Copilot evolves: Microsoft launches MAI-Voice-1, MAI-1-preview with diverse voice choices

Microsoft has unveiled its first two in-house models, underscoring the tech giant’s ambition to broaden its AI portfolio beyond its long-standing reliance on OpenAI systems. The new models—named MAI-Voice-1 and MAI-1-preview—were officially announced on Thursday and have already begun appearing within Microsoft’s product ecosystem. Notably, MAI-Voice-1 introduces a versatile voice wardrobe offering a range of accents, emotions, and styles to suit any script.

Mustafa Suleyman, CEO of Microsoft AI, took to X (formerly Twitter) to share the announcement. He wrote, “Excited to share our first MicrosoftAI in-house models: MAI-Voice-1 and MAI-1-preview. Details and how you can test below, with lots more to come.”

The MAI-Voice-1 model focuses on speech generation and is built for speed. According to Microsoft, it can generate a full minute of audio in under one second using just a single GPU, positioning it as one of the fastest and most efficient voice AI solutions available today.

The technology is already integrated into Copilot Daily, a feature that narrates news updates through AI voices, and is also being used in tools that create podcast-style discussions to simplify complex topics.

The tool comes loaded with a wide range of voices to bring any script to life. To kick things off, you can pick between two modes—Emotive and Story. Next, you get to choose the voice itself, with accents spanning American, British, and Australian varieties.

But the real fun begins with style. Want your AI to sound joyful, confused, curious, relieved, or even a little shy? No problem. Prefer something more narrative, like a news anchor or audiobook narrator? That’s there too.

For those who like things quirky, there’s an entire buffet of eccentric options—from vampire, pirate, and British butler, to ghost, robot, Bostonian cop, or even animal impersonations like an owl or a cow.

In parallel, the company has introduced MAI-1-preview, a large-scale text model trained on 15,000 Nvidia H100 GPUs. Designed for instruction-based queries and conversational responses, this system provides what Microsoft calls “a glimpse of future offerings inside Copilot,” signalling that it will play a more central role in the company’s AI assistant in the near future.

The MAI-1-preview has been released in a limited capacity, supporting select text-based tasks. Microsoft has also made it available for public testing on LMArena, an AI benchmarking platform, allowing researchers and developers to assess its performance.