Meta's Muse Voice Transcribe: $0.18/Hour, 20+ Speakers

Meta Superintelligence Labs put a price tag on its new transcription model this week: $0.18 per hour of processed audio for Muse Voice Transcribe, a real-time speech-to-text system built to handle streaming transcription, endpoint detection and speaker diarization in one pass. No separate post-processing pipeline required.

The pitch is capacity plus speed. Muse can track more than 20 speakers as they talk, not after the recording ends, and Meta's launch materials claim support for audio sessions running past an hour, multilingual code-switching mid-sentence, and biasing toward specific languages or keywords. Training spanned more than 70 languages, with 25 getting heavier validation ahead of the initial release.

That 20-plus-speaker number sounds big until you check what competitors already publish. Speechmatics lists 50 speakers by default, scalable to 100. Amazon Transcribe caps diarization at 30, streaming included.

So Muse isn't setting a ceiling record. What it's doing instead is bundling a lot of separate capabilities, high speaker counts, low latency, code-switching, cheap API access, into a single model aimed squarely at enterprise builders working on meetings, call centers and live assistants.