A method of Eliminating FSK Noise in cockpit voice recordings

This paper presents a novel method to eliminate Frequency Shift Keying (FSK) noise from cockpit voice recordings (CVRs) using multi-channel similarity. FSK signals, which encode time information, often introduce noise into CVR audio, interfering with voice recognition. Traditional filtering methods degrade audio quality by removing valuable signal components. Our method detects and replaces FSK noise segments by leveraging the similarity between CVR channels, preserving the original audio integrity. Automated processing enables rapid and batch handling of recordings. Tested on a 3.5-hour CVR recording, the method significantly reduced FSK noise, improving auditory comfort with minimal impact on the original content.