Meta Muse Voice Transcribe Brings Real-Time AI Transcription for 70+ Languages with Multi-Speaker Detection

Meta Unveils First Real-Time Audio Perception Model

Meta Superintelligence Labs has launched Meta Muse Voice Transcribe, marking the company's entry into real-time audio transcription with an AI model that processes speech as it happens rather than after recording completion1

1

. The real-time transcription system combines live speech transcription, speaker separation, and endpointing within a single architecture, eliminating the need for separate post-processing steps2

2

. Mark Zuckerberg demonstrated the model's capabilities on X, showcasing its ability to automatically distinguish between multiple speakers while seamlessly switching between languages mid-conversation1

1

.

Advanced Speaker Detection and Multilingual AI Support

The real-time audio transcription model can identify and track more than 20 speakers in recordings longer than an hour, making it suitable for complex, multi-participant conversations2

2

. Meta trained the system on over 70 languages, with 25 undergoing extensive validation for the initial release, including Chinese, French, Hindi, Japanese, Spanish, and Vietnamese3

3

. The model provides native support for five major Indian languages—Hindi, Tamil, Telugu, Kannada, and Malayalam—expanding accessibility across diverse linguistic markets2

2

. Real-time speaker detection happens as audio streams in, allowing the system to identify when someone starts or stops talking without delay4

4

.

Adaptive Delay Technology Powers Processing Speed

Muse Voice Transcribe processes incoming audio in 80-millisecond chunks, using adaptive delay to determine when sufficient information exists to transcribe each word accurately1

1

. The AI model commits faster on simpler words while waiting slightly longer on complex phrases, balancing speed with accuracy through this intelligent processing approach4

4

. Meta employed reinforcement learning during training to optimize the trade-off between transcription accuracy and lower latency, helping the system make split-second decisions about when to finalize each transcribed token3

3

. This adaptive delay system allows the model to handle messy, real-world audio conditions while maintaining high accuracy across hour-long sessions1

1

.

Code-Switching Capabilities Set New Standard

The model excels at handling code-switching, where speakers mix multiple languages within a single sentence—a common occurrence in multilingual conversations1

1

. Users don't need to manually change language settings when speakers switch between languages, as the system automatically detects and adapts to linguistic shifts in real time2

2

. Meta demonstrated this capability by showing the model switching between English and Mandarin within the same sentence, maintaining accuracy throughout the transition3

3

. The system supports language, keyword, and context biasing, helping it identify words using information from both the audio stream and the broader conversational context2

2

.

Availability Through Meta Model API and Product Integration

Meta has made Muse Voice Transcribe available through the Meta Model API at $3 per 1,000 audio minutes, which translates to approximately $0.18 per hour of transcription2

2

. The model is immediately rolling out across several Meta products, including voice dictation in Meta AI, Muse Code, and Meta AI for Mac4

4

. Because the Meta AI for Mac app can power voice-enabled features across other applications, Muse Voice Transcribe will now handle dictation features on third-party services1

1

. Developers can access the model to integrate speech transcription capabilities into their own applications, with a demo version available on Meta's research blog1

1

.

Strategic Positioning Against Competition

Meta's release follows Google's launch of Gemini 3.5 Transcribe by less than a week, intensifying competition in the real-time audio transcription space1

1

. While Google integrates its audio model into Android and Chrome, Meta's integration strategy remains focused on its own ecosystem and developer access through APIs1

1

. Meta positions Muse Voice Transcribe as a building block for personalized AI assistants capable of understanding accents, interruptions, overlapping speech, and multilingual conversations—capabilities increasingly demanded in modern voice interfaces3

3

. The model represents the latest release from Meta Superintelligence Labs, which has recently introduced its first dedicated coding agent, an open-weight model, and the Meta AI Mac app1

1

. As real-time speech systems evolve beyond basic transcription, speaker identification and faster response times are becoming critical differentiators in AI-powered voice tools3

3

.