Cantonese -> Chinese / Mandarin

This is a rather niche language / dialect specific case, and may not be fixable above the model level (or maybe possible based on how the model is configured / parameterized?). But here goes:

When I dictate in Cantonese (a more colloquial dialect of Chinese), the model changes the text to formal written Chinese (basically the written form of Mandarin). Is there a way to instruct the model to not have this behavior?