Audio examples from WSJ-mix dataset

• Data
  
• MixIT (TF-GridNet)
  
• RemixIT w/ CS (TF-GridNet)
  
• Self-REmixing w/ CS (TF-GridNet)
  
• RemixIT w/o CS (TF-GridNet)
  
• Self-Remixing w/ CS -> w/o CS (TF-GridNet)
  

NOTE: Output order of the three separated signals are aligned with the ground-truth audios (two speech and a noise).

Input mixture and clean speeches

Mixture 1

Mixture 2

Mixture 1 Clean speech1

Mixture 1 Clean speech2

Mixture 2 Clean speech1

Mixture 2 Clean speech2

MixIT

Mixture 1 Estimated speech1

Mixture 1 Estimated speech2

Mixture 2 Estimated speech1

Mixture 2 Estimated speech2

RemixIT w/ CS

Mixture 1 Estimated speech1

Mixture 1 Estimated speech2

Mixture 2 Estimated speech1

Mixture 2 Estimated speech2

Self-Remixing w/ CS

Mixture 1 Estimated speech1

Mixture 1 Estimated speech2

Mixture 2 Estimated speech1

Mixture 2 Estimated speech2

RemixIT w/o CS

Mixture 1 Estimated speech1

Mixture 1 Estimated speech2

Mixture 2 Estimated speech1

Mixture 2 Estimated speech2

Self-Remixing w/o CS

Mixture 1 Estimated speech1

Mixture 1 Estimated speech2

Mixture 2 Estimated speech1

Mixture 2 Estimated speech2