End-to-end amp modeling

The following audio samples are provided as supplementary material for our paper
End-to-End Amp Modeling: From Data to Controllable Guitar Amplifier Models

• LSTM is a control-dependent neural network operating directly on audio waveforms. The model adjusts its response to amp knob positions provided as additional inputs, and can generalize to the full range of continous control of unseen positions. We use a 32-cell LSTM network for demonstration purposes, as this configuration is well studied in the literature and known to be real-time feasible.
• SPICE is an offline white-box circuit simulation method, based on LTSpice transient analysis. Making this model required careful circuit analysis and component value measurements. Furthermore, the model requires offline rendering and is several orders of magnitude away from real-time operation.
• Reference is a Matchless DC-30 tube amplifier used as a target device for modeling. The amplifier has five controls, volume, bass, treble, tone cut, and master volume, all of which must be recreated in an accurate model.

Reference recordings were made directly from the amplifier's output using a reactive loadbox. For consistency, all samples have the same DC-30 cabinet impulse response applied to them. Test samples vary in musical content, and the amplifier controls were adjusted to fit each context.

	LSTM	SPICE	Reference
Sample 1
Sample 2
Sample 3
Sample 4
Sample 5

Find out more about Neural DSP and experience controllable neural amp models yourself at neuraldsp.com.