MAI-Voice-1

MAI-Voice-1 is Microsoft AI's debut in-house speech generation model, announced August 28, 2025 alongside MAI-1-preview as the opening of the MAI model family under Mustafa Suleyman. Microsoft's headline claim: a full minute of high-fidelity audio generated in under one second on a single GPU.

The model powers Copilot Daily, Podcasts, and the Copilot Labs audio expressions experience, where users pick a voice, style, and mode to generate expressive audio. It supports single and multi-speaker scenarios and has since been rolled out to developers via Microsoft Foundry and MAI Playground, with custom voice creation from a short audio snippet and pricing reported around $22 per million characters. It remains closed weights, available only through Microsoft's services.