Corti’s Symphony for Speech-to-Text Shatters OpenAI Accuracy in Medical Dictation

Symphony for Speech-to-Text Sets New Accuracy Benchmark

Corti, the Copenhagen‑based health‑AI firm, unveiled its latest clinical speech model today, promising the lowest word‑error rate ever recorded for medical dictation. In a Reuters interview, CEO Andreas Cleve emphasized that the system must be trustworthy for physicians, patients and the broader health ecosystem.

Head‑to‑head numbers

On English medical terminology the model logged a 1.4% WER, while OpenAI’s competing offering posted 17.7% WER. ElevenLabs and Whisper lingered above 17% . The gap translates into a ↑ 93% reduction in errors versus generic APIs.

“If a model mishears a dosage, every downstream decision is compromised,” Cleve told VentureBeat.

Entity recall reached 98.3% for dosages, dates and measurements, dwarfing the best generic baseline at 44.3%. For developers, that difference can mean the line between a time‑saving tool and a liability.

Beyond English

German tests delivered a 2.4% WER versus the runner‑up’s 13.0% . In French the model achieved 3.9% WER compared with 10.6% elsewhere, proving its multilingual mettle in Swiss hospitals.

Quarter‑over‑quarter sign‑ups have risen ↑ 30%, indicating strong developer appetite for vertical AI solutions. Corti positions the API as a plug‑in for EHR vendors, telehealth platforms and any workflow that needs real‑time, error‑free clinical text.

Availability starts today via the Corti API console, with full documentation and an open research paper outlining the methodology.

Intel provided by Dr. Aris Thorne (Artificial Intelligence Researcher).