Feedback delay networks (FDNs) are a compact way to model recursive audio systems, commonly used for artificial reverberation and signal decorrelation. Their design involves structured matrices, acoustic decay shaping, and numerical simulation, which have typically required separate tools. The new pyFDN toolbox aims to consolidate these tasks into a single Python workflow.
Version 0.5.0 of pyFDN includes numerical design and analysis routines alongside block-based time-domain processing. It also integrates with FLAMO, enabling frequency-domain differentiable processing—useful for optimization-based approaches. The library supports explicit network construction and JSON-based presets, allowing researchers to move from design to rendering and reuse without switching contexts.
The authors position pyFDN as a reference for both research and teaching, emphasizing reproducibility mechanisms built into the software. By lowering the barrier to working with FDNs, the toolbox could make advanced reverberation techniques more accessible to the audio processing community.