I have trained the model for 200k steps, and still, the synthesised results are extremely bad. The sampling rate I have used is 22050 Hz and the batch size used is 16.

This is how my loss curve looks after 200k steps. Can you help me with what can I do now to improve my synthesized audio results?