Skip to content

Scribe v2 Medical is now available to everyone

ElevenLabs

The clinical speech recognition model is generally available on the batch Speech to Text API using the model identifier scribe_v2_medical.

Key points

  • Achieved a 7.0% word error rate on the Eka Medical ASR benchmark, reducing WER by 18% compared to base Scribe v2
  • Reached 5.88% WER and 86.2% dosage accuracy on the Omi Health Medical Speech to Text benchmark
  • Maintains standard transcription quality with an unchanged 5.3% WER on CommonVoice non-medical speech samples
  • Supports HIPAA compliance for enterprise customers with Business Associate Agreements and Zero Retention Mode enabled
  • Supports keyterm prompting to reduce transcription errors on isolated drug and condition names

Why it matters

Healthcare developers can transcribe clinical conversations and medical dictation with higher accuracy on drug names and dosages without degrading standard speech transcription.

Sources