mispronounces some words

[pakdamanAli]

Checks

• This template is only for research question, not usage problems, feature requests or bug reports.
• I have thoroughly reviewed the project documentation and read the related paper(s).
• I have searched for existing issues, including closed ones, no similar questions.
• I am using English to submit this issue to facilitate community communication.

Question details

Hello,
I fine-tuned the F5 model for the Persian language with 100 epochs, and 60 h data
but it still mispronounces some words that are even present in the training data. Where could the problem be?

[brambox]

I may be wrong but in my experience it maybe is an internal model flaw. it will always do these here and there. Slowing speed or trying other reference audio can help but its always there no mater how much you train or hours you feed it. actually it may get worse with more data. But if someone of developers or people who knows better how it works can give us some help is welcome.

[HT5757]

I’ve found that the pronunciation is much better when use_ema is unchecked