Musimorphose, a national project composed of researchers in music and in data mining and data visualization was focusing on music and the influence of digital age.
Archipelago of music digital library
While collaborating with Pascale Kuntz and Nicolas Greffard, a tool for analysing and visualizing music libraries was conceived. This software uses an archipelago of music islands. Music styles are islands and we link theirs heights with the corresponding number of tracks.
“Echonest” (part of Spotify now), a company specialized in sound analysis, provides an API that gives, for almost any tracks title, a vector of 12 sound descriptors. Styles distributions are available for artists as well. We used this data to place islands.
In our tool, we perform the location of islands either using a Self Organizing Map algorithm (SOM) or with Principal Composant Analysis (PCA).
Archipelago of music digital libraries comparison
Another tool produces an Archipelago of music digital libraries in comparison. It takes 2 digital music libraries. A facet approach is used and islands can be style, artist, album or tracks. Every island shows the cardinal of the difference between the two libraries for the corresponding concept. The island color depends on the sign of the result.
Exploration and filtering are made via mouse clicks on islands. If the display level is “music style” then a click on a music style island would produce a new archipelago of artists of this style. In the same way, a click on an artist island would produce the archipelago of the artist albums.
In addition, we enhanced the exploration of the comparison of two libraries A and B by the choice of the shown set, which can be A union B, A inter B, A minus B, B minus A.
Two users can therefore explore fastly, styles in common, tracks in common, tracks which are only in one library, etc.
We have published this work in :
- GT IHM 2013
- SFC 2014
- DI2014
- Séminaire Cartomuse 2014
- RCIS 2015
_
A few videos are on my vimeo page :