Designing a Real-Time ATC Transcription Tool to Reduce Pilot Cognitive Load and Readback Errors

Abstract

Radio communication between pilots and Air Traffic Control (ATC) requires pilots to distinguish whether the transmission is directed at them, memorize information quickly, and prepare accurate readbacks, raising cognitive load on pilots during critical phases of flight. Miscommunication errors are a significant problem in aviation human factors, and most automatic speech recognition (ASR) applications primarily focus on supporting controllers. Unlike prior work, this study developed an application that listens to ATC transmissions, transcribes them, matches messages to a selected callsign, and displays relevant information to pilots. The application presents matched clearances, suggests a readback, and itemizes extracted information such as altitude, heading, runway, frequency, route, and squawk command. The application saves previous transcriptions, supports aircraft-related information, a custom aviation language dictionary, and imported hints from flight plans. An ATC-fine-tuned Whisper Small English model converted to CTranslate2 int8 was used for transcription, and the application was tested on 150 radio recordings from the ATCO2 corpus. 40 out of 150 transcriptions were exact. On average, word error rate was 20.9%, processing time was 1.91 seconds (real-time factor 0.547), and 94.0% of recordings processed more quickly than their audio duration. This processing time outperformed the ~3-second industry baseline. Errors often involved unnecessary words, and some changed the callsign or command itself. The tool was more accurate for longer transmissions because word error rate decreased as audio length increased. By visually organizing extracted clearances and storing audio history, the system reduces cognitive workload by easing information overload and multitasking. Additional improvements in extraction of safety-critical information and testing on pilots’ workload are necessary before determining its usefulness.