We're looking at usual patterns in a subspace of the EchoNest dataset made out of 43K songs.
We try to find near neighbors using chromas features.
The goal is to see whether:
- we can do similarity using existing features, without the audio
- we can do it large-scale
- we can get something meaningful
What a synthetized chroma sounds like: http://www.columbia.edu/~tb2332/BMHD/chroma_synthetized_example.wav
http://www.ee.columbia.edu/~ronw/
http://www.ee.columbia.edu/~thierry/
Python
Some Matlab lines
The Echo Nest API