Microsoft unveils MAI-Voice-1 and MAI-1 Preview AI Models

Microsoft Unveils New In-House AI Models: MAI-Voice-1 and MAI-1-Preview

In a significant move, Microsoft has announced the introduction of two proprietary AI models—MAI-Voice-1 and MAI-1-preview—marking a shift from its previous reliance on OpenAI’s frameworks. Traditionally, with each new release from OpenAI, Microsoft promptly integrated those models into its Azure platform and various products. Now, with these latest innovations, the tech giant is poised to enhance its AI offerings further.

Introducing MAI-Voice-1

The MAI-Voice-1 model is a sophisticated speech generation tool that is already available through Copilot Daily and Podcasts. Microsoft has rolled out the Copilot Audio Expressions experience, enabling users to experiment with the model’s capabilities. Users can input text, choose from various voice options, and select desired styles to create high-quality, expressive audio outputs, complete with the ability to download their creations.

This new voice model is remarkably efficient, capable of producing one minute of audio in less than a second using a single GPU, emphasizing its fast performance for real-time applications.

MAI-1-Preview Enters Public Testing

Additionally, Microsoft has initiated public testing for MAI-1-preview on LMArena, a prominent platform used for community evaluation of AI models. This is the first instance of an end-to-end foundation model developed entirely in-house by Microsoft, showcasing a potential evolution in the Copilot suite of offerings. The ongoing development cycle aims to enhance model performance continuously, with further updates expected soon.

Equipped with a mixture-of-experts (MoE) architecture and pre-trained on an extensive dataset, MAI-1-preview utilizes almost 15, 000 NVIDIA H100 GPUs. Notably, this model is adept at following user instructions and delivering relevant answers to common queries. Users can anticipate its integration into select text-based functionalities within Microsoft Copilot over the next few weeks.

Strategic Collaboration and Future Outlook

Microsoft has reassured users that MAI-1-preview is not designed to replace OpenAI’s models within Copilot. The company envisions a collaborative future where it harnesses the strengths of its own models, those from partners, and contributions from the open-source community. Moreover, MAI-1-preview is available via API for trusted testers, widening the scope for feedback and enhancement.

Commenting on the advancements in their AI portfolio, the Microsoft AI team stated:

We have an exciting roadmap of compute at MAI, with our next-generation GB200 cluster now operational. And we have an ambitious mission we truly believe in. We’re also fortunate to partner with incredible product teams giving our models the chance to reach billions of users and create immense positive impact.

By advancing its foundational AI models alongside leveraging existing technologies from OpenAI, Microsoft aims to establish a robust and versatile AI infrastructure for all its products and services.

Source & Images