Transcribe audio and video in 64 languages

When you upload an audio or video file, the transcription engine analyses the first few seconds of speech and identifies the language automatically — no manual selection is needed. This works across accents, dialects, and mixed-language recordings. The detected language is shown alongside the transcript so you can verify it was identified correctly.

OpenAI Whisper covers 57 of the 64 languages listed here and is the established choice for broad multilingual coverage. MAI-Transcribe 1.5, a preview model, covers 43 languages and is typically faster than Whisper on the languages it supports. Gemini 3.5 Transcribe, also a preview model, has the widest coverage at 60 languages. The 7 languages Whisper cannot transcribe (Assamese, Bengali, Gujarati, Malayalam, Odia, Punjabi, Telugu) are covered by both MAI-Transcribe and Gemini, so picking either one works for them.

Languages are rated across three accuracy tiers based on the volume and quality of training data. High accuracy languages (including English, Spanish, French, German, Chinese, Japanese, and Portuguese) produce clean transcripts even from informal conversational speech. Good accuracy languages work reliably for clear recordings in a quiet environment. Basic accuracy languages have limited training data — results are usable but benefit from clear pronunciation and minimal background noise.

All 64 languages are supported in the same upload interface — MP3, WAV, M4A, FLAC, MP4, MOV, and other common formats are accepted up to 2 GB. Results are available as plain text, SRT subtitles, or VTT subtitles; the subtitle formats let you import captions directly into video editing software. Whisper provides no coverage for 7 of them (Assamese, Bengali, Gujarati, Malayalam, Odia, Punjabi, Telugu); MAI-Transcribe 1.5 and Gemini 3.5 Transcribe both handle those, and a job in one of them runs on MAI-Transcribe unless you pick Gemini yourself. Each card in the grid below lists the models that cover that language. All 3 models detect the spoken language automatically and return output in the language of the recording, so no manual language selection is needed before uploading. Transcription results are typically ready in under a minute; longer recordings are split and processed in parallel, then merged automatically before delivery. Files up to 2 GB are accepted and are deleted from the service within 24 hours — they are never stored permanently or used to train AI models.