Microsoft Debuts New AI Voice Models for Real-Time Conversations

Microsoft has expanded its artificial intelligence portfolio with the launch of new voice-focused models aimed at improving real-time conversations between people and AI systems. The new technology includes MAI-Transcribe-2-Streaming, a speech-to-text model capable of processing spoken language continuously, alongside MAI-Voice-2.1 and MAI-Voice-2.1-Flash for generating spoken responses. Microsoft is positioning the models as building blocks for a new generation of voice agents that can listen and respond with much less delay.

The MAI-Transcribe-2-Streaming model is designed to begin transcribing speech while a person is still talking instead of waiting until an entire sentence has been completed. Microsoft says the system can produce its first transcription results just over 100 milliseconds after receiving audio and continuously refine the transcript as more speech arrives. It supports 60 languages and automatic language detection, making it suitable for applications such as live captions, dictation, customer-service systems and conversational AI assistants.

Microsoft has also introduced MAI-Voice-2.1, a text-to-speech model designed for multilingual applications. The model supports 23 languages and 26 regional locales while allowing a consistent voice identity to be maintained across different languages. The company says the system is designed to produce natural accents appropriate to the language being spoken, potentially allowing businesses to deploy voice assistants across international markets without having to create a separate voice system for every language.

A faster version, MAI-Voice-2.1-Flash, is aimed at applications where response speed and operating costs are particularly important. Microsoft says the Flash model can generate audio with very low latency and is designed for high-volume voice applications. Combined with the streaming transcription model, the technology could allow AI agents to begin understanding a person’s request before they have finished speaking and prepare responses or actions much earlier in a conversation.

The launch highlights Microsoft’s broader push to make voice interaction a central part of its AI strategy. The company is also integrating voice capabilities into Microsoft Foundry, where developers can build and deploy voice agents using different AI models, tools and enterprise data. Microsoft says these voice agents support natural turn-taking, interruptions and spoken interactions across more than 80 languages and 140 locales, signalling a growing shift from traditional text-based chatbots toward AI systems that can communicate more naturally through speech.