ApexSVC converts singing from one voice to another without training a model per voice. Zero-shot means you hand it a reference sample and a song, and it does the rest.
It runs on PyTorch with CUDA and came out of the same rabbit hole as our RVC cover tooling, except this one skips the whole train-a-model-first step.
Open source under AGPL-3.0. GPU strongly recommended.
more projects