Meta Launches Transcription: 20 Speakers and 70 Languages in Real Time

Meta has released Muse Voice Transcribe — a single model that simultaneously identifies 20 or more speakers and works with 70+ languages, including code-switching — when someone mixes words from different languages within the same sentence. The model handles sessions up to an hour long, costs $3 per 1,000 minutes, and has been available via API since September 1, 2026.

Until now, accurate transcription of multilingual meetings with multiple participants required several specialized tools and manual stitching of results. Now it's a single API call, and the transcription market for international teams, call centers, and medical interviews is getting both cheaper and more accurate.