FURI | Spring 2026

Enhancing ASR Models for Reliable EMS Radio Transmissions

The objective of this research is to determine if fine-tuned Automatic Speech Recognition (ASR) models can accurately transcribe noisy Emergency Medical Services (EMS) radio communications. By training models with domain-specific data and simulated noise, results indicate a significant reduction in word error rates. This enhanced accuracy streamlines hospital preparedness and reduces critical medical errors, ultimately improving patient outcomes during high-stakes transfers. Future work should focus on developing a comprehensive pipeline to predict patient outcomes directly from transcribed EMS communications.