Reproduce “Differentiable Attenuation Filters for Feedback Delay Networks” 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
Ilias Ibnyahya, Joshua D. Reiss
A differentiable parametric equalizer (PEQ) is introduced for attenuation filters within Feedback Delay Networks (FDNs), offering precise control over frequency-dependent reverberation decay. This design achieves state-of-the-art T60 matching accuracy while significantly reducing computational overhead by approximately 62% compared to previous methods, enabling efficient gradient-based optimization.
Key insights
- Accurate and differentiable conversion of analog prototype filters to digital Second Order Sections (SOS) is crucial for maintaining spectral fidelity across the audio range.
- Sharing frequency and Q-factor parameters across delay lines significantly reduces computational complexity and the number of trainable parameters without sacrificing the ability to precisely shape the frequency-dependent decay.
- Formulating the attenuation filter design as a supervised regression task against a target T60-derived magnitude response allows for efficient, gradient-based optimization of the parametric equalizer's characteristics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.