[Language Detection] Whisper Large-v3 cannot detect cantonese audio as "yue" language

[josephwong14wkh]

I am using whisper large-v3 to transcribe an audio spoken in Cantonese. However, I found that whisper detected the language as "zh" (Chinese), instead of "yue" (Cantonese), and the transcription becomes written-language format, instead of spoken-language format. Is this behavior normal? Is there any method to solve this issue?

[samarasimhapeyala]

Can you try passing the language(of audio) argument as well --language yue or --language cantonese

[josephwong14wkh]

Thank you for your response. When I pass the language argument --language yue, the text is transcribed in written-language format, rather than spoken-language format. But in my case, the transcription is not very important. I just want whisper to differentiate whether the audio is spoken in mandarin or cantonese. However, it seems what whisper cannot distinguish these two languages (both large-v3 and large-v3 turbo version). Is there any idea for language detection part?