Meta Unveils 'Muse Voice Transcribe' Real-Time AI Speech Model Supporting 5 Indian Languages
Tech giant Meta, through its Superintelligence Labs, has launched a new artificial intelligence model named 'Muse Voice Transcribe' designed for real-time speech-to-text processing. Unveiled on September 1, 2026, the model marks Meta's first specialized real-time audio perception system capable of instantly transcribing live audio, identifying distinct speakers, and detecting conversation endpoints without requiring post-processing steps.
The model supports over 70 global languages, with 25 validated at launch—including five major Indian languages: Kannada, Hindi, Tamil, Telugu, and Malayalam. A key feature of the model is its native support for code-switching, which allows it to track and transcribe conversations when speakers alternate between English and regional Indian languages within a single sentence or discussion. Furthermore, the system can distinguish between more than 20 distinct speakers in real-time streaming audio and handle recordings lasting over an hour.
To balance processing speed and accuracy, Meta incorporated an adaptive delay mechanism trained via reinforcement learning. This allows the model to process standard words rapidly while holding complex terms briefly for additional context before rendering text. The model has been made available to software developers via the Meta Model API at a rate of $0.18 per hour of audio, and is also integrated into Meta AI desktop applications. The rollout promises to enhance automated meeting summaries, live captioning, and enterprise voice assistants across diverse multilingual regions.
