Meta launches its first real-time audio AI model
What's the story
Meta has launched Muse Voice Transcribe, a cutting-edge model for real-time audio perception. The innovative system brings multilingual and streaming transcription capabilities to Meta AI for Mac, Muse Code, and developers via the Meta Model API. The new tool enables system-wide dictation on Mac devices, making it easier than ever for users to convert spoken language into written text in real time.
Tech specs
Model can transcribe speech in real time
Muse Voice Transcribe combines streaming automatic speech recognition with speaker diarization and endpointing.
This means it can transcribe speech in real time, identify different speakers in recordings with over 20 voices, and detect when a person has stopped speaking - all without any post-processing.
The model has been trained on more than 70 languages, with support for audio longer than an hour and native code-switching within or between sentences.
Intelligent processing
Adaptive delay system decides how long to listen before committing
Muse Voice Transcribe also features an "adaptive delay" system, which decides how long to listen before committing each word.
This way, it can quickly process easier speech while using more audio context for difficult words.
The model has already been integrated into Meta AI for Mac and Muse Code, allowing users to dictate into any application by simply holding the Fn key on their Mac devices.
Market impact
Muse Voice Transcribe ranks 1st on Artificial Analysis leaderboard
As of September 1, Muse Voice Transcribe ranks first on the Artificial Analysis streaming speech-to-text leaderboard.
The model is available through the Meta Model API at $3 per 1,000 audio-minutes ($0.18/hour).
This competitive pricing makes it an attractive option for developers looking to integrate advanced voice dictation capabilities into their applications or services.