Word repetion in english after finetuning.

[hassan-webm]

Checks

• This template is only for question, not 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 confirm that I am using English to submit this report in order to facilitate communication.

Question details

I fine-tuned the F5 TTS model with 350M parameters on 215 hours of Spanish and 35 hours of English. After approximately 200 epochs (or 280k steps), inference on the checkpoint shows good performance in Spanish , but English exhibits word repetition issues. Could this be due to catastrophic forgetting? If so, what would be an appropriate ratio of English data to the target language to prevent forgetting? And to fix this should I add more English data to the corpus and continue training?

Example:
Ref text: "Yesterday, I watched a documentary about the diverse wildlife in the Amazon rainforest, which left me amazed by the beauty and complexity of nature."
https://jumpshare.com/s/jLfTAbmg2efuPzVvBB9s

Also if you notice it pronounced, "yesterday" as "gesterday" and also mispronounced amazed then repeated it.

[tcreek]

I have experienced word repetition just using it as is.