Microsoft has started previewing MAI-Realtime, a proprietary voice AI model that handles two-way conversation in real time. The move signals the company's push to build its own voice technology rather than rely on outside vendors.
What MAI-Realtime does
MAI-Realtime is a bidirectional voice stack — meaning it can listen and speak simultaneously, much like a human conversation. Microsoft developed the model in-house, and the preview gives developers and enterprise customers early access to test it. The company hasn't said when a full commercial release might come, but the preview is open now.
For years, Microsoft has leaned on third-party voice services for products like Azure AI and Copilot. MAI-Realtime could change that. By controlling the entire voice pipeline, Microsoft can tailor the experience for its own platforms — Teams, Dynamics 365, Azure — without sharing data or paying licensing fees to external providers. That's a direct play for enterprise customers who want tighter integration and fewer middlemen.
The model also promises to improve user experience. Real-time bidirectional voice makes interactions feel more natural, which matters for customer service bots, virtual assistants, and meeting transcription tools. If MAI-Realtime works as advertised, it could make Microsoft's voice offerings more competitive against similar products from Amazon and Google — though the facts don't name those competitors.
Reducing reliance on external providers
Microsoft has been investing heavily in AI infrastructure, from large language models to speech recognition. MAI-Realtime fits that pattern. The company already uses its own Azure Speech services, but those are built on a mix of internal and external technology. This new stack is entirely Microsoft's, from the model architecture to the inference engine. That gives the company more control over performance, cost, and data privacy — a selling point for regulated industries like healthcare and finance.
The preview is limited. Microsoft hasn't disclosed pricing, latency benchmarks, or which regions are supported. Developers who want to try it need to apply through Azure AI Studio. The company is likely gathering feedback before a broader rollout.
Microsoft hasn't set a timeline for general availability. The preview will run for several months, and the company will probably refine the model based on early user reports. For now, the question is whether MAI-Realtime can deliver on its promise — and whether enterprise customers will adopt it fast enough to justify Microsoft's investment.
