Use MAI-Voice-2.1 audio in a Sume lip sync clip: import it first

Yes: you can make a talking clip on Sume from audio you made elsewhere, including MAI-Voice-2.1, as long as the file becomes Sume-hosted first. Sume's lip sync routes take audio_url on the Sume media host, up to 10 MB, and the H3 Max route also needs 5 to 14.8 seconds of audio and a measured duration_seconds. MAI-Voice-2.1 is not listed in the Sume TTS Router, so Sume cannot generate that voice for you.

Microsoft's page (read 2026-10-13) says MAI-Voice-2.1 supports instant voice matching from short reference clips without fine-tuning and is reachable in the MAI Playground, Copilot Audio Expressions and Foundry. OpenRouter lists it too. Sume lip sync limits come from the repo's H3 Max lip sync notes (origin/main b28f3b0).

Which model takes which audio?

Fabric stays the default talk model and has no 5 second floor. H3 Max is the explicit alternative inside its 5 to 14.8 second window; the API refuses duration_seconds outside it and never clamps. Keep the original wav or mp3 untouched: the packet guidance is not to cut or re-synthesize audio to make it fit.

Is it cheaper to stay inside Sume?

That depends on the voice you need. Sume's Sonic rows bill about $0.0475 per 1,000 characters before cent rounding; Microsoft lists MAI-Voice-2.1 at $22 per 1M characters, which is $0.022 per 1,000 before any reseller fee. The two prices are not like for like, because the Sume figure includes the 1.25 house margin and a one-cent job minimum.