Meta Superintelligence Labs has released Muse Voice Transcribe, a single autoregressive model that performs streaming automatic speech recognition (ASR), speaker diarization for 20+ speakers, and endpointing in one pass. The model is available as a hosted API on the Meta Model API under the name muse-voice-transcribe-1.0.

  • It achieves 3.1% final-transcript word error rate at 0.16s latency on Artificial Analysis AA-WER Streaming benchmarks.
  • Reinforcement learning trains an adaptive delay policy that varies per word to balance accuracy and speed.
  • The model supports audio inputs exceeding one hour and native code-switching across 70+ languages.
  • Pricing is $3.00 per 1,000 audio minutes, undercutting competitors like Cartesia Ink-2 and ElevenLabs Scribe v2 Realtime.

The system powers dictation in Meta AI for Mac and Muse Code, offering a unified solution that reduces latency and failure modes associated with multi-model stacks.