RemFX: Evaluation Datasets

Evaluation datasets for paper General Purpose Audio Effect Removal. 

These datasets are initially sourced from VocalSet, GuitarSet, DSD100, and IDMT-SMT-Drums datasets before being processed in our dataset generation script.

The datasets are named according to the number of effects applied (0-5). For example, 2-2.zip contains 2 effects applied to each input audio example. The targets are left untouched. The audio effects applied are from the set (Distortion, Delay, Dynamic Range Compressor, Phasor, Reverb) and randomly sampled without replacement for each example. 

Each example is stored in a folder (0-999) and contains four files

• input.wav (clean)
• target.wav (effected)
• wet_effects.pt (serialized PyTorch file containing a list of the present effects (order))
• dry_effects.pt (unused for this dataset)

The wet effects list is in the order of Reverb, Chorus, Delay, Distortion, Compressor

More information can be found on the github repo and our audio examples.