Yesterday's GitHub update was great!!
But I'm having a problem. The Huggingface spaces you have generate very natural and too close to the given reference audio.
But when i installed the GitHub version it was a little different like a bit more fast speech and doesn't respect given reference audio, sounds too robotic, and also it takes 3-4 generations to get a perfect audio (not the previously said problems but the audio is morphed into non-verbal sounds).
is there any custom configuration you did in the Huggingface spaces?
my config is this
max_length=2048,
top_p=1,
temperature=0.8