Existing work
	Machine Learning Systems can be made to predict F0, but… • Opaque results • Require large amounts of training data • Internal representation is typically very complex • Models of similar speakers aren’t internally similar