In the context of rapid advancements in Large Audio Language Models (LALMs), fine-grained audio reasoning—such as recognizing event order, repetitions, and duration—remains challenging.
Existing post-training methods heavily depend on costly external labels or provide only coarse semantic signals. To address this issue, we introduce Audio-Zero, the first label-free self-evolution framework in the field of LALMs, aimed at enhancing fine-grained auditory perception and reasoning.
Audio-Zero constructs an auditory self-play game from unlabeled audio contrast pairs: most players hear a reference audio, while one odd listener hears a subtle variant. The model first generates clues describing what it hears and then identifies the odd listener by reasoning over inconsistencies among the clues. Since the odd listener is known by construction, the game provides verifiable rewards without any annotated answers.
Experiments with Qwen2-Audio-7B-Instruct and Qwen2.5-Omni-7B on TREA, MMAU Test-mini, and MMAR show that Audio-Zero improves fine-grained audio reasoning while preserving broad audio understanding. Evolutionary and diagnostic analyses further reveal that increasingly fine-grained auditory descriptions emerge naturally from game pressure.
Blogger's Review: The innovation of Audio-Zero lies in its label-free self-evolution mechanism, breaking the traditional model's dependency on annotated data, and showcasing the vast potential for fine-grained audio reasoning applications. This approach not only enhances the model's reasoning capabilities but also brings new insights to the audio processing domain. Its cleverly designed self-play game warrants further exploration and application.