search

MTG / dunya

The Dunya music browser
http://dunya.compmusic.upf.edu
GNU Affero General Public License v3.0
29 stars 20 forks source link

Feature extraction for all the files in Carnatic collection #97

Closed sankalpg closed 9 years ago

sankalpg commented 10 years ago

Currently features are extracted for only the files which are a part of the dunya-compmusic collection. But for research purpose certainly we need other files a well.

alastair commented 10 years ago

I'm not sure of the best way to do this, since all of the feature calculation code assumes that we're also showing the data in dunya. The easiest way is probably to create a new collection in musicbrainz - "Carnatic not in dunya" and import it into the dashboard, but not into dunya.

Algorithms that calculate aggregate features over all of the audio will need to remember to check all the files in 2 collections. I should be able to make this an option.

alastair commented 9 years ago

Carnatic collection has imported: http://dunya.compmusic.upf.edu/dashboard/collection/f8bf7d1e-70d2-44f6-a3cb-5a6ded00be1f

This solves the problem of the audio and derived files being available in the docserver. We can't get this information from the carnatic API. If we want it to be available (e.g. have these recordings show up when we get all tracks played in raaga x) then we'll need to import it into the main dunya database except flag it somehow so it doesn't show up in the interface.

Currently we don't have this separation for any other styles. Is it needed for hindustani or makam?

alastair commented 9 years ago

We will store a bootleg flag on carnatic releases and recordings. Dunya won't show these items, but you can set a flag on the API to include or exclude them. If there is additional data - e.g., an artist or a raaga that only exists from our bootleg collection, we will show this in dunya anyway, even if there is no other related information to show.

For aggregate data (e.g. raaga pitch profile) we can choose if we want to use all data or just the non-bootleg. We should look at how good the pitch data is first.

alastair commented 9 years ago

Done! You can access bootlegs through the api if you're a staff member (look at the python client). On search, and other places in dunya you only see bootlegs if you're staff, otherwise it looks like they don't exist.