Cantonese (yue) dictation writes standard Chinese instead of spoken Cantonese
When I dictate with the language set to Cantonese (Cloud mode), the text comes out as Standard Written Chinese rather than what I actually said. For example, saying 我唔想返工 gives 我不想上班.
It happens in the voice model, not the cleanup step: in the history entry, rawResult and llmResult are both 我不想上班 (model S1-Voice).
Cantonese speakers write messages and chats in written Cantonese (唔, 係, 喺, 佢, 嘅, 咗), so the rewrite into standard Chinese makes Cantonese dictation unusable for everyday writing. It is possible: SenseVoice, for example, keeps colloquial Cantonese characters.
Could Cantonese have an option to output written Cantonese, ideally in traditional characters with Hong Kong forms, or make that the default?
Pending