Saved in:
Bibliographic Details
Main Authors: Zhang, Zihan, Cheng, Xize, Jiang, Zhennan, Fu, Dongjie, Chen, Jingyuan, Zhao, Zhou, Jin, Tao
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.10509
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915800727683072
author Zhang, Zihan
Cheng, Xize
Jiang, Zhennan
Fu, Dongjie
Chen, Jingyuan
Zhao, Zhou
Jin, Tao
author_facet Zhang, Zihan
Cheng, Xize
Jiang, Zhennan
Fu, Dongjie
Chen, Jingyuan
Zhao, Zhou
Jin, Tao
contents 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/.
format Preprint
id arxiv_https___arxiv_org_abs_2510_10509
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MARS-Sep: Multimodal-Aligned Reinforced Sound Separation
Zhang, Zihan
Cheng, Xize
Jiang, Zhennan
Fu, Dongjie
Chen, Jingyuan
Zhao, Zhou
Jin, Tao
Sound
Artificial Intelligence
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/.
title MARS-Sep: Multimodal-Aligned Reinforced Sound Separation
topic Sound
Artificial Intelligence
url https://arxiv.org/abs/2510.10509