Repository navigation
NAM Parametric Plugin 0.4.4
HyperA2 playback now uses the specialized A2 audio backend for both Full and Lite,
with NeuralAudio-style multi-frame weight reuse for small x64 buffers and the
faster Eigen kernel for larger buffers. Updating knobs and switches preserves
convolution history without recreating the audio network.
- Reduced processing cost: a repeated synthetic-model benchmark on i7-12700H
measured 16–32% less CPU time for Full and 53–65% for Lite. Actual gains depend
on CPU, buffer size and hypernetwork size. - Preserves calibration, metadata, Full/Lite selection, Eco de-click, crossfade
mode, resampling and static export. No retraining or file conversion is needed. - Supports existing
.pnamand.namfiles; other architectures retain the
generic backend. No added block latency or AVX2 requirement. - Shares the optimized core and validated HyperA2 loader with the
NeuralAudio PNAM hardware fork.
Validation includes Full/Lite PyTorch golden audio, generic/optimized audio under
control automation, all 30 combinations of three independent switches mixed with
knobs, calibration, de-click/crossfade and static export.
Packages: Windows x64 standalone/VST3, macOS universal (Intel + Apple Silicon)
standalone/VST3/AU, and Linux x86_64 standalone/VST3.