Guitar amps, but make them neural networks

What it does

This is the training half of the Neural Amp Modeler ecosystem: you feed it audio, it learns to mimic a guitar amplifier’s behavior, and it spits out a .nam file you can load into the companion plugin. The actual real-time inference lives in a separate repo—this one stays firmly in the “offline, GPU-hungry” lane.

The interesting bit

The project splits the problem cleanly: Python here for the messy work of training and exporting, C++ plugin there for the stage-ready performance. It’s a pragmatic admission that training and inference have different constraints, and trying to cram both into one codebase usually means both suffer.

Key highlights

• Trains neural network models that emulate analog guitar amplifier behavior
• Exports to a dedicated .nam format for use with the companion plugin
• Partner repo (NeuralAmpModelerPlugin) handles real-time standalone and plugin playback
• Documentation hosted at Read the Docs; buildable locally on Linux or Windows
• Active CI via GitHub Actions

Caveats

• README is sparse on training details—no mention of required data format, model architectures, or hardware expectations
• The split-repo structure means you’ll need both repositories for a complete workflow; this one alone won’t make sound

Verdict

Worth a look if you’re into DSP, guitar tone, or want a concrete example of separating training infrastructure from real-time inference. Skip it if you just want to plug in and play—you want the plugin repo instead.