AI Achieves Record Accuracy in Medical Speech

Corti, a Copenhagen-based artificial intelligence company, has introduced Symphony for Speech-to-Text, a new generation of clinical-grade speech recognition models. These models are specifically engineered for real-time dictation and audio processing, achieving the highest recorded accuracy for this use case. In recent evaluations, Corti's new models demonstrated up to a 93% reduction in word error rates compared to leading generalist speech models on medical terminology. Its performance resulted in a 1.4% word error rate on English medical terms, significantly outperforming competitors. This advancement is crucial as healthcare moves into an "agentic era" where AI agents actively assist in clinical decision-making, making accurate data inputs paramount. In contrast to general-purpose AI, Corti's specialized models are designed to handle medical acronyms, complex dosages, and noisy environments, mitigating the risks associated with transcription errors. The company also reported an exceptional 98.3% recall rate on formatted clinical entities. Corti's technology also aims to surpass legacy systems like Dragon Medical One, demonstrating superior accuracy in real-world medical dictation and medical term recall. The company is making these models available via an API, enabling developers and vendors to build advanced healthcare tools. Corti's models have also shown strong performance in multilingual settings, achieving low word error rates in German and French, proving their viability in diverse international markets. This launch is part of Corti's broader strategy to demonstrate the strength of vertical AI specialization against horizontal tech giants, showcasing gains in medical coding, clinical reasoning, and speech-to-text accuracy. The company is observing a significant increase in platform sign-ups as developers opt for clinical-grade models over generalist APIs.