Meta Superintelligence Labs releases Muse Voice Transcribe ASR model

Meta Superintelligence Labs released Muse Voice Transcribe, a real-time audio perception model for streaming speech-to-text. The model handles speaker diarization with more than 20 speakers, turn-endpointing, and multilingual speech with code-switching, all within a single model. It processes audio in 80ms chunks and dynamically decides after each chunk whether to emit text or continue listening. Meta reports a 3.1% final-transcription word error rate and first-place rankings on Artificial Analysis streaming speech-to-text and public diarization benchmarks. Accuracy can be improved through language, keyword, and context biasing. The model is available through the Meta Model API, Meta AI for Mac, and Muse Code.