Meta's new real-time audio model is the foundation for AI assistants that never stop listening
decoder ·
Meta's Superintelligence Labs released Muse Voice Transcribe, a real-time audio model that handles speech recognition, speaker separation for up to 20 people, and sentence boundary detection in a single system. It processes audio in 80-millisecond chunks and uses reinforcement learning to dynamically decide how long to listen before transcribing each word, balancing speed and accuracy. The model supports over 70 languages and can process recordings longer than an hour without post-processing. Independent testing by Artificial Analysis reported a 3.1 percent word error rate on English, with results arriving about 0.16 seconds after a speaker finishes. Meta priced the model at $0.18 per hour, undercutting competitors such as Cartesia, ElevenLabs, and Deepgram. Muse Voice Transcribe powers voice dictation in Meta AI and Muse Code and is available through the Meta Model API. Meta framed the release as a step toward personal superintelligence and AI glasses, but it did not disclose technical details or release model weights.
decoder