Welcome to the audio examples page. Here you’ll find the results of our experiments, as well as audio examples of music source separation.

All the audios presented below are taken from the evaluation datasets: ChoralSingingDataset or Cantoría.

Source separation results - Joint learning

BC1song

For these results, we trained the different models on the BC1Song dataset, and evaluated on the ChoralSingingDataset dataset. These results are presented in the table 1a of the paper, and audio examples are shown below.

Audio Mix Example

Separated voices from Wiener filtering

Original
UMSS
U-Net
SFSF
SFTSFT
SFSFT
WUP
Soprano
Alto
Tenor
Bass
Re-synthesized mix


BCBSQ

For these results, we trained the different models on the BCBSQ dataset, and evaluated on the ChoralSingingDataset dataset. These results are presented in the table 1b of the paper, and audio examples are shown below.

Audio Mix Example

Separated voices from Wiener filtering

Original
UMSS
U-Net
SFSF
SFTSFT
SFSFT
WUP
Soprano
Alto
Tenor
Bass
Re-synthesized mix



Generalization capabilities on a new dataset: Cantoría

For these results, we trained the different models on the BCBSQ dataset, and evaluated them on the Cantoría dataset. These results are presented in the table 3 of the paper, and audio examples are shown below.

Audio Mix Example

Separated voices from Wiener filtering

Original
UMSS
U-Net
WUP
Soprano
Alto
Tenor
Bass
Re-synthesized mix