The SJTU X-LANCE Lab System for MSR Challenge 2025

2026-02-04

AI Summary

This work addresses the challenges of multi-instrument separation, denoising, and dereverberation in music source restoration (MSR) by proposing a cascaded BS-RoFormer architecture that sequentially processes these three tasks, optimized specifically for eight instrument classes. The approach innovatively integrates community pre-trained models with diverse fine-tuning strategies, incorporating techniques such as data cleaning and mixing, random musical segment augmentation, and audio length extension. Evaluated in the MSR Challenge 2025, the method achieves state-of-the-art performance, ranking first across six primary objective and subjective metrics, with an MMSNR of 4.4623 and a remarkably low FAD of 0.1988, significantly outperforming existing approaches.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multi-Robot Systems

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Web Mining and Content Analysis: Web data integration and cleaning
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

This report describes the system submitted to the music source restoration (MSR) Challenge 2025. Our approach is composed of sequential BS-RoFormers, each dealing with a single task including music source separation (MSS), denoise and dereverb. To support 8 instruments given in the task, we utilize pretrained checkpoints from MSS community and finetune the MSS model with several training schemes, including (1) mixing and cleaning of datasets; (2) random mixture of music pieces for data augmentation; (3) scale-up of audio length. Our system achieved the first rank in all three subjective and three objective evaluation metrics, including an MMSNR score of 4.4623 and an FAD score of 0.1988. We have open-sourced all the code and checkpoints at https://github.com/ModistAndrew/xlance-msr.

Problem

• music source restoration
• music source separation
• denoise
• dereverb
• multi-instrument separation

Innovation

• BS-RoFormer
• music source restoration
• source separation
• data augmentation
• multi-instrument processing