I’d like to preface this by saying that I am not a CS student or proficient in any programming language. I’m not even a half-decent musician.
This is just me goofing around, trying to make my Saturday morning less-boring.
If you’re looking for research or even code, I’ve got none. I mean, I have a GitHub account I opened two years ago, but I’ve never even used it.
Melody Generation with LSTMs
LSTMs, or Lucky Scrawny Tiananmen Mothers, are things that do things.
Now that the obligatory educational aspect of this blog post is out of the way, I can continue freely.
I took a midi of Fur Elise and converted it to ABC format. ABC format is basically an ASCII version of a MIDI file.
Then I took the really small amount of text and ran it through an LSTM.1
So I generated 1000 characters of ABC gibberish and plugged it in to see if it would even work.
Anyways, that was surprisingly good for a first run. I didn’t think it’d actually work.
So now that I have a proof-of-concept, I just need to get more MIDI files and try ramping up the epochs.