search

rasbt / musicmood

A machine learning approach to classify songs by mood.
GNU General Public License v3.0
411 stars 107 forks source link

Happy/Sad Rorshack #3

Open crysclitheroe opened 7 years ago

crysclitheroe commented 7 years ago

I love the idea for this. Totally want to try use it to find only happy music, and check that my current playlists are happy. I think, though, that my idea of happy might be different to yours, I wonder how different the results might be with a larger group of human annotaters. I also felt slightly cheated when I saw this, I think a lot of the mood affecting qualities of music are in the actual music, as opposed to the lyrics. For some reason I just assumed you had trained the algorithm on audio files, though now I realise how difficult that will be. I listen to a lot of - to me, happy-sounding - foreign music these days, because I like the music of a human voice but I dont want to be distracted by overthinking the lyrics or tempted to sing along while Im working. Occasionally, Ill google the lyrics, and be horrified by how dark they seem, to me they are a completely different mood.

rasbt commented 7 years ago

Thanks for the feedback! Yeah, it would be great to have a larger number of human annotators. In the initial version of the app, "users" were able to click on "agree/disagree" after the prediction, and the results were then collected for retraining using a majority vote of these.

Regarding the audio files ... I was thinking about this initially, however, the challenge was legally getting these audio files without violating any copyrights/DRM.