Nanashi v1.7 (RMVPE) (RVC v2, 250 Epoch)

[rapid escarp]

Epoch: 250
Batch Size: 16
Dataset Length: 11 hours of Brazilian singing voices
F0 pitch extraction: RMVPE
Sample Rate: Only 32k
FP16
Fine-tune on Titan Medium
Link: https://huggingface.co/shiromiya/nanashi-pretrain/tree/main/v1.7

Model made with this pretrain:
https://discord.com/channels/1159260121998827560/1265495016428666963

Image credits: https://www.pixiv.net/en/users/5015345
Old versions changelog: https://github.com/ShiromiyaG/nanashi-pretrain/blob/main/versions.md

[clear canyon]

👏👏👏👏

[dim totem]

Pretrain brasileiro ebaaa

[elder scarab]

Ayo? @dim totem level 1 !!!

[dim totem]

Cade o barboza numa hora dessas

[vague gazelle]

[weary scarab]

Added this pretrain to the index in the pretrains guide !

I put long datasets needed for this pretrain as the model sample dataset was of 35 mins

[rapid escarp]

Okay, thanks @weary scarab ! I'll make a model with a small dataset as a test in the future, then I'll get back to you if the model looks good

[weary scarab]

Alright, let me know

[rapid escarp]

Nanashi v1.5 (RMVPE) (RVC v2, 50 Epoch)

[rapid escarp]

Nanashi v1.5 (Rigel Version) (RMVPE) (RVC v2, 50 Epoch)

[rapid escarp]

Nanashi v1.7 (RMVPE) (RVC v2, 250 Epoch)

[weary scarab]

btw updated it with the newer versions too

[rapid escarp]

Thank you! Btw, you put this version as v1.5 Rigel. I'm going to test it later with models with a smaller dataset, to see if it's any good. Also, this is the last version before 2