MAI-Voice-2.1: 23 languages and voice matching; Sume takes a voice id

Microsoft's MAI-Voice page lists 23 languages for MAI-Voice-2.1 and describes "instant voice matching" that captures a voice from a short reference clip with no fine-tuning. It does not state how long the clip must be. Sume TTS 1.0 has no reference-audio field: you send a transcript plus a ready voice id or an avatar, and the voice must be one Sume already holds.

Microsoft's claims are from its MAI-Voice page, read 2026-10-01; Sume's from the OpenAPI schema behind the API reference.

What does a Sume voice selector accept?

Either avatar_id / avatar_handle (Sume resolves that avatar's voice, which works when its voice status is ready) or voice.id. A voice id must be a TTS voice UUID or a Voices library id starting voi_. Anything else fails with 400 invalid_voice_id before a job is queued or credits are reserved.

How do the two models compare on input?

What does a request with a stored voice look like?

Pass the voice and the language the text is written in.

curl -X POST https://api.sume.com/v1/tts-1.0/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "transcript": "Bonjour et bienvenue.",
    "voice": { "id": "voi_0123456789abcdef0123456789abcdef" },
    "language": "fr"
  }'

Can I clone a voice from a clip on Sume?

Not through the TTS request. Voice cloning API covers what exists today. If you hold a consented voice elsewhere, check the rules of both vendors before reusing it.