Rune Factory 3 -- Tuna, Generate By so-vits-svc 4.0
-
raw_data from tieba
-
filter by wav.title attributes which contains 'myce'
-
trained by so-vits-svc
- about 58000 steps, V100 GPU * 1 for 10+ hours (cloud server)
- cuda 12 and torch 1.13.1
- pretrained model and env from hugging_face
- D_0.pth and G_0.pth
- 44k
- kmeans cluster used
- about 58000 steps, V100 GPU * 1 for 10+ hours (cloud server)
-
music vocal extraction, try the following:
- online : melody ML and vocal remover
- local: deploy uvr5-gui on linux, 3080Ti 3-4 minutes per song
- window size : 320
- VR-HP-3 model
-
inference command:
python inference_main.py -m "logs/44k/G_[step].pth" -c "configs/config.json" -n "[song].wav" -t 0 -s "[speaker]" -cm "logs/44k/kmeans_10000.pth" -cr 0.4
-
Mix vocal and intrumental/bass/drums ...: Audacity
-
Picture Generate NovelAI
-
Video maker : kdenlive
无知时诋毁图恒宇
懂事时理解图恒宇
成熟时成为图恒宇
Rune Factory3 for NS ! Looking forward to new game also.