This is reimplementation of the NSynth model as described in Neural Audio Synthesis of Musical Notes with WaveNet Autoencoders [arxiv:1704.01279].

The original TensorFlow v1 code can be found under in github:tensorflow/magenta.

Python package requirements for this code are torch >= 1.3.1 and librosa>=0.7.1. Further to load audio with librosa you need libsndfile which most systems should already have.

To replicate the original experiment you will have to download the NSynth dataset under https://magenta.tensorflow.org/datasets/nsynth#files in json/wav format.

A training script can be found in train.py which will train the model with the same parameters as in the original paper.

python ./train.py --datadir SOME_PATH/nsynth/ --device=cuda:0 --nbatch=32

The required argument datadir should be de full path to the NSynth dataset. (The path should contain folders nsynth-test, nsynth-train and nsynth-valid each of whom have to contain the folder /audio/ and the file examples.json.)

The other default arguments will give the same setting as in the original paper.