Full song used in the model demo
Hello! Congratulations on creating the model—it helps me a lot with separating female and male vocals from songs. It comes very close to what MedleyVox does, although its main downside is really the quality; even when using AI-based repair/restoration, you can still notice the loss.
Have you ever considered making instrument-specific separation models in the future? I’ve been having trouble finding this kind of thing that runs locally, although I also believe training such models must be a lot of work. I thought about this possibility myself, but I’m not sure it would be a good idea with an RTX 3060.
Anyway, I wanted to ask if it would be possible for you to provide the link to the song you used in the demo. I really liked it and would love to listen to it in full. This kind of music with perfect vocal doubles is exactly what I enjoy the most.
Thanks!
Hello, I am not planning to train an instrument separation model mainly because I only have vocal datasets and not instrument datasets. Using RTX3060 for inference of such models is sufficient, but it is far from enough for training purposes.
The example song uses a cover of a song, the link is https://www.youtube.com/watch?v=seOX2QIMVH0
Thank you very much!