So-Vits-Svc

so-vits-svc is an open-source AI singing voice conversion project that takes a singing vocal recording and converts it to sound as though it were sung by a different, trained target voice, while preserving the original melody, timing, and pitch contour. It builds on SoftVC (a content-encoding approach for separating vocal content from timbre) combined with VITS, a text-to-speech and voice conversion architecture, to achieve conversion quality that improved substantially on earlier singing voice conversion approaches when it was released. Using it requires training or obtaining a model for the specific target voice you want to convert to — the project provides the training and inference code, but a usable voice model is trained on a dataset of recordings of that target voice. This makes it a more involved, technical process than a plug-and-play consumer app: users typically need some comfort with running Python-based ML projects and either training their own voice models or sourcing pre-trained community models. so-vits-svc is aimed at hobbyist producers, VTuber/content-creation communities, and ML-savvy users experimenting with AI singing voice conversion, most commonly for cover songs where a vocal is converted to sound like a different singer or character voice. It's an open-source research-adjacent project rather than a hosted, consumer-friendly product — running it requires local setup and, in most cases, an already-trained voice model.