Parametric Neural Amp Modeling with Active Learning
This work addresses the challenge of data efficiency in creating virtual guitar amps for audio engineers and musicians, though it is incremental as it builds on existing parametric modeling and active learning techniques.
The paper tackles the problem of efficiently training parametric guitar amp models by introducing Panama, an active learning framework that uses an ensemble-based strategy to minimize required data points. With only 75 datapoints, the models achieve perceptual quality matching that of the leading non-parametric amp modeler, NAM, as shown in MUSHRA listening tests.
We introduce Panama, an active learning framework to train parametric guitar amp models end-to-end using a combination of an LSTM model and a WaveNet-like architecture. With \model, one can create a virtual amp by recording samples that are determined through an ensemble-based active learning strategy to minimize the amount of datapoints needed (i.e., amp knob settings). Our strategy uses gradient-based optimization to maximize the disagreement among ensemble models, in order to identify the most informative datapoints. MUSHRA listening tests reveal that, with 75 datapoints, our models are able to match the perceptual quality of NAM, the leading open-source non-parametric amp modeler.