Fig. 1
From: Blind extraction of guitar effects through blind system inversion and neural guitar effect modeling
Depiction of the proposed idea: a previously, and separately, trained Hybrid Transformer Demucs (HT Demucs) network estimates for a set of processed reference input guitar signals \(x^i_{eff}(n)\), each processed by the same effect \(H_{eff}\), the clean, unprocessed reference signal \(\hat{x}^i_{clean}\). These estimates are then used, together with the given processed signals \(x^i_{eff}(n)\), to train a gated convolutional network with temporal feature-wise linear modulation (GCNTF). The result is an estimate \(\hat{H}_{eff}\) of the actual guitar effect \(H_{eff}\) used to process the \(x^i_{clean}\), which then can be applied to any guitar signal