Using spotify's web API seems to be a viable and easy way to get several audio features for music, however I cannot find any documentation about what low-level features make up the given features like "danceability" and so on. Does anyone have experience with that?
I've done a bit of work with it, and I know that the "danceability" metrics, etc. are proprietary. These are calculated by Echonest's algorithms, which is now owned by Spotify, but even when they weren't the specifics were closed off. Your best bet is maybe Tristan Jehan's dissertation, which talks about the concepts that would later go onto be echonest features.
Hi Paul, this article might be of interest for you: https://www.researchgate.net/publication/305278539_Close_reading_big_data_The_Echo_Nest_and_the_production_of_rotten_music_metadata
Best regards, Florian
Article Close reading big data: The Echo Nest and the production of ...
Hi Florian, thanks for this. The paper indeed was interesting, alhough the empirical part mainly describes how the echo nest retrieves meta data about artists from blog posts and does not go into detail about audio feature extraction.
Thanks Daniel for your comment. I'll take a look at the thesis. Which can be found here, if anyone else is interested http://web.media.mit.edu/~tristan/phd/dissertation/
I think, a reason why there's still little insight on those specific data and algorithms your searching for could be - not only because of legitimate copyright issues, but also because they're kind of delicate business information worth a lot of money. But maybe I am mistaken.
But may I recommend you to get in contact with Meinard Müller (read his article: Müller, Meinard: Automatische Erschließung von Musikdaten.
In: Auhagen W., Bullerjahn C., v. Georgi, R. (Hrsg.): Musikpsychologie - Anwendungsorientierte Forschung. Band 25. Hogrefe, Göttingen 2015, Seite 243-245.) or also maybe with Klaus Frieler, Christoph Reuter or Felix Christian Thiesen . Their experts when it comes to music information retrieval. When there's some documentation available on Spotifiy's algorithms or calculation methods, they shurely will know.
Best wishes for continued success!
Florian
Anyone interested in using spotify audio features has now the opportunity to use the spotifyr package for R written by Charlie Thompson. This is very useful and and easy way to get structured data outputs for musicians, bands, or spotify users: https://github.com/charlie86/spotifyr
Hey Guys, if you are intersted in what you can do with Spotify's audio features, I conducted a sentiment analysis of European chart music using the valence Score from Spotify in this blog post. https://paulelvers.com/post/emotionsineuropeanmusic/
