Singing Voice Enhancement

[mickszh]

Hi, thank you for the brilliant work! I just have a question on singing voice enhancement. I tried the serivce out with a piece of clean, studio-recorded singing voice to see if the model will give me an almostly identical wave. Because in my case, I cannot determine if the input waves are already clean or not, and I need it for both singing voice and speech to get finally clean and intelligeable results. It seemed that the clean singing voice is partially eliminated by the model, which introduces unintelligeability. I guess that's because there is little singing data in the dataset. I'm wondering if there is a pretrained model that works well on both singing and speech inputs? Thank you so much!