Transcribe Thai audio to text

Thai works in the free trial: three transcriptions of up to 10 minutes per calendar month, no account, as text with timestamps. It runs on Whisper Large v3 Turbo, a model trained on roughly 100 languages.

With a day pass, credit or Pro, Microsoft MAI-Transcribe-2 takes over. It separates speakers, so the transcript shows who said what — useful for interviews, meetings and conversations in Thai. Microsoft reports an average word error rate of 5.2% across the model's 60 languages (FLEURS benchmark); individual languages score above or below that.

Thai is a tonal language: the same syllable changes meaning with pitch. That makes a clean recording especially important — reverb, background music or heavy compression blur exactly the pitch contours the meaning depends on.

Written Thai does not put spaces between words. The model decides where words begin and end, so it is worth checking technical terms and names in the transcript.