Problem
Research questions and friction points this paper is trying to address.
Historical Music Restoration
Orchestral Music
Unsupervised Learning
End-to-End Restoration
End-to-End Historical Music Restoration in Latent Space
🤖 AI Summary
This study addresses the challenge of unsupervised restoration of early orchestral historical recordings, which is hindered by the absence of authentic paired data. To this end, we propose an end-to-end music restoration framework based on latent space flow matching. By simulating a high-fidelity degradation chain to synthesize training data, this work establishes the first supervised benchmark for orchestral Historical Music Restoration (HMR). Our primary contributions include the release of a 9.3-hour royalty-free test set alongside open-source code. Comprehensive evaluations demonstrate that the proposed method significantly outperforms existing baselines across both subjective and objective metrics. Ultimately, this research provides a high-quality benchmark and an efficient generative solution for historical audio restoration.
📝 Abstract
Historical music restoration (HMR) has almost exclusively focused on constrained problems such as Super-Resolution or the restoration of solo pieces, under-exploring the general task of restoring orchestral historical music, which has multiple instruments. This under-exploration is largely because the HMR domain, early-20th-century recordings, has no pre-degradation ground-truth pairs, making the restoration task unsupervised and more challenging. This paper presents a supervised end-to-end orchestral HMR benchmark by exploring both the synthetic degradation functions and the end-to-end generative deep-learning restoration methods. We simulate the historical recording degradation chain more faithfully than prior work, which makes orchestral restoration into a tractable supervised problem. A latent flow-matching model trained on the resulting synthetic pairs outperforms existing HMR baselines on intrusive, non-intrusive, and subjective evaluations. We also curate and release a 9.3-hour license-free, unpaired, historical classical-music test set, along with code and audio demos.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Historical Music Restoration
Latent Flow-Matching
Synthetic Degradation
End-to-End Generation
Orchestral Audio
🔎 Similar Papers
No similar papers found.
