Microsoft AI Unveils New Voice Models for Building Conversational Experiences

Microsoft AI has released two new models for building voice agents: MAI-Transcribe and MAI-Voice. According to Microsoft, these models offer fast, low-cost, and accurate audio understanding and generation capabilities.

The MAI-Transcribe-2-Streaming model is a top-rated transcription model that delivers real-time transcripts in 60 languages with continuous language detection. It produces partial transcripts in just over 100 ms of receiving audio, allowing voice-enabled applications to act on speech before the speaker finishes speaking.

Microsoft AI's strongest multilingual text-to-speech model yet is MAI-Voice-2.1, which supports 23 languages and 26 locales while enabling a single 'voice' across all languages with a native accent. The MAI-Voice-2.1-Flash variant is optimized for high-volume, latency-sensitive workloads and can generate 45 seconds of audio with an end-to-end latency of just 150 ms.

The new models are designed to work together to keep conversations moving naturally in voice agent experiences such as customer service agents, multilingual assistants, or interactive learning and media. Users can experience them in the new Chatter demo on the MAI Playground.