Speech Separation
Speech separation refers to the task of extracting all overlapping speech sources from a mixed speech signal. As a specific scenario of source separation problems, speech separation primarily focuses on isolating multiple simultaneously occurring speech signals rather than other interfering signals such as music or noise. This technology holds significant application value in speech recognition in multi-speaker environments, hearing assistance devices, and audio editing.
WSJ0-2mix
SepReformer-L
WHAMR!
TF-Locoformer (M)
Libri2Mix
MossFormer2 (w speed perturb)
WSJ0-3mix
Gated DualPathRNN
LRS2
TDFNet-small
WHAM!
MossFormer2
WSJ0-5mix
Gated DualPathRNN
LRS3
IIANet
VoxCeleb2
RTFS-Net-4
WSJ0-4mix
Libri5Mix
Hungarian PIT
Libri10Mix
GRID corpus (mixed-speech)
Libri20Mix
LibriCSS
Conformer (large)
iKala
U-Net
Libri15Mix
Hungarian PIT
TCD-TIMIT corpus (mixed-speech)
WSJ0-2mix-16k
MossFormer2