ElevenLabs Announces Speech Recognition Model "Scribe" — a Voice Synthesis Company Enters Transcription
ElevenLabs announced its first speech recognition (ASR) model, “Scribe.” Having previously focused on voice synthesis, dubbing, and conversational agents, the company now moved seriously into the reverse direction — converting speech to text.
Details
- Language support: Supports transcription in 99 languages
- Accuracy: Reportedly achieved 96.7% transcription accuracy in English, which the company claimed was best-in-class at the time
- Additional features: Also supports word-level timestamps, speaker diarization, and tagging of audio events
- Availability: Offered as a standalone API that can be combined with the company’s existing voice synthesis and agent products
- Reaction at the time: Covered by several international outlets, including TechCrunch, as a voice synthesis company launching its own ASR model
What happened next
With the launch of Scribe, ElevenLabs became one of a small number of companies offering both voice “generation” and “recognition” through its own models. The company went on to enhance real-time performance with Scribe v2 and Scribe v2 Realtime, leading into an expanded 2026 product lineup that includes Dubbing v2 and Music v2.