MUGEN benchmark measures multi-audio understanding in LALMs

MUGEN is a comprehensive benchmark for evaluating multi-audio understanding in large audio-language models across speech, general audio, and music. Experiments reveal current LALMs still struggle with multi-audio tasks.