Meta releases speech recognition that transcribes a 20-person conversation with almost no delay
Prepared by Sintelligent with the help of SI. Publication time restored from the build log. . Updated: .
Live transcription of multilingual meetings is becoming instant and accurate, and language barriers in conversation keep melting away.
Muse Voice Transcribe delivers the final text in about 0.16 seconds, distinguishes more than 20 speakers, is trained on 70+ languages (25 of them carefully validated) and understands language switching within a sentence. On Artificial Analysis rankings it took first place in streaming recognition (3.1% word error rate) and in speaker diarization. It is available through the Meta Model API, Meta AI for Mac and Muse Code.
Sources 1