Microsoft launches streaming MAI voice models in Foundry

Microsoft released three voice models in Microsoft Foundry: MAI-Transcribe-2-Streaming transcribes live speech with first hypotheses in the low hundreds of milliseconds, while MAI-Voice-2.1 and MAI-Voice-2.1-Flash handle expressive multilingual text-to-speech. The transcription model debuted at #1 for accuracy on Artificial Analysis.

• MAI-Transcribe-2-Streaming returns first hypotheses in hundreds of milliseconds
• Model debuted at #1 for accuracy on Artificial Analysis
• MAI-Voice-2.1 is an expressive multilingual TTS model
• MAI-Voice-2.1-Flash targets high-volume, low-latency voice apps