Mamba2 Meets Silence: Robust Vocal Source Separation for Sparse Regions

TL;DR

This paper presents a robust vocal source separation model using Mamba2, a state space model, which outperforms existing methods in isolating vocals from music by capturing long-range dependencies efficiently.

Contribution

The paper introduces a novel Mamba2-based model with band-splitting and dual-path architecture for improved vocal separation, outperforming Transformer-based approaches.

Findings

Achieved a cSDR of 11.03 dB, the best reported to date.

Demonstrated stable performance across different input lengths.

Outperformed recent state-of-the-art models in vocal isolation.

Abstract

We introduce a new music source separation model tailored for accurate vocal isolation. Unlike Transformer-based approaches, which often fail to capture intermittently occurring vocals, our model leverages Mamba2, a recent state space model, to better capture long-range temporal dependencies. To handle long input sequences efficiently, we combine a band-splitting strategy with a dual-path architecture. Experiments show that our approach outperforms recent state-of-the-art models, achieving a cSDR of 11.03 dB-the best reported to date-and delivering substantial gains in uSDR. Moreover, the model exhibits stable and consistent performance across varying input lengths and vocal occurrence patterns. These results demonstrate the effectiveness of Mamba-based models for high-resolution audio processing and open up new directions for broader applications in audio research.