-
Bespoke Score-Informed Nets 2020Purpose-built neural networks that separate sources in a mix by synthesizing their own training data. [demo]
-
audioLIME 2020Listenable explanations of black-box audio models using sources as interpretable components. (With Verena Haunschmid) [code]
-
Cerberus 2020Neural network to isolate instruments in a musical mix and simultaneously transcribe the notes of each instrument. [demo]
-
Slakh2100 2019Dataset for music source separation with 145 hours of mixtures, isolated sources, and MIDI. [website]
-
WHAM! 2019Dataset of nonstationary background noise for speech separation. [website]
-
SDR Prediction 2017Predicting source separation evaluation metrics without ground truth. [demo]
-
nussl 2015 - Now
-
SampSyn 2012