Malay Subtitle Generator

Bahasa Melayu on video is a mixed language. A creator in Kuala Lumpur will say a sentence that begins in Malay, contains an English verb, and ends with lah, and that is not sloppy — it is how the language works among people who grew up with both. A transcription model that has been told the audio is Malay will try to spell the English words as Malay, and the result is unreadable. The model here treats the switch as normal and writes each word in the language it was said.

Spoken Malay also differs from written Malay in ways that matter for subtitles. Contractions, dropped affixes and the discourse particles that carry tone are everywhere in speech and rare in print. The transcript writes what was said in a form a Malaysian reader recognises, and the editor lets you standardise to formal Bahasa Melayu if the video calls for it — a corporate piece, say, rather than a vlog.

Malay and Indonesian are close enough to confuse a general model and different enough that a Malaysian audience notices when the transcript uses Indonesian spelling or vocabulary. The model separates the two, so a Malaysian video comes back in Malaysian Malay, and an Indonesian speaker in the same video is transcribed in Indonesian rather than being smoothed into one.