LOFI Dataset

The training dataset is synthesized using multiple sources.

• Chords and melodies are obtained from Hooktheory
• Lyrics are obtained by scraping Google or Musixmatch
• Audio features are obtained using the Spotify API

To build the training set and add lyrics and audio features:

1. Download the Hooktheory dataset from this repo and copy the event folder into this directory, renaming it hooktheory.
2. Register application at the Spotify Developer Dashboard.
3. Write client id into the file spotify_client_id and secret into spotify_client_secret.
4. Set add_lyrics (true/false), add_spotify (true/false) and lyrics_provider (google/musixmatch) inside prepocessor.py.
5. Run python prepocessor.py.
6. The dataset will be built into the folder processed.

You don't need lyrics if you are running Lofi2Lofi only. This will create a larger dataset, as tracks with no lyrics will get discarded if add_lyrics is true