Ov2Super

Ov2Super, an experimental refinement of RVC v2 pre-trained models. Unleash the full potential of your models through fine-tuning, pushing the boundaries of innovation and performance. This new pretrain allows users to train good RVC voice-models with way less amonut amount of data than before and with way less required epoch count, thus umproving the speed of training.

To use these pretrained models simply download them and put them into the pretrained_v2 folder of your rvc fork, then you need to type in the name of the pretrain in training section of uors user interface to use them, as simple as that. Minimum recommeneded length of the dataset is 1 minute, but you can get good results even with 10 seconds! The minimum epoch count for the models to sound good is also 2-3 times lower than needed before.