NEURAL AMPLIFIER MODELLING WITH SEVERAL GAN VARIANTS
On EGDB dataset
We demonstrate one audio sample for each amplifier on different setting.
Input is a clean audio extract from EGDB test-set.
Target is the corresponding amplifier rendered audio extract from EGDB test-set.
Baseline is the audio generated from the model we trained on the baseline method.
MPD+MSD is the audio generated from our model with replaced discriminator: MPD+MSD.
Input
Ground Truth
Baseline
MPD+MSD
Mesa
Fender
Marshall
Application to Unseen Data
We demonstrate that our model possesses the capability to render data from other guitar dataset. We use the Mesa model to render audio in this Experiment.
Input is the audio extract from Guitarset in "mic" setting.
MPD+MSD is the audio generated from our model with replaced discriminator: MPD+MSD.