Said artist already has models made on him but none of them are good. I have some stems and a lot of unreleased material that can be used for a dataset.
I can provide all songs and vocal stems i want included in a zip file. All this artists songs have harmonies / layered vocals in the chorus, and theyre too strong to seperate, so for the model makers sake, just get rid of any layered vocal parts in the song and just focus on seperating the single vocal verses and such as the layered vocals will never be seperated enough for a good quality dataset.
I also need the model trained, hopefully someone who has experience a lot in training and understands the best settings to use and also maybe someone who trains multiple models and compares them to see which one is best? (I saw someone else in a shop post who did this practice that is why)
artist is english btw
I can pay in many payment methods. Thank you for taking the time to read this.