Microsoft AI Releases Impressive Transcription and Voice Models

First reported by Thurrott at · Updated · 2 sources

Microsoft's latest MAI-Transcribe and MAI-Voice models are available for building voice agents, according to Thurrott. MarkTechPost identifies MAI-Transcribe-2-Streaming as the company's first real-time speech-to-text model. It also reports that the model holds the top position among 38 models on the Artificial Analysis AA-WER Streaming benchmark.

  • MarkTechPost reports 2.5% word error rates, with final transcripts at 0.13 seconds and first partials at 0.12.

Covered by 2 publishers within 13 hours of the first report.

Reporting2

Thurrott Microsoft AI Releases Impressive Transcription and Voice Models · Paul Thurrott
MarkTechPost Microsoft AI Releases MAI-Transcribe-2-Streaming: #1 Real-Time Speech-to-Text Model on Artificial Analysis · Michal Sutter

Related

Microsoft releases Exchange V2 patches for authorization flaws
Skip the Wait. Microsoft's 26H2 Update Fixes the Parts You Hate Most About Windows 11
Skip the Wait. Microsoft's 26H2 Update Fixes the Parts You Hate Most About Windows 11
New Intel Arc driver adds Gears of War E-Day and Ace Combat 8 support

Topics Microsoft