This model cannot be used with HiFiGAN. It was created using RefineGAN and can only be used with certain forks that support RefineGAN.

This model was created as a test version for reference purposes by model developers. Please note that RefineGAN has not been fully updated yet, so keep this in mind.

DataSet -
SID 0 : Dialogue style / Vocal 10 mins
SID 1 : Recitative style 10 mins
SID 2 : Furious tone 10 mins
SID 3 : Sorrowful tone 10 mins

The feature index generates different accents based on the speaker channel. By utilizing the values of the feature index and the speaker channel, the tone of the model can be adjusted.

Train Info -
Batch size per GPU : 16
Epochs : 117
Steps : 13000+
Pretrained Model : KLM 5.0 x2 32k
Sample Rate : 32k
Emb. Model : Contentvec / RVMPE
FP 32

Model Link - https://huggingface.co/SeoulStreamingStation/RVC_Voice_Models/resolve/main/STELLA_KLM50_RFG.zip?download=true

Additional Details:

Tags: No tags available

Download Link: https://huggingface.co/SeoulStreamingStation/RVC_Voice_Models/resolve/main/STELLA_KLM50_RFG.zip?download=true