Paul Mestemaker
Published on

How to: clean up audio quality in a video



  1. Extract audio to mp3 using ffmpeg.

  2. Clean audio with machine learning (e.g. something like or if you want to get your hands dirty

  3. Replace audio in original video with ffmpeg.


I've been on an AI/ML kick recently and have been trying out a ton of new tech and products. I recently discovered Modal, which seems like AWS Lambda for machine learning (but super fast config/deploy and great developer experience). I signed up for their service and was greeted by an action-packed Loom that sounded totally blown out. I looked up the author on Twitter and saw this tweet:

So I figured I could put some goodwill into the community and help their onboarding experience by making the video sound a bit better.

Step 1: download the video from Loom

There's a button in the UI. Just click it. :)

Step 2: Extract audio from video with ffmpeg

$ ffmpeg -i loom_original.mp4 blown_out_audio.mp3

If command lines aren't your thing, you can also use something like Audio Extractor.

Step 3: Clean up audio

Drag and drop blown_out_audio.mp3 onto and wait about 3 minutes. This auto-saves as blown_out_audio (enhanced).wav.

If you don't want to use a 3rd party service, you could also use an ML model like this one: and set up your own pipeline.

Step 4: Replace audio in original video

$ ffmpeg -i loom_original.mp4 -i "blown_out_audio (enhanced).wav" -c:v copy -map 0:v:0 -map 1:a:0 new_hotness.mp4


We are in a golden age of machine learning. There are going to be thousands of services launched in the next year to help with image/text/audio/video generation/manipulation/editing/enhancement. Given the amount of open source code and models that exist, these services will likely be completely free addons to existing products or incredibly cheap & accessible products.


  1. A Complete Guide to Speech Enhancement