Microsoft launched public testing of its first in-house foundation model. MAI noted that these models represent the first steps in a larger roadmap. The company plans to orchestrate a range of specialised AI models designed for different use cases and user intents, combining its in-house work with contributions from partners and the open-source community.
Microsoft AI (MAI) has introduced two new models as part of its mission to build AI that empowers people and organisations everywhere. The goal of the company is to create “AI for everyone” responsible, reliable, and tailored to individual needs, while serving as a trusted companion for everyday use.
The first release, MAI-Voice-1, is Microsoft’s new speech generation model designed to provide high-fidelity, expressive audio. Already powering Copilot Daily and Podcasts, the model is also available to experiment with through Copilot Labs, where users can try storytelling as well as guided meditation demos.
MAI-Voice-1 can generate a full minute of audio in under a second on a single GPU, making it one of the most efficient speech systems currently available. Microsoft described voice as the “interface of the future” for AI companions, noting the model’s ability to handle both single- and multi-speaker scenarios.
MAI observed that these models represent the first steps in a larger roadmap. The company plans to orchestrate a wide range of specialised AI models designed for different use cases as well as user intents, combining its in-house work with contributions from partners and the open-source community.
Developers and testers can also apply for API access to MAI-1-preview, which Microsoft says is an important step toward building a flywheel of continuous improvement.