MAI-Transcribe-2 on OpenRouter: wav, base64, transcript JSON

OpenRouter's page for microsoft/mai-transcribe-2 shows a dedicated endpoint, POST https://openrouter.ai/api/v1/audio/transcriptions, that takes the model id and the audio as base64 with its format, and returns JSON with the transcript and usage. The page lists WAV as the processed audio format. A verbose response can add language, duration and timing when the upstream provider supports it.

Microsoft's own page lists 60 languages, diarisation, timestamps and keyword biasing for the model, plus a streaming variant. Facts here were read on 2026-10-13 from both pages.

How do I get WAV from a video?

Sume's audio detach returns sample-exact wav (pcm_s16le) from a hosted video, and sample_rate: 16000 with channels: "mono" is the speech-to-text shape. Detach once, download the file, then encode it for the transcription call. The detach source cap is 1800 seconds and the output cap is 900 seconds, so cut longer audio with range.

What if I only want captions?

You do not need a separate transcript. Sume's POST /v1/video-captions runs speech-to-text itself on a public HTTPS video and burns the captions; a flat job estimate covers clips up to 60 seconds (see the rate card). Send script_text to keep your spelling of product names, or words if you already hold timed words from another transcriber. Send only one of script_text, words, cues and segments.

curl -X POST https://api.sume.com/v1/video-captions \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: caption-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/clean.mp4",
    "style": "slam",
    "script_text": "Say hello to the Sume developer platform."
  }'