Audio Synthesis with Matrix Product States

• TensorFlow
• Python 3.6.6

• model.py

It defines our RNN model and contains other methods, to sample from the trained model for example.

• training_estimators.py

It loads the training data, and runs the training. A custom estimator is used.

• reader.py

It picks out a subset of the data, that contains a desired feature: pitch, instrument ...

• make-small-dataset.py

It imports reader.py to make the dataset.

1. Sampling
2. What about initialization? Glorot variance depends on size.
3. Monitor change of state to see if it is small
4. Exploring latent space