Meta’s Muse Voice Transcribe enables real-time dictation on Mac

Meta launched Muse Voice Transcribe, described as its first real-time audio perception model, which brings streaming automatic speech recognition to Mac users. The model combines speaker diarization and endpointing, allowing it to transcribe speech as it happens, separate multiple speakers, and detect when a person has finished talking without a separate post-processing step. Users can trigger dictation system-wide by holding the Fn key, making it available across any Mac application.

Adoption for engineers involves using the Meta Model API to embed the transcription capability into software. This requires handling API calls, managing usage-based costs, and accounting for the model’s adaptive delay mechanism that varies listening time based on speech difficulty. Integration also means relying on Meta’s infrastructure for model updates and maintenance.

The model’s performance may be limited in noisy acoustic conditions or with strong accents, and only twenty-five of the seventy-plus languages it was trained on have been validated at launch, which could affect accuracy for unsupported languages. Additionally, the solution is presently tied to macOS and does not offer an offline or cross-platform option, restricting deployment to Apple’s desktop environment.

Because the announcement appears in a single feed, there is no independent corroboration of claims such as leaderboard rankings or language coverage. Engineers should seek third-party evaluations or wait for broader community feedback to validate the model’s real-world latency, accuracy, and suitability for specific voice-driven applications.