Meta rolls out Muse Voice AI to all smart glasses, showcases low latency

Meta just gave every pair of its smart glasses a significant upgrade. The company’s Muse Voice Transcribe model, introduced on September 1, is now available across its entire smart glasses lineup, including Ray-Ban Meta and Oakley Meta devices, bringing real-time voice AI that processes audio faster than most people can blink. The headline number: 0.16 seconds of final-transcript latency after someone finishes speaking. For context, the average human blink takes about 0.3 seconds. So this thing transcribes what you said before you finish blinking at the person you said it to. Under the hood Muse Voice Transcribe works by chopping incoming audio into 80-millisecond chunks and processing them in near real-time. That granular approach delivers a 3.1% word error rate on English benchmarks, which puts it in competitive territory with the best speech-to-text systems available today. The model handles speaker diarization for more than 20 speakers simultaneously, and it does so without requiring separate post-processing. Language support spans more than 70 languages. For developers looking to build on top of it, Meta is offering the model through its Model API at $3 per 1,000 audio minutes...