Test of 12 open speech models on 208 drug names finds best gets only 169 right
Maziyar Panahi tested 12 open-source speech transcription models on 208 drug names from simulated GP consultations; the best model, Nemotron 3.5, got only 169 right, revealing large gaps in medical-domain reliability.
2026-11-17 ~ 2026-11-17 · 3 related posts
- 12 open speech models tested on 208 drug names: even the best gets 27 wrong — MaziyarPanahi · 2026-11-17
2 near-duplicate retellings: MaziyarPanahi · MaziyarPanahi