PennyZhang9/MultiAct

A benchmark for hierarchical activity recognition and captioning from long-form audio with multi-level annotations.