Meta Superintelligence Labs releases Muse Voice Transcribe, a real-time model doing streaming ASR, diarization, and endpointing at 3.1% WER.