Release2026-09-22
ElevenLabs released Scribe v2 Medical, a speech-to-text model fine-tuned for clinical audio. It cuts word error rates on clinical audio by about 35% versus base Scribe v2. It leads on MedDictate, MedTerm, and Eka Medical ASR benchmarks. It maintains general speech accuracy at 5.3% WER. It is HIPAA-eligible with Zero Retention Mode for enterprise customers. Customers report higher accuracy on clinical terminology in production. It is available via the ElevenLabs Speech to Text API with model ID `scribe_v2_medical`.
What changed
Scribe v2 Medical was in limited or beta availability.
What it unlocks
Transcribing clinical audio with lower error rates on drug names, dosages, and medical terms.
- 35% lower WER on clinical audio vs base Scribe v2
- 4.9% overall WER on MedDictate
- 77.7% term recall on MedTerm
- 6.50% semWER on Eka Medical ASR
What you need to act on it
- an ElevenLabs API key
- HIPAA eligibility requires a Business Associate Agreement and Zero Retention Mode
Sources