Waveform generation for text-to-speech synthesis using pitch-synchronous multi-scale generative adversarial networks.