all pre and post processing layers and keep UNet portion with the weights, convert the core portion into CoreML, and adapt the pre post process codes to it. #VocalExtractor#LetsUnmix#Spleeter#CoreML
I was asked many times from iOS developers how I could convert Spleeter into CoreML. The early version of Vocal Extractor actually did not use Spleeter but it had a model trained by myself and it was based on UNet architecture and all pre and post processes were done by Swift
codes on iOS which were also developed by myself. As Spleeter is coincidently and fortunately based on UNet, similar to my approach, it was possible to reuse the pre and post processing codes to adapt the Spleeter pre trained model. The way to convert it into CoreML is to strip