Ghoost voices at the end of sentence in audio generation.

[Tortoise17]

I have trained an additional language other than English with almost 10,000 hours. Still when I use it to produce the audio voice, there are strange voices at the end of sentence. This happens specially, when there is text of more than 1 sentence. Is there any solution to fix or any hint which I could properly set while training?

Same length I tried for English model and it has no such issue. Kindly guide