Meta Releases Muse Voice Transcribe Real-Time Speech Transcription Model
Meta released Muse Voice Transcribe, which can perform real-time streaming speech recognition, speaker separation for more than 20 speakers, and endpoint detection, and supports more than 25 languages and code-switching.
Meta released Muse Voice Transcribe, calling it the first real-time audio perception model developed by Meta Superintelligence Labs.The model can simultaneously perform streaming automatic speech recognition, speaker separation for more than 20 speakers, and endpoint detection. It natively supports more than 25 languages, intra-sentence code switching, and language, keyword, and context bias.
Muse Voice Transcribe is now available through the Meta Model API, Meta AI for Mac, and Muse Code. The one-click voice dictation feature in Meta AI and Muse Code has also switched to this model. The official page does not claim to have open-source model weights, therefore this product should not be described as an open-source release.
refer to:Meta AI Research Release Page