Neural Audio Spring Reverb is a differentiable model of analog springs designed for the Neutone plugin platform and presented at DAFx 2024. The project combines circuit-inspired DSP with deep learning to capture the chaotic behaviour of hardware tanks while remaining lightweight enough for real-time performance.
The toolkit includes:
- PyTorch training scripts and datasets built from impulse responses of multiple spring tanks.
- A Neutone-compatible VST/AU plugin with macro controls for modulation, color, and excitation modes.
- Evaluation notebooks comparing the neural model with physically modelled and convolutional baselines.
Explore the full toolkit, research notes, audio examples, and downloadable plugin builds on the project website.
AI-assisted spring reverb modeling project presented at DAFx 2024, including VST builds, training scripts, and evaluation notebooks.