Accuracy & Metrics
Relaya's Scribe achieves clinical-grade accuracy across Indian medical specialities. Here's how we measure it.
Evaluation Methodology
- Evaluated against 10,000+ manually transcribed Indian clinical consultations
- Blind evaluation by licensed physicians (not model developers)
- Speciality-stratified sampling across general practice, cardiology, orthopedics, pediatrics, dermatology, and 7 others
- Tested across 15 Indian English accent regions and Hindi dialects
- Monthly re-evaluation with fresh clinical recordings (never reused test sets)
- Comparison benchmarks against Google Medical ASR and Whisper Large v3
Continuous Improvement
Every week, our clinical evaluation team reviews a random sample of generated notes against physician gold-standard annotations. Error patterns are categorised, prioritised, and fed into our model improvement pipeline.
Accuracy improvements ship via model updates that require zero action from users. When we improve Hindi code-switching accuracy by 2%, every Relaya user benefits immediately. no app update needed.
We maintain a public accuracy dashboard (available to all customers in their settings) showing real-time model performance metrics for their specific speciality and language mix.
See the accuracy yourself
Start a free trial and test Relaya with your own clinical conversations.
Start free trial