Microsoft releases real-time transcription and speech synthesis models

On 2 October 2026, Microsoft AI released MAI-Transcribe-2-Streaming, a real-time transcription model for voice agents. According to The Decoder, Microsoft says the model ranks first for accuracy on Artificial Analysis, supports 60 languages, returns its first partial result in just over 100 milliseconds, and will be offered at an introductory price of $0.54 per hour of audio by the end of the year. The report did not disclose details of the speech synthesis model, and the story is still at an early stage.

Written by AI from the reports · updated 14 d ago