Abstract
In general music composed by recurrent neural networks (RNNs) suffers
from a lack of global structure. Though networks can learn note-by-note
transition probabilities and even reproduce phrases, attempts at
learning an entire musical form and using that knowledge to guide
composition have been unsuccessful. The reason for this failure seems
to be that RNNs cannot keep track of temporally distant events that
indicate global music structure. Long Short-Term Memory (LSTM) has
succeeded in similar domains where other RNNs have failed, such as
timing & counting and CSL learning. In the current study I show
that LSTM is also a good mechanism for learning to compose music.
I compare this approach to previous attempts, with particular focus
on issues of data representation. I present experimental results
showing that LSTM successfully learns a form of blues music and is
able to compose novel (and I believe pleasing) melodies in that style.
Remarkably, once the network has found the relevant structure it
does not drift from it: LSTM is able to play the blues with good
timing and proper structure as long as one is willing to listen.
Note: This is a more complete version of the 2002 ICANN submission
Learning the Long-Term Structure of the Blues.
Users
Please
log in to take part in the discussion (add own reviews or comments).