New Delhi — Meta has launched Muse Voice Transcribe, its first real-time audio perception model, offering native support for five major Indian languages along with multilingual transcription and advanced speaker recognition capabilities.
Developed by Meta Superintelligence Labs, Muse Voice Transcribe can provide streaming transcription, distinguish more than 20 speakers in recordings lasting over an hour, handle code-switching between languages and perform speaker diarization within a single model without requiring additional post-processing, the company said.
The model was trained on more than 70 languages spoken across multiple countries, with 25 languages validated at launch.
Meta said Muse Voice Transcribe ranked first on the Artificial Analysis streaming speech-to-text leaderboard as of Sept. 1, 2026.
The technology is available through the Meta Model API and is already being used for dictation in Meta AI for Mac and Muse Code. It offers real-time automatic speech recognition, speaker diarization for more than 20 participants and endpoint detection.
The model can also switch seamlessly between languages and improve transcription accuracy using language, keyword and contextual information.
One of its key features is what Meta calls “adaptive delay,” which adjusts how long the system waits before transcribing individual words based on their difficulty.
“The longer the model waits to predict, the more accurate the transcript, but the higher the latency. Muse Voice Transcribe has ‘adaptive delay,’ dynamically changing delay for each word based on difficulty,” Meta said.
The company said the capability is enabled through reinforcement learning, combining word error rate and delay measurements to balance transcription accuracy with speed.
“Muse Voice Transcribe is an autoregressive multimodal model from the Muse Spark family,” Meta said.
Audio is processed in 80-millisecond segments, or 12.5 chunks per second, with each segment transformed into a single soft token. At each stage, the model determines whether to continue listening to incoming audio or generate a text token.
Meta said the adaptive-delay approach allows Muse Voice Transcribe to achieve a strong balance between speed and accuracy when measured by the time required to produce a final transcription. (Source: IANS)
