Background: Understanding speech in noisy environments relies on the ability to focus on a target talker while filtering out competing input, a process typically associated with cortical auditory stream segregation. However, the auditory system is not strictly feedforward: descending efferent pathways, including the medial olivocochlear (MOC) bundle, project from higher auditory centers to the outer hair cells in the cochlea and regulate cochlear gain. This raises the possibility that selective attention can shape auditory processing at the earliest stages of the peripheral system. Otoacoustic emissions (OAEs), which originate from the cochlea’s nonlinear mechanics, offer insight into these peripheral dynamics. While distortion product OAEs (DPOAEs) are usually elicited with pure tones, more complex, speech-like stimuli have rarely been used.

Methods: In this work, we employed speech-derived DPOAEs (speech-DPOAEs) to determine whether attention to speech modulates cochlear responses and whether such modulation depends on whether harmonic components are resolved or unresolved along the basilar membrane. The stimuli were generated from male and female speech by selecting harmonic overtones n and m (n < m) of the fundamental frequency to elicit cubic distortion products at 2n–m. Resolved harmonics are expected to produce distinct excitation peaks on the basilar membrane, whereas unresolved harmonics are assumed to overlap. The design allowed speech-DPOAEs from both voices to be measured simultaneously.

Results: Forty normal-hearing adults (18–31 years) were presented with two competing voices in one ear, while speech-DPOAEs were simultaneously evoked and recorded in the contralateral ear. Participants alternated their attention between the male voice, the female voice, or a visual distractor across repeated trials. We found that speech-DPOAEs corresponding to resolved harmonics were significantly reduced when the associated voice was attended relative to when it was ignored, indicating that attention reduces cochlear output for spectrally resolved speech elements. In contrast, no effect emerged for unresolved harmonics when both target and competing voices contained unresolved components in the same frequency range. Comparable modulation was also observed when auditory attention was contrasted with visual engagement, suggesting that intermodal attention can likewise influence cochlear activity.

Conclusion: The present results offer the first direct evidence that selective attention to speech in noise can act on the cochlear amplifier itself, consistent with efferent MOC control of outer hair cell function.