Why removing 'um' from a recording is harder than it sounds

A developer created "erm," a tool that automatically removes filler words (um, uh, er) from audio recordings by combining OpenAI's Whisper speech-to-text model with additional audio analysis techniques. The naive approach of simply transcribing and cutting filler words produces poor results with audible clicks and mismatches, so the tool uses multiple detection passes and audio refinement methods to identify missed fillers and create seamless edits.