Microsoft releases streaming transcription and two multilingual voice models

Microsoft launched MAI-Transcribe-2-Streaming and two text-to-speech models on October 1, making all three available through Microsoft Foundry. The company says its streaming model transcribes speech in 60 languages, detects language changes continuously and returns preliminary text while a speaker is still talking. Transcription costs $0.54 per hour of audio at an introductory rate through the end of the year. MAI-Voice-2.1 and MAI-Voice-2.1-Flash support 23 languages and 26 locales, with prices of $22 and $15 per 1M characters, respectively. Microsoft says both preserve a speaker's identity across languages and support voice cloning from a few seconds of reference audio. The voice models are also available through OpenRouter; LiveKit availability is coming later.