Reproduce “Style Transfer of Audio Effects with Differentiable Signal Processing” 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
Christian J. Steinmetz, Nicholas J. Bryan, Joshua D. Reiss
DeepAFx-ST enables the transfer of production styles between audio recordings by learning to control traditional differentiable audio effects. The system employs a self-supervised training strategy to overcome data scarcity and demonstrates interpretable parameter prediction, along with robust performance across varying audio sample rates.
Key insights
- The self-supervised training strategy successfully generates paired data from any audio collection, overcoming data scarcity challenges for deep learning in audio production.
- Traditional DSP-based differentiation methods (Automatic Differentiation, SPSA) maintain performance across varying sample rates, outperforming neural network-based approaches.
- The system learns to predict interpretable control parameters for standard audio effects, providing transparent and controllable style transfer.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.