Plan of action:
- Create a script to collect files based on keywords "instruments list"- "genre list" and copy into different folders in each combinations to train
- Update the fine tune model script to run to a set number of checkpoints then output a named model based on the genre/instrument combo
- Find a way to run this in bulk and run it!
Then:
- I have checkpoints for key word combos
- Create an api/web service that takes keywords and generates samples from one of the chose models...
Finally:
- Integrate with interface that let's you search, playback and download clips?