Artificial intelligence (AI) is rapidly improving voice transcription by making it faster and more accurate. Tech giant Meta on Tuesday, September 1, announced a new speech-to-text model that can transcribe conversations as they happen. Named Muse Voice Transcribe, the model can handle multiple languages within the same conversation. This is also the first real-time audio perception model released by Meta Superintelligence Labs (MSL).

Meta said that the model supports over 70 languages, including five major Indian languages like Hindi, Tamil, Telugu, Malayalam, and Kannada. The company claims that Muse Voice Transcribe is designed to handle code-switching, where speakers move between languages during a conversation, without needing separate models or additional processing.

In India, users are prone to switch between English and regional languages. Muse Voice Transcribe can come in handy and could potentially make real-time transcription more useful in the Indian context. The model produces streaming transcription, meaning text will be generated as a person speaks instead of waiting for the entire recording to be processed. It is also capable of separating speakers in recordings involving more than 20 voices, and can handle recordings longer than an hour, according to Meta.

Story continues below this ad

Meta said that these capabilities are managed by a single model without a separate post-processing step. Muse Voice Transcribe has been trained across over 70 languages, with 25 languages validated at launch. The company also claimed that the model ranked first on the Artificial Analysis streaming speech-to-text leaderboard as of September 1, 2026.

One of its distinct features is how it strikes a balance between speed and accuracy. Instead of using a fixed setting for how long it listens before producing each word, the model ascertains this on a word-by-word basis, allowing it to respond quickly to straightforward speech while spending more time on words that are harder to recognise.

Muse Voice Transcribe is available through Meta’s Model API starting today. Meta said the model is already being used for dictation in Meta AI for Mac and Muse Code. The API is priced at $3 per 1,000 audio minutes, which, according to Meta, is equivalent to about $0.18 per hour.

The launch comes at a time when speech-based AI systems are becoming increasingly important for applications such as transcription, dictation, coding and voice assistants. Supporting code-switching and a wider range of regional languages can also help such systems better comprehend how people communicate in multilingual markets such as India. Meta has also published additional technical information about Muse Voice Transcribe in a research blog accompanying the launch.