Downloads · 30 days
0
full_totakeke
AlpacaSundae/full_totakeke
full_totakeke is a machine learning model from AlpacaSundae. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as openrail.
Trained using all 73 songs from acww. (overkill but seems to work better than when I hand selected 5 songs, maybe the next model will just be scales etc of the sound bank instead)
Access
Public
Updated Jul 4, 2023
Repo size
82.6 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.zip82.6 MB · 100%
At a glance
- License
- openrail
- Access
- Public
- Created
- Jul 4, 2023
- Updated
- Jul 4, 2023
- SHA
- 95d92ebb
Model card
From the Hugging Face model README
Trained using all 73 songs from acww. (overkill but seems to work better than when I hand selected 5 songs, maybe the next model will just be scales etc of the sound bank instead)
Muted the instrument tracks in sf2 and converted to wav in python, but I left the whistling in as I thought it would be ok but it get's weird with silences so maybe it will be remade without whistling one day.
I typically just use mangio-crepe set to 64 hop length and I set the pitch down an octave.
For some songs I'll generate two octaves to get clarity in higher and lower parts of the song. It seems that too high/low pitch in the often lets words slip through too much. Usually need to cut bits where the input was silence after generation as well due to weird artefacts mentioned.
idk what im doing
- License
- openrail
- Created
- Jul 4, 2023
- Updated
- Jul 4, 2023