Muse Voice Transcribe in LiveKit Agents

Meta's real-time speech-to-text model is now a ready-made plugin in LiveKit Agents, an open-source framework developers use to build voice agents.

Install the livekit-agents[meta] package in Python or @livekit/agents-plugin-meta in Node.js. The default model is muse-voice-transcribe-1.0
Streams live audio and returns running transcripts, and detects on its own when a speaker has finished talking
A keywords setting steers spelling for names and industry vocabulary, and a language setting can favor one of 25 supported languages per stream, including English, Spanish, Hindi, Japanese, and Mandarin Chinese
The plugin does not support telling speakers apart, transcribing recorded files, or changing keywords mid-stream, and each live session caps at 60 minutes
Requires a Meta Model API key. Neither Meta nor LiveKit lists pricing for it

Voice agents that need to turn a caller's speech into text as they talk, including names and brand terms.

You already build voice agents on LiveKit and want Meta's transcription model without writing your own connection to it.

Developers building phone agents, customer-facing voice assistants, or live support tools.

Join the newsletter to get ahead with AI.

I share strategies, new modules, free guides, and tools to get you better at using AI in your work.