[Tacotron2+Waveglow/PyTorch] Audio Clicks and Noisy Audio

[notdxlin]

Related to Tacotron2 + Waveglow/PyTorch

I am currently attempting to train Tacotron2 + Waveglow models for speech synthesis using the Biaobei data set. I'm using a downsampled version of the audio from 48k to 16k, but am noticing some audible clicks and noise in the synthesized speech audio. Attached is the spectrogram for an utterance I created after training Tacotron for 1500 epochs and 1200 for Waveglow. Does anyone have an idea what I might be doing wrong? I'm not sure if I need to supply more information, if so please let me know!

Environment
Please provide at least:

• Container version (e.g. pytorch:19.05-py3): pytorch: 1.6.0 python 3
• GPUs in the system: (e.g. 8x Tesla V100-SXM2-16GB): Quadro RTX 5000 x2
• CUDA driver version (e.g. 418.67): 11.0
  nvcc: NVIDIA (R) Cuda compiler driver
  Copyright (c) 2005-2020 NVIDIA Corporation
  Built on Wed_May__6_19:09:25_PDT_2020
  Cuda compilation tools, release 11.0, V11.0.167
  Build cuda_11.0_bu.TC445_37.28358933_0