Microsoft AI releases MAI-Transcribe-2-Streaming and the MAI-Voice-2.1 voice models
In one sentence On October 1 Microsoft AI published its first streaming transcription model, MAI-Transcribe-2-Streaming (60 languages, first words in just over 100 ms, 0.54 dollars per hour), alongside the MAI-Voice-2.1 and MAI-Voice-2.1-Flash text-to-speech models.
Microsoft AI, the division building Microsoft's in-house models, added three voice models to its lineup. The most interesting for anyone building voice assistants is MAI-Transcribe-2-Streaming: it transcribes speech in real time while the person is still talking, instead of waiting for the recording to end.
The model covers 60 languages and detects the spoken language automatically and continuously, which helps in conversations that switch between languages. The first words appear in just over a tenth of a second. Microsoft places it first on the Artificial Analysis leaderboard for accuracy.
Alongside it come two models that do the opposite, turning text into speech: MAI-Voice-2.1, which keeps the same voice across 23 languages and can clone one from a few seconds of audio, and the Flash variant, faster and cheaper. All are closed models used through the cloud.
Companies
Microsoft
Tools
MAI-Transcribe-2-Streaming, MAI-Voice-2.1, MAI-Voice-2.1-Flash
Tags
Sources