让声音分离更懂人话，用强化学习对齐语义意图。
MARS-Sep: Multimodal-Aligned Reinforced Sound Separation
- 用强化学习将分离任务看作决策过程，学习可解释的掩码策略。
- 在多模态查询下显著提升音质与语义一致性，尤其在图像/文本提示时效果突出。
- 适合需要精准语义控制的声音分离场景，如影视剪辑、智能音箱降噪。
通用声音分离面临核心错位：优化低层信号指标的模型常产生语义污染结果，无法有效抑制听觉上显著但声学相似的干扰源。本文提出偏好对齐视角，类比大模型与人类意图对齐。为此，我们提出MARS-Sep，一个基于强化学习的框架，将分离重构为决策过程。不同于直接回归真值掩码，MARS-Sep学习由偏好奖励模型引导的分解式贝塔掩码策略，并通过稳定截断的信任区域代理进行优化。奖励信号来自逐步对齐的音频-文本-视觉编码器，直接激励输出与查询提示的语义一致性。在多个基准测试中，该方法在文本、音频、图像查询分离任务上均取得一致提升，显著改善信号指标与语义质量。代码已开源：https://github.com/mars-sep/MARS-Sep，音频样例见：https://mars-sep.github.io/。
原文摘要 · Abstract (English)
Universal sound separation faces a fundamental misalignment: models optimized for low-level signal metrics often produce semantically contaminated outputs, failing to suppress perceptually salient interference from acoustically similar sources. We introduce a preference alignment perspective, analogous to aligning LLMs with human intent. To address this, we introduce MARS-Sep, a reinforcement learning framework that reformulates separation as decision making. Instead of simply regressing ground-truth masks, MARS-Sep learns a factorized Beta mask policy that is steered by a preference reward model and optimized by a stable, clipped trust-region surrogate. The reward, derived from a progressively-aligned audio-text-vision encoder, directly incentivizes semantic consistency with query prompts. Extensive experiments on multiple benchmarks demonstrate consistent gains in Text-, Audio-, and Image-Queried separation, with notable improvements in signal metrics and semantic quality. Our code is available at https://github.com/mars-sep/MARS-Sep. Sound separation samples are available at https://mars-sep.github.io/.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。