounds. We found that neural activity at single STG sites was highly selective for particular zones in the formant space. Furthermore, this formant tuning shifted dynamically to adjust for speaker-specific spectral context. Despite this formant selectivity, local cortical responses were not sufficient to reliably discriminate between vowel categories. Instead, vowel category representations could be accurately decoded when using the entire population of formant encoding electrodes. Together, our results reveal that vowels are locally encoded in STG with complex acoustic tuning in two-dimensional formant space. As a population code this gives rise to phonological vowel perception.