Exploring properties of songs and predicting track features like its genre from spatially scaled notes.
Music genres are often composed of particular pitch patterns that can be used for prediction. The Spotify API provides features for entire tracks, e.g. its loudness or acousticness scores, as well as the sequence of the individual pitches (notes). Totaling 3600 tracks across techno, rock, jazz and classicsal generes were analyzed and used for both classical Machine Learning and Deep Learning modeling methods. Validation accuracy of both approaches were similar suggesting that more sophisticated network architectures are needed to increase the model performance.
- keras deep learning framework
- Tensorflow deep learning framework
- tidymodels machine learning framework
- tidyverse data wrangling
- R targets pipeline system
- spotifyr REST API calls
- quarto notebook documentation
Keywords:
- Spatial data analysis
- deep learning
- REST APIs
Create a file .env in the main directory to define the environment variables SPOTIFY_CLIENT_ID and SPOTIFY_CLIENT_SECRET