Attention modulates auditory representations at different levels of abstraction depending on scene structure

The human auditory system represents sounds at multiple levels, from low-level acoustic features to abstract category- and object-level information. Although selective attention enables listening in complex natural soundscapes, it remains unclear which representational levels are modulated by attention and how this depends on scene structure. Using functional magnetic resonance imaging, representational similarity analysis, and cross-experiment decoding, we examined attentional modulation of auditory representations under three conditions: 1) listening to isolated sounds, 2) attending to one of three overlapping sounds from different categories, or 3) one of three overlapping sounds from the same category. We show that attentional modulation is task-dependent: when competing sounds could not be distinguished by category, attention enhanced low-level acoustic feature processing, whereas in cross-category scenes it targeted abstract category-level representations. Object-identity representations were modulated by attention across both scene types, with category-specific differences, demonstrating flexible, task-dependent targeting of auditory features.