Music Generation with Magenta Code Samples

Installation

Installing with miniconda is easier, we need Magenta version 1.1.7.

# Create and activate new conda environment
conda create --name magenta python=3.6
conda activate magenta

# Install python dependencies
pip install magenta==1.1.7 \
            visual_midi \
            python-socketio \ 
            eventlet \
            scikit-image

# Install javascript dependencies
npm install

Usage

# Start application, then go to http://127.0.0.1:5000/ and click stuff
python app.py

# TODO start live
# TODO check cello sounds for the synth like https://www.youtube.com/watch?time_continue=97&v=rU2ieu5o5DQ

# TODO
curl --output "checkpoints/wavenet-ckpt.tar" "http://download.magenta.tensorflow.org/models/nsynth/wavenet-ckpt.tar"

# TODO
python nsynth.py --checkpoint="checkpoints/wavenet-ckpt/model.ckpt-200000" --wav1="sounds/83249__zgump__bass-0205-crop.wav" --wav2="sounds/371192__karolist__acoustic-kick-long.wav" --sample_length="80000"

Magenta.js demos

• IRCAM demo 01: Showcase a simple usage of GanSynth
• IRCAM demo 02: Showcase a combined usage of MusicVae and GanSynth

Freesounds

CC0 1.0 Universal (CC0 1.0) Public Domain Dedication:

• The wav file “sounds/83249__zgump__bass-0205.wav” is from https://freesound.org/people/zgump/sounds/83249/
• The wav file “sounds/371192__karolist__acoustic-kick.wav” is from https://freesound.org/people/karolist/sounds/371192/