jabogithub / psc2

code used for collaborative musical improvisation with ML models

Psc2

psc is a señor swesearcher in google brain, and is experimenting with using machine learning models (including generative models for music) as part of the live performance. most of the code used will be available in this repo.

if you want to see the code i use for my talks, you can see it here.

NEW: i built a web app to make it easier for you to try out this idea! check it out here.

the white paper explaining the solo setting is available here.

if you want the full-blown system, continue reading.

Installation

1. Install SuperCollider

2. Clone this repo.

3. Create a virtualenv and activate it. We need Python2 (and not Python3) because pyosc is not compatible with Python3. This step is optional but recommended:

   virtualenv --system-site-packages -p python2 venv
   source venv/bin/activate
   cd Psc2
   pip install -r requirements.txt
   

   If the last command does not work, then:

   1. pip install absl-py

   2. pip install pyosc.

   3. pip install tensorflow Full instructions here.

   4. pip install magenta Full instructions here.

4. Download Melody RNN attention_rnn model here, and update the base_models_path parameter in Psc2/modes/mlsplainer.py.

5. Download DrumKit RNN model (for NeurIPS demo) here.

6. Open SuperCollider, open Psc2/server.sc and run the main group (enclosed in parentheses).

7. From the root directory, run:

   python setup.py install
   

8. Start the python server:

   python Psc2/server.py