Interactive Exploration of Musical Space with Parametric t-SNE

This paper presents a user interface for the exploration of music libraries based on parametric t-SNE. Each song in the music library is represented as a stack of 34-dimensional vectors containing features related to genres, emotions and other musical characteristics. Parametric t-SNE is used to construct a model that extracts a pair of coordinates from these features for each song, preserving similarity rela- tions between songs in the high dimensional-feature space and their projection in a two-dimensional space. The two-dimensional output of the model will be used for projecting and rendering a song catalogue onto a plane. We have investigated different models, which have been obtained by using different structures of hidden layers, pre-training technique, features selection, and data pre-processing. These results are an extension of a previous project published by Moodagent Company, which show that the clustering adaptation of genres and emotions, that is obtained by using parametric t-SNE, is by far more accurate than the previous methods based on PCA. Finally, our study proposes a visual representation of the resulting model. The model has been used to build a music-space of 20000 songs, visually rendered for browser interaction. This provides the user with a certain degree of freedom to explore the music-space by changing the highlighted features and it offers an immersive experience for music exploration and playlist generation.