This project is an assignment for the course:

Music and Audio Processing Lab @ UPF, Music Technology Group

Feature Extraction and Embeddings:

Utilizing Essentia, extracts features and embeddings from audio files that are used to "navigate" through the collection of files with the following apps.

• Playlist Builder: Web app for generating playlists based on user-defined audio descriptor queries, allowing users to create playlists with to specific characteristics.
• Music Similarity Comparison: Web app to explore audio similarity by computing cosine similarity of embeddings of songs.

Setup:

Data and Weights

Audio files shall be placed in the audio folder.

The following weights need to be downloaded from essentia and placed in the weights folder. https://essentia.upf.edu/models.html

• danceability-discogs-effnet-1.pb
• discogs-effnet-bs64-1.pb
• emomusic-msd-musicnn-2.pb
• genre_discogs400-discogs-effnet-1.pb
• msd-musicnn-1.pb
• voice_instrumental-discogs-effnet-1.pb

Requirements

It encouraged the usage of a virtual environment 😇

pip install -r requirements.txt

Usage

To run the script to extract descriptors:

python extract_main.py

To run the app to create playlists based on sonic descriptors:

streamlit run playlist_app.py

To run the app to compare songs based on embedding similarity:

streamlit run embeddings_app.py