Audio Samples for:
Self-Supervised Learning-Based Source Separation for Meeting Data
AMI Corpus (Real Overlapped Speech)
The separation model was fine-tuned from WavLM+ through semi-supervised PIT&MixIT on AMI and R-LibriMix datasets.
Before Separation
After Separation
Before Separation
After Separation
LibriCSS dataset (Simulated Overlapped Speech)
The separation model was fine-tuned from WavLM+ through supervised PIT on R-LibriMix datasets.
Mixture
Speaker 1
Speaker 2
R-LibirMix dataset (Synthetic Overlapped Speech)
The separation model was fine-tuned from WavLM+ through supervised PIT on R-LibriMix datasets.
Mixture
Speaker 1
Speaker 2