Bert 不识别 'ー'

[ffftian]

我之前有训练vits和so-vits-svc的经验，在我尝试用一致的训练集时他对ー报错，我改成汉字的一还是无法解决问题，日语中拥有大量的ー。理论不应该错在这上。

[meimisaki]

I encountered the same problem. The Japanese-Extra branch doesn't handle it correctly in the clean_text function. This error will be triggered when we feed it with something like "〜". As a workaround you can simply filter it out from the dataset.