stefan-balke/hamr2015-lstm-music-gen

Generate Melodies + Chords by using LSTMs

HAMR 2015: DeepComposer

Authors

This is a little hack we did at the HAMR@ISMIR 2015 hackday in Málaga!

• Anna Aljanaki, Stefan Balke, Ryan Groves, Eugene Krofto, Eric Nichols

Hack Goals

• Collect several symbolic song datasets, with melody and possibly chords.
• Represent data in a common vector format appropriate for input to a neural net.
• Develop an LSTM architecture for generation ration melody/chord output.
• Make music!

Installation

Python Requirements

pip install -r requirements.txt

Datasets

The datasets we use are included. In the following we would like to give credits to the authors:

• Temperley's Rock Corpus
  • http://theory.esm.rochester.edu/rock_corpus/
  • 200 songs
• Essen folk song collection (ESAC): http://www.esac-data.org
• WeimarJazzDatabase: http://jazzomat.hfm-weimar.de

Documentation

Further documentation can be found on: http://labrosa.ee.columbia.edu/hamr_ismir2015/proceedings/doku.php?id=deepcomposer