Spaces:
Running
Running
Pronunciation Issue: "nan4 xiong1 nan4 di4" being spoken as "nan2 xiong1 nan2 di4"
#2
by
sa556693828
- opened
Hi, I encountered a pronunciation issue when using the TTS model.
I provided the exact pinyin input: nan4 xiong1 nan4 di4 for the sentence “难兄难弟”, where both instances of "难" should be pronounced in the fourth tone (nan4). However, in the generated audio, both were spoken in the second tone (nan2), which changes the intended meaning.
Is there a way to enforce tone accuracy strictly based on the provided pinyin input? Or is this a known limitation of the model?
Thank you in advance!
Thanks for flagging this! You can fix it in either of these ways:
Use the API’s pronunciation_dict to correct the pinyin, e.g. 难兄难弟/(nan4)(xiong1)(nan4)(di4).
Or skip the API and input (nan4)(xiong1)(nan4)(di4) directly, then select Chinese as the language.