Meta launches Muse Voice Transcribe, SOTA streaming speech-to-text

Meta Superintelligence Labs released Muse Voice Transcribe, its first streaming speech-to-text model, claiming SOTA performance and native speaker diarization and endpointing. It tops AA-WER Streaming with 3.1% WER at 0.16s after end of speech, trained on 70+ languages, and is available on the Meta models API.