Meta unveils Muse Voice Transcribe, real-time AI that separates speakers and languages
Meta launched Muse Voice Transcribe, a real-time audio model that can differentiate up to 20 speakers and transcribe multiple languages, including code-switching, with pricing of $3 per 1,000 minutes.
Meta Superintelligence Lab introduced Muse Voice Transcribe, the company’s first real-time audio perception model capable of speaker diarization and endpointing within a single system. The model can manage dictation for more than 20 speakers and simultaneously process multiple languages, even handling code-switching within sentences. Trained on over 70 languages, with 25 officially validated at launch, it adapts its listening delay to improve accuracy on difficult words.
Meta made the service available today through the Meta AI Mac app, Muse Code, and the Model API, charging $3 for each 1,000 minutes of audio. A demo is posted on Meta’s research blog, and the rollout comes less than a week after Google unveiled Gemini 3.5 Transcribe. While Google plans to embed its model in Android and Chrome, Meta has not disclosed any immediate integration into its flagship services.
Why it matters
The tool lets developers and users add accurate, multilingual live transcription to apps, expanding AI accessibility and competition.
In this story
