FSA-GRPO: Teaching Auditory LLMs Few-shot Demonstrations

Arxiv CS.SD (Sound/Audio ML)

FSA-GRPO adapts auditory LLMs to low-resource tasks like children's speech recognition. The method explicitly trains models to perform inference with few-shot demonstrations.

few-shot

Arxiv CS.SD (Sound/Audio ML)·Haolong ZhengSiyin WangXulin FanZengrui JinMark Hasegawa-Johnson

8 days ago