[benjismith]

Very interesting. It's pretty good with song structure, formatting, themes, etc... The one major thing it seems to be missing is rhythm and rhyme.

I can imagine decorating the data with a pronunciation dictionary, so that you can coerce some better phonology from the model... There's a pretty good dataset from Carnegie Mellon here:

http://www.speech.cs.cmu.edu/cgi-bin/cmudict

[mathigatti]

That's a great idea, thanks for the feedback, I will try to do that :)

[gremlinsinc]

I'd like to be able to remove/save/hide some songs, so I can play around w/ new keywords and get rid of really bad ones... So, maybe an X that deletes the song on that line...

[mathigatti]

That would be pretty useful and easy to implement, thanks!