Inconsistent Pronunciation of "BAN/ban" and "BUFF/buff" in GPT-SoVITS

[Strive-for-excellence]

I have noticed an issue with GPT-SoVITS where the pronunciation of certain words differs depending on their capitalization. Specifically:

Description:

"BAN" and "ban" are pronounced differently.
"BUFF" and "buff" are pronounced differently.
This inconsistency can affect the usability and naturalness of the model, especially when dealing with context-sensitive or case-sensitive text-to-speech tasks.

Steps to Reproduce:

Input "BAN" and "ban" in the model.
Input "BUFF" and "buff" in the model.
Compare the pronunciations for each pair.

Expected Behavior:

The pronunciation of these words should remain consistent regardless of capitalization, unless explicitly designed to convey different meanings.