Reproduce “ITO-Master: Inference-Time Optimization for Audio Effects Modeling of Music Mastering Processors” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Junghyun Koo, Marco A. Martínez-Ramírez, Wei-Hsiang Liao, Giorgio Fabbro, Michele Mancusi, Yuki Mitsufuji
The ITO-Master framework introduces Inference-Time Optimization (ITO) applied to the reference embedding for mastering style transfer, enabling dynamic, fine-grained control over audio effects during inference. This approach consistently outperformed existing baselines in objective metrics and expert listening tests, allowing for flexible adjustment of mastering characteristics, including text-guided artistic steering.
Key insights
- Inference-Time Optimization (ITO) applied to the reference embedding (z_ref) effectively refines mastering style transfer and is significantly more efficient than optimizing the full differentiable effect chain parameters.
- Co-training the reference encoder (Φ) with the mastering style converter (Ψ) greatly enhances the accuracy and adaptability of the style transfer process.
- White-box differentiable models, while offering interpretability, also yield perceptually consistent audio outputs as indicated by superior Fréchet Audio Distance (FAD) scores.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.