Microsoft has announced an exclusive early preview of “MAI-Realtime,” a new bidirectional voice model that is currently undergoing testing on the MAI Playground platform with limited access. This model aims to provide a more natural conversational experience compared to its predecessor, MAI Voice 2, and features two available voices, Victoria and Grant. Notably, MAI-Realtime supports automatic language detection and can switch languages mid-conversation, accommodating a total of fourteen languages, including English, Spanish, Chinese, and Arabic.

Microsoft: Microsoft develops advanced AI technologies as part of its ongoing MAI initiative focused on generative and voice models. The company is sharing an exclusive early preview of its new bidirectional voice model MAI-Realtime, which builds on prior efforts and integrates web search along with additional tools.
MAI Voice 2: MAI Voice 2 is an earlier Microsoft voice model that served as a benchmark for naturalness in conversational AI. The new MAI-Realtime model is positioned as a significant improvement over it in terms of fluidity and tool operability.
MAI-Realtime: MAI-Realtime is Microsoft’s upcoming bidirectional voice AI model designed for more natural conversational experiences. It is currently undergoing limited testing on the MAI Playground with two voices available and supports automatic language detection plus mid-conversation switching.

Development Stage: MAI-Realtime is in early preview testing on the MAI Playground platform with limited access and offers two distinct voices named Victoria and Grant.
Multilingual Capabilities: The model supports automatic language detection and switching among English, German, Spanish, French, Italian, Portuguese, Japanese, Korean, Chinese, Dutch, Hindi, Indonesian, Arabic, Russian, Turkish, Vietnamese, and Thai.