Speech synthesis utilizing audio waveform difference signal(s)

Techniques for neural speech synthesis that model audio waveform difference signals, from the WaveRNN/efficient neural audio synthesis line of work at DeepMind.