Meta Superintelligence Labs has launched Muse Voice Transcribe, an autoregressive audio perception model that unifies speech recognition, speaker diarization for over 20 speakers, and endpoint detection into a single pass. Designed to eliminate the latency and complexity of multi-model audio pipelines, the system processes audio inputs in 80ms chunks without requiring secondary post-processing steps.
The model is available exclusively as a hosted API on the Meta Model API under the identifier ‘muse-voice-transcribe-1.0’, priced at $3.00 per 1,000 audio minutes ($0.18 per hour). Meta confirmed that the architecture already powers voice features across internal products, including dictation in Meta AI for Mac and Muse Code. No weights have been released, precluding open-source or self-hosted deployments.
By consolidating transcription, speaker separation, and end-of-turn detection into one continuous stream, Meta offers developers a streamlined building block for conversational applications. The release signals a shift toward integrated multimodal architectures that replace complex, stitched-together speech processing systems.
Why it matters
Developers can reduce voice application latency and operational overhead by consolidating multi-stage speech pipelines into a single API endpoint.
Meta’s decision to withhold model weights locks enterprise deployments into hosted infrastructure for real-time audio workflows.
Source: marktechpost.com
