Supplementary Material of HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis