← Retour au blog
tech 12 June 2026

Removing 'um' from a Recording: An Underestimated Challenge

Removing 'um', 'uh', and other filler words from an audio recording is more complex than it seems. Discover why and how a local CLI can transform this tedious task.

Article inspired by the original source
Removing 'um' from a recording is harder than it sounds ↗ doug.sh

Introduction

In today's digital world, audio quality plays a crucial role, whether for podcasts, online videos, or conferences. Disfluencies, those little filler words like 'um', 'uh', can quickly degrade the listening experience. Yet, removing them from a recording is more complex than it seems.

The Challenge of Disfluencies

Linguists call these small interruptions disfluencies. They often serve to fill a void during a pause or hesitation. However, in a professional or public context, they can give an impression of improvisation or lack of preparation.

Manually removing these disfluencies from a recording is a tedious task. Traditional audio editing tools require spotting each occurrence and carefully cutting each segment, which can take hours for a one-hour recording.

The Technological Solution: A Local CLI

Doug, a developer passionate about efficiency, created a tool named 'erm'. This command-line interface (CLI) is designed to automate the process of cleaning audio recordings by removing disfluencies.

How Does It Work?

The tool relies on Whisper, an OpenAI speech-to-text transcription model. Whisper can generate transcripts with precise timestamps for each word, allowing for targeted disfluency removal.

However, Whisper has its limitations. It tends to omit disfluencies during transcription, complicating their detection. To overcome this issue, the 'erm' tool performs multiple passes on the audio to identify missed disfluencies, particularly by analyzing unusual pauses in the speech.

Technical Challenges

  1. Incomplete Transcription: Whisper may ignore disfluencies, leading to omissions in the transcript.
  2. Audible Cuts: Cutting audio at arbitrary points can create clicks or perceptible changes in background sound.
  3. Sound Level Matching: Changes in background noise between cut segments can be detected by the human ear, requiring subtle adjustments.

To solve these issues, 'erm' adjusts audio segments to eliminate clicks and harmonizes background noise across cuts.

The Impact of 'erm' on Productivity

By automating the cleaning process, 'erm' saves a considerable amount of time. Content creators can focus on producing quality material instead of spending hours on post-production. The speed and efficiency of this tool can transform a production team's workflow.

Conclusion

Removing 'um', 'uh', and other disfluencies is essential for producing professional and engaging audio. Thanks to tools like 'erm', this process becomes not only faster but also more precise.

Let's discuss your project in 15 minutes.

disfluencies audio editing Whisper CLI tool productivity
Deepthix newsletter · 100% AI · every Monday 8am

An AI agent reads tech for you.

Our AI agent scans ~200 sources per week and ships the best articles to your inbox Monday 8am. Free. One click to unsubscribe.

Visit the newsletter page →

Want to automate your operations?

Let's talk about your project in 15 minutes.

Book a call