Microsoft has stepped into the spotlight with its own artificial intelligence tools. On August 28, the company rolled out its first in-house AI models: MAI-Voice-1 and MAI-1 preview. This move shows a new direction for Microsoft, which has relied on OpenAI’s technology for some time.
One of the new offerings, MAI-Voice-1, is an impressive audio model. It can create a full minute of speech in less than a second. What’s even better is that it only needs a single graphics processing unit, or GPU, to do this. This makes it very efficient.
MAI-Voice-1 is already at work in some parts of Microsoft’s Copilot system. For example, Copilot Daily uses this AI to act like a news anchor, summarizing the day’s events. It even creates conversations that sound like a podcast, explaining different news topics. Users can try MAI-Voice-1 themselves in Copilot Labs. There, you can type what you want the AI to say and even change its tone and speaking style.
The other new model, MAI-1 preview, offers a peek into future Copilot features. This AI is a heavyweight, trained using more than 15,000 NVIDIA H100 chips. It is designed to understand everyday commands, making it more intuitive for users.
Microsoft plans to use MAI-1 preview for specific tasks within Copilot AI. While Copilot currently uses models from OpenAI, MAI-1 preview will add another layer of capability. The company has already put MAI-1 preview through its paces using LMArena, a special platform for testing AI. This new release marks a notable step for Microsoft as it builds out its own AI capabilities.