Meta Unveils Muse Voice Transcribe: An AI Breakthrough in Speech-to-Text
The tech giant has introduced a groundbreaking artificial intelligence tool that has already secured the top spot in global indices, offering real-time transcription for up to 20 speakers and support for 70 languages.
Meta is introducing a new method for speech-to-text translation with the launch of Muse Voice Transcribe. The new model, developed at Meta Superintelligence Labs, provides real-time processing capabilities and is capable of identifying and separating over 20 different speakers during a single conversation. The technology supports 70 languages and breaks down linguistic barriers by switching naturally between languages (code-switching) even within the same sentence.
Beyond handling multi-user conversations, the model is suitable for working with long recordings exceeding one hour without the need for additional processing. The model automatically adjusts its transcription pace, determining when to produce text immediately and when to take a short "waiting beat" to better understand context and keywords. These performance metrics have already earned it first place in the Artificial Analysis index for real-time transcription and in other public tests.
The new tool is designed for diverse applications and is currently integrated into voice dictation within Meta AI and the Muse Code development environment. For instance, users of the Meta AI application on Mac computers can press the Fn key to speak freely instead of typing, with the text appearing rapidly on the screen. The model is currently available via the Meta Model API, the Mac application, and Muse Code.
Meta continues its expansion with plans for a major collaboration with the American company Anthropic. The joint deal, currently in early-stage negotiations, involves the supply of computing power for AI and is estimated at 10 billion dollars. This represents a significant move by Meta into a sector dominated by strong competitors such as Microsoft, Amazon, and Google.
