Save a pedal setting as a playable Neural Amp Modeler file

The trainer learns how the pedal transforms a test signal. Keep its controls fixed throughout: the snapshot does not learn every knob position.[1][5]

This documentation-based workflow follows NAM’s 0.13 trainer guide; we did not record or train a pedal. It assumes an existing recording setup: a music workstation, audio interface, reamp box, pedal, and a DI box for returning its output to a microphone input.[2][3]

Make the recording pair

Route the test track to a spare interface line output, then through the reamp box into the pedal. Feed the pedal’s output through the DI into the interface. Record that input on a separate mono track whose output goes to the main mix, never back to the reamp send. Start with the send level low and check for clipping.[3][4]

Install Miniconda using the NAM installation guide, then install the trainer with pip install neural-amp-modeler. Windows or Linux users wanting NVIDIA acceleration should first follow the guide’s link to the current PyTorch installation instructions. Launch the trainer with nam.[2]

Choose Download input file. Play input.wav through the pedal and record the complete return. Export that recording as WAVE, preserving the input’s 48 kHz, 24-bit format and duration.[1]

Train, then inspect the match

Select the input WAV, recorded output WAV and destination folder in the trainer. Choose Train. Inspect its alignment plot: the two impulse responses should overlap, with the black dashed line just before their onset. If they do not, revisit the recording rather than accepting that alignment.[1]

Close the alignment figure to continue. At completion, inspect the plot comparing the model’s prediction with the recording. Closing that plot saves the model.[1]

Put the saved sound back into a session

For playback, the official NAM Users page supplies Windows and macOS installers for Gateway, the project’s snapshot-model player. Its underlying plugin supports VST3 and AudioUnit hosts. Install the appropriate player, add it to a track in your workstation and load the saved .nam file into its model slot.[5][6]

Try a dry guitar passage through the model and compare it with a recording of that same passage through the unchanged pedal. Listen to quiet notes as well as harder attacks. This listening check is a proposed comparison, not a result from our testing.

Keep the paired WAVs, model and a note of the pedal and interface settings together. Naming the file for the captured setting makes the library useful when you return to a song months later.

Report an error or suggest a correction