Authors: Yu Wang, Daniel Stoller, Rachel M. Bittner, Juan Pablo Bello.
Here, we show 30 examples randomly sampled from the MUSDB18 test set. Each example contains 3 seconds of the original mixture, the output separated by the best performing 5-shot conditioned model, and the output separated by the instrument class-conditioned baseline. We show 10 examples for each of the three target instruments (vocals, drums, and bass).