Efficient Domain Adaptation of Whisper for Air Traffic Control Transcription and Contextual Error Rate Evaluation

Accurate speech-to-text transcription is critical for Air Traffic Control (ATC) to ensure safety and operational efficiency. However, existing Automatic Speech Recognition (ASR) systems struggle with domain-specific phraseology, high noise levels, and real-time constraints. This paper explores domain adaptation strategies of OpenAI’s Whisper ASR model, applying Transfer Learning and Low-Rank Adaptation (LoRA) on ATC benchmark and a Federal Aviation Administration (FAA) corpus. Experimental results show that fine-tuning significantly reduces Word Error Rate (WER) from 55.2% to 6.8%, yet WER alone fails to capture safety-critical transcription errors. To address this limitation, we introduce Contextual Error Rate (CER) - a novel metric that prioritizes errors with operational impact rather than surface-level differences. Our findings demonstrate that CER provides a more reliable assessment of ASR performance in safety-critical ATC applications, accurately flagging errors that could affect flight operations. This work highlights the need for context-aware evaluation metrics in ATC speech recognition. By combining fine-tuned ASR models with CER-based assessments, we enhance transcription reliability, paving the way for safer and more efficient speech-to-text automation in air traffic management.