Dominant mid-to-high-frequency harmonics predict neural encoding and perception of concurrent vowels with large f0 differences.

Concurrent vowel perception experiments have revealed the importance of fundamental frequency (f0) differences in speech stream segregation. Understanding neural processes that support speech streaming using f0 differences remains an active area of perceptual and neurocomputational modeling research. This study simultaneously measured subcortical neural encoding [frequency following responses (FFRs)] and cued vowel identification accuracy of 12 concurrent vowel mixtures with large f0 differences (>8 semitones) to assess whether f0-based neural channel selection predicted perception. Results indicated that the strength of f0-based neural channel selection for one vowel often depended on the identity of the competing vowel. FFR f0 amplitudes and cued identification accuracy were predicted by which vowel had more dominant spectral energy in the mid- to high-frequency range (>1500 Hz). Attending to the target vowel in each concurrent vowel mixture did not modulate neural representation of target or nontarget vowels. Together, these findings suggest that speech streaming using f0 differences may be supported by f0-based neural channel selection from dominant unresolved harmonics, at least for competing vowels with large f0 differences.