The pipeline is managed with python. The webAPI can be accessed with a web view.
webAPI → youtube-dl
↓
webAPI ← keras ← vamp
Current known issues: The webserver does not serve python files.
The deep learning model uses convolution on top of recurrent cells. It achieves an accuracy (precision) of 85% on three genres.
The datset is included in the form of pickled feature vectors extracted via sonic-annotator. The original music files and categorization were collections of "best-of" music CDs.
youtube-dl, sonic-annotator with plug-ins
Detailed guide in tutorial.md
- Vamp plugin host sonic-annotator (GNU license)
- QM Vamp plugin
- BBC Vamp plugin
On Mac OS copy the plugins to /Library/Audio/Plug-Ins/Vamp/.
allow execution of unsigned library in security settings after a failed attempt
Simply run:
sudo python3 -m pip install -r requirements.txt