Is "multi-segment language detection" based on accent or actually spoken speech?

[toanhn]

When I use "multi-segment language detection" https://github.com/SYSTRAN/faster-whisper?tab=readme-ov-file#multi-segment-language-detection.
I found that it detects language based on accent rather than actual spoken speech.
Ex: I have an audio file, a speaker is Malaysian and he says English, if I use "native language detection" with the first 30 seconds of the audio file then it detects the speech is in English. But when I use "multi-segment language detection" then it detects the speech is in Malay.

Does anyone have any idea on this issue? Thanks.