Muse Voice Transcribe: Meta's real-time speech-to-text model

Melvin Vivas · X post · 2026-09-02 · Open on X

Topics: Industry Trends & Job Market · Level: beginner

Summary

The creator highlights Muse Voice Transcribe from Meta Superintelligence Labs, a real-time audio model. It does streaming speech recognition, speaker diarization for 20+ speakers, and endpointing, and handles multiple languages, including switching languages mid-sentence.

Key points

• Real-time streaming speech-to-text (ASR)
• Tells apart 20+ speakers (diarization)
• Endpointing: detects when a speaker has finished
• Handles multiple languages, including switching mid-sentence (code-switching)
• Meta Superintelligence Labs' first real-time audio perception model

Resources mentioned

• Muse Voice Transcribe · tool · developer.meta.com · paid
  Meta Superintelligence Labs' real-time multilingual speech-to-text model with diarization and endpointing.