Exporting complete Roon library data for data science / machine learning experiments

Hello,

I’m exploring a personal data science and machine learning project and would like to use my Roon library as a real-world dataset.

I’m therefore looking for ways to extract as much information as possible from my Roon database, including:

Albums, artists, tracks and compositions
Genres, tags, playlists and collections
Favorites and ratings
Play counts
Listening history
Timestamps (first play, last play, playback chronology, etc.)
Any user-generated or Roon-generated metadata available

The objective is not to migrate away from Roon, but rather to experiment with various machine learning techniques on my own listening habits and music collection, for example:

Building recommendation models
Exploring clustering and similarity analysis
Predicting listening preferences
Studying listening patterns over time
Creating embeddings and graph-based representations of artists, albums, genres and listening behavior

I am aware that Roon backups allow full restoration into another Roon instance, but I’m specifically interested in exporting the data into formats that can be analyzed outside of Roon (CSV, JSON, SQL, API access, etc.).

Has anyone managed to extract most or all of these data?

If so:

Which approach did you use?
What data could be retrieved?
Was listening history included?
Are there any community tools, scripts or APIs you would recommend?

Thanks in advance for any guidance or shared experiences.

Best regards,

I didn’t think there was anything like that. I’d also be interested if there was or there was a workaround or API. There is a limited Export feature which I assume you have already tried but its not exporting history, playcounts, timestamps etc.

Look at using either Last.fm or ListenBrainz. I doubt the API will have all you need.

I copied my Roon data directory (stop Roon first) in order to safely dump the levelDB data. I used a couple of open source libraries and tools to examine the data but found it more difficult than useful to rebuild the relationships and query for the type of patterns you describe above. Please share if you get anywhere with your research.