Microsoft Tests MAI-Realtime Voice AI to Eliminate Conversation Pauses
Microsoft is quietly testing MAI-Realtime, its latest voice AI model that can talk and listen simultaneously. This technology aims to eliminate the pause between conversation turns found in current mainstream voice assistants. In contrast, MAI-Realtime allows for a more natural flow of communication, enabling users to interrupt mid-sentence or change subjects without losing context.
The model currently supports 17 languages and can detect language switches during conversations. It also integrates tools like live web search, allowing users to access information seamlessly while chatting with the AI. Notably, MAI-Realtime is designed for conversation purposes only and will not engage in singing or producing non-speech sounds.
This development marks a significant step towards Microsoft's goal of owning the voice model, safety policy, and billing process, rather than relying on OpenAI's services. The company has been working to create its own speech lineup, including MAI-Voice-1, MAI-Voice-2, and MAI-Transcribe-1, under Mustafa Suleyman's Microsoft AI division.
Google also has a similar technology called Gemini Live, which offers full-duplex conversation mode. The competition among tech giants is pushing the development of more natural and seamless interactions between humans and AI assistants.