Spaces:
Running
on
Zero
Running
on
Zero
worse persian output
#9
by
devops724
- opened
i tested this model on Persian, it generate output in this language but in accent of Chinese
is there any finetune guide for this model
So, you tell me that Qwen 3 TTS generates Persian but with a Chinese accent?
He is wrong actualy... It's not persian with Chineese Accent... That's Arabic Accent. I tried to finetune it on persian but the Loss remains above 12 ... I guess tokenizer did not include persian letters completely maybe?