Microsoft expands AI model family MAI with three audio models

Microsoft’s new streaming transcription model recognizes 60 languages and posts first words in about 100 milliseconds, refining them as more context arrives, which lets voice agents start thinking mid-sentence. Its two companion voice models speak 23 languages with native accents and can clone a voice from seconds of audio, after a consent review.

Summary written with AI by HowTechMoves, based on reporting by heise online.