Clean up recorded speech with AI
Reduce room noise and electrical hum, even out the level and compare the finished recording with the source.
Clean up recorded speech with AI when a good take has a steady buzz beneath the voice. The opening silence may sound harmless on laptop speakers, then become distracting in headphones. With audio-cleanup the agent measures the recording, applies filters for the selected preset and checks the result against its loudness and peak targets.
You can run the analysis before making a new file. On a video, the selected audio track is processed while the first video stream is copied. This suits a single-speaker lesson or interview where the picture is already right.
What it does
- Reduce a consistent mains hum under a spoken recording.
- Try a light pass on a voice that already sounds close.
- Compare the loudness and peaks before and after processing.
- Lift a quiet podcast voice and listen for any remaining hiss.
- Keep the video picture while producing a new version of its soundtrack.
Install and use
- Install the skill, Python and FFmpeg locally. Supply a recording and choose the intended audio track if there is more than one.
- Describe the problem and the destination. Ask for an analysis first or select a suitable voice preset.
- Listen to a quiet passage and a louder phrase in the new file. Keep the version you prefer under a distinct filename.
FAQ
Source
Made by Dave Saunders (nemock). Original skill: audio-cleanup, published on GitHub under the MIT licence.
View on GitHub