Microsoft Foundry now offers MAI-Voice-2 in public preview, a first-party voice model from the Microsoft AI team. MAI-Voice-2 generates natural speech across more than 10 languages and supports voice cloning from a short reference sample, plus voice prompting so developers can shape tone, pacing, and style with plain instructions. Customers can build multilingual virtual agents, accessibility experiences, and audio content workflows without stitching together separate cloning and prompt-engineering tools. Because MAI-Voice-2 is sold directly by Microsoft, it ships with the same billing, content safety, and responsible AI policies as other Microsoft-published models in Foundry. Developers can deploy it through the Foundry catalog and call it alongside other speech and language models in a single project.
Learn more.