Voice perception reflects graded sensitivity to acoustic cues along the auditory hierarchy.

Humans readily recognize voices in noisy, challenging environments, yet how the brain does this remains poorly understood. A key question is whether the brain is tuned to detect "voice-like" qualities present in sounds. To test this, we asked whether voice perception relies on sensitivity to acoustic patterns shared by human vocalizations. We developed a computational framework that identified the acoustic features most characteristic of human voices and examined how these features predict listener judgments of "voice-likeness" in both natural sounds and synthetic stimuli we designed to vary systematically in acoustic structure. We then recorded neural activity from implanted electrodes in the auditory brain regions of patients undergoing evaluation for epilepsy, asking whether these same acoustic patterns predict neural responses. Across the auditory cortex, neural activity aligned with this continuous scale of voice-likeness, and the strength of this alignment increased along the hierarchy from early to higher auditory areas but not in frontal auditory areas. These findings reveal that auditory association cortex implements a coding framework characterized by graded sensitivity to vocal acoustic patterns. This organization emerges progressively across auditory processing stages, suggesting that voice recognition reflects a fundamental principle: the brain uses continuous, graded representations to interpret the acoustic world and guide perception.