folkRNN-synthesis

A synthesis pipeline for machine folk tunes.

Use

• make the following directories:
  • ./stomps, containing individual stomp samples (16k mono)
  • ./ambiences, containing ambience sounds (16k mono)
  • ./impulses, containing impulse responses
  • ./render/models, containing the trained instruments

For each instrument, there should be a folder ./render/models/<instrument name>. That folder must contain two folders: control, with the trained control model, and synthesis, with the trained synthesis model.

To start the synthesis, call create_parts <source midi folder> <destination audio folder>. The midi files have to be monophonic. The script will create:

• <dest>/audio, containining individual wav stems for each instrument
• <dest>/midi, with the individual microtimed and ornamented midi stems
• <dest>/songs, with the rendered complete tunes