Causal-U-Net
Causal-U-Net copied to clipboard
YangangCao
Experimental results
Hi, thank you for your sharing. How about the results of your experiments, please? I tried it, but the results looked terrible. The loss (sisnr or mae) hardly decrease in the first epoch.
Hello, my result sounds also not well, I am contacting with author of paper. I will let you know if I have any new information.
Ok. By the way, Xinlei Ren also winned the L3DAS21 challenge, and publiced their interactive demo at https://replicate.com/l3das/l3das22_challenge. I have experienced this demo, and it sounds not well. You can try it, and maybe it is beneficial to you.
Hi, there is just a baseline model in https://replicate.com/l3das/l3das22_challenge, I can't find Xinlei Ren's model.
ok, thanks, I have received Xinlei Ren's email, I will look into it and do some modifications according email. I will share the results with you after some days
According my training and testing, I have given up MMUB, you can try EaBNet (https://github.com/Andong-Li-speech/EaBNet). it's better than MMUB, I am training it.
Hi. Thank you for telling me about the EaBNet. I tried it and the results are really good. By the way, have you tried personalized speech enhancement. How do you thank about this track?
Hi, I had never tried personalized speech enhancement. by the way, can you share your script of training and data synthesis, and training logs？ In my training, the EaBNet is not good. my email: [email protected]. thanks very much!
Hello would the performance be enhanced by adding the beamformer and post filter?
Hi. Thank you for telling me about the EaBNet. I tried it and the results are really good. By the way, have you tried personalized speech enhancement. How do you thank about this track?
Hi, I have tried too, EaBNet is not good, would mind share your script of training and data synthesis, and training logs? my email: [email protected]
Hi. Thank you for telling me about the EaBNet. I tried it and the results are really good. By the way, have you tried personalized speech enhancement. How do you thank about this track?
Hi, I have tried EaBNet, but unfortunately, I couldn't get good results. Would mind sharing your script of training and data synthesis, and training logs? My email: [email protected]
