Meta has made Muse Voice Transcribe available through the Meta Model API, so developers can use the model for speech transcription.
Muse Voice Transcribe handles recordings longer than an hour
Photo Credit: Meta
Meta has launched Muse Voice Transcribe with native support for five major Indian languages, including Hindi, Tamil, Telugu, Kannada and Malayalam, as part of support for more than 70 languages globally. The model is the first real-time audio perception model developed by Meta Superintelligence Labs. It combines live speech transcription, speaker separation and endpointing in a single system, while also handling code-switching and conversations with more than 20 speakers.
Muse Voice Transcribe can turn speech into text as someone is speaking, rather than processing the entire recording afterwards, the company explained in a blog post. It can also identify individual speakers and detect when someone starts or stops talking. Transcription, speaker separation and endpointing happen as the audio comes in, without a separate post-processing step.
It can distinguish more than 20 speakers in a recording and work with audio longer than an hour, according to Meta. This allows it to handle long conversations without limiting the transcription to shorter clips.
The model processes audio in 80-millisecond chunks and determines when it has heard enough information to transcribe each word. It can wait longer for words that are harder to recognise and commit simpler words sooner. Meta says reinforcement learning helps the model decide these delays while keeping transcription errors low.
Meta trained the model on more than 70 languages and extensively validated 25 of them for the initial release. Its language support includes five major Indian languages, including Hindi, Tamil, Telugu, Kannada and Malayalam, the company added in a press release.
It can also transcribe conversations in which speakers switch between languages, including when the switch happens within a sentence. Users do not have to change the language manually each time someone switches languages. The model supports language, keyword and context biasing, which helps it identify words using information from the audio and the wider conversation.
Meta has made Muse Voice Transcribe available through the Meta Model API, so developers can use the model for speech transcription in their own applications. Meta AI for Mac and Muse Code already use it for dictation. The API is priced at $3 (roughly Rs. 300) per 1,000 audio minutes, which works out to around $0.18 (roughly Rs. 17) per hour.
Get your daily dose of tech news, reviews, and insights, in under 80 characters on Gadgets 360 Turbo. Connect with fellow tech lovers on our Forum. Follow us on X, Facebook, WhatsApp, Threads and Google News for instant updates. Catch all the action on our YouTube channel.