lora-svc

Why we included this project

Singing voice conversion is the trick here: you take a recording of someone singing and re-render it in a different voice without losing the melody or the words. lora-svc strings together Whisper for the linguistic content, a speaker encoder for timbre, and a neural source-filter BigVGAN vocoder for the final synthesis. For hobbyists and small teams, the practical part is that you can fine-tune a pretrained model on a small set of clips from the voice you want to clone, after the usual prep work of separating the accompaniment and slicing the audio into short segments. A GUI script covers the train-and-infer flow, and a pitch-shift argument lets you transpose the result. One honest caveat: LoRA-style adapter fine-tuning is referenced but not fully implemented here, so treat this as a working conversion toolkit rather than a finished adapter framework.