Natural high-quality faster-than-real-time text-to-speech synthesis with minimal data. MIT 2019.