I would begin with a clean recording and restrained processing, then compare at matched listening level. Removing all background sound can damage speech or make edits obvious. Keep a short room-tone recording and the original audio so cleanup decisions remain reversible.
On this page
Our recommendation
I prefer a little natural room sound to heavily damaged speech when the brief permits it. Improve capture conditions first, use targeted processing and keep the original available for a fair comparison.
Identify the noise and the speech
Listen to pauses, quiet syllables and word endings. Separate a steady background from intermittent clicks, handling noise or another voice. A single global noise-reduction setting may address one type poorly while damaging another part of the recording.
Preserve an unprocessed copy and choose a representative loop. Include both a quiet phrase and a louder one. Do not compare a louder processed version with a quieter original and call the louder result clearer; match the listening level as fairly as practical.
Review the transitions
Check for watery consonants, clipped word endings, changing room sound and sudden silence between phrases. Use room tone appropriately to support natural edits, without covering a meaningful sound or pretending that unrelated audio came from the same event.
Render a small proof through the final audio codec and listen on more than one relevant playback route. Compression and a small speaker can reveal different problems from the editor’s monitoring. Keep loudness adjustment separate from the judgement about cleanup artifacts.
A controlled cleanup pass
Loop a sentence ending in a soft consonant followed by a pause. Compare the original with a mild cleanup setting at similar loudness. If the consonant disappears or the room audibly swells when speech starts, reduce the processing or address the problem with a more targeted edit rather than increasing the suppression.
Before you act
- Use a loop containing quiet speech and pauses.
- Match comparison loudness.
- Listen to the actual encoded proof for artifacts.
Use the Audio file size calculator to apply the guide to your own setup.
Keep exploring
- Peak normalization vs loudness normalization
- Check loudness gain against the true-peak limit before rendering
Method note: this is an original planning procedure with illustrative inputs. It does not report a physical product test. Check the exact product, edition and official instructions before acting.
Build on what you just learned
Your next useful step.
Check channels, loudness and peak headroom.
Resolve 21 subframe audio edits: solve the click without moving the picture
Use fine audio timing when a frame-sized move is too coarse for a click, fade or transient.
Read the guide →
Generative Extend: a few extra frames still need an editorial review
I would use generated extensions for a deliberate transition or timing need, then inspect the new frames and identify them in the project record.
Read the guide →