List your company
ResearchFormerly WizWand

Speaker Diarization on Libri2Mix

1.37DERUME ASR init. (0.33, 0.33, 0.34)
1.19482.37743.564.7426Aug 28, 2025
Updated 5mo ago

Evaluation results

MethodLinks
UME ASR init. (0.33, 0.33, 0.34)Training set=LibriMix (460 hours), Mixture type=mixclean: speech only, Mode=max, Weighting scheme=w/ RWSE, Training weights (λasr, λdiar, λsep)=(0.33, 0.33, 0.34), ASR initialization=true, Collar tolerance=None2025.081.37
UME ASR init. (0.33, 0.33, 0.34)Training set=LibriMix (460 hours), Mixture type=mixboth: speech + noise, Mode=max, Weighting scheme=w/ RWSE, Training weights (λasr, λdiar, λsep)=(0.33, 0.33, 0.34), ASR initialization=true, Collar tolerance=None2025.082.14
UME (0.1, 0.1, 0.8)Training set=LibriMix (460 hours), Mixture type=mixboth: speech + noise, Mode=max, Weighting scheme=w/ weighted sum, Training weights (λasr, λdiar, λsep)=(0.1, 0.1, 0.8), ASR initialization=false, Collar tolerance=None2025.082.19
UME (0.33, 0.33, 0.34)Training set=LibriMix (460 hours), Mixture type=mixboth: speech + noise, Mode=max, Weighting scheme=w/ weighted sum, Training weights (λasr, λdiar, λsep)=(0.33, 0.33, 0.34), ASR initialization=false, Collar tolerance=None2025.082.26
UME (0.1, 0.1, 0.8)Training set=LibriMix (460 hours), Mixture type=mixboth: speech + noise, Mode=max, Weighting scheme=w/o weighted sum, Training weights (λasr, λdiar, λsep)=(0.1, 0.1, 0.8), Collar tolerance=None2025.082.28
UME (0.33, 0.33, 0.34) + ASR init.Training set=LibriMix (460 hours), Mixture type=mixboth: speech + noise, Mode=max, Weighting scheme=w/ weighted sum, Training weights (λasr, λdiar, λsep)=(0.33, 0.33, 0.34), ASR initialization=true, Collar tolerance=None2025.082.45
UME (λdiar = 1.0)Training set=LibriMix (460 hours), Mixture type=mixboth: speech + noise, Mode=max, Weighting scheme=w/o weighted sum, Training weights (λasr, λdiar, λsep)=(0, 1.0, 0), Collar tolerance=None2025.082.91
WavLM Large [32]Training set=LibriMix (100 hours), Mixture type=mixclean: speech only, Mode=max, Collar tolerance=None2025.083.24
EEND [2] (reprod.)Training set=LibriMix (460 hours), Mixture type=mixboth: speech + noise, Mode=max, Collar tolerance=None2025.084.62
wav2vec 2.0 Large [30]Training set=LibriMix (100 hours), Mixture type=mixclean: speech only, Mode=max, Collar tolerance=None2025.085.62
HuBERT Large [31]Training set=LibriMix (100 hours), Mixture type=mixclean: speech only, Mode=max, Collar tolerance=None2025.085.75