Guide
How to clean a voiceover: noise, filler, breaths, and loudness
People ask “how to clean a voiceover,” “remove background noise from VO,” “de-um audio,” “level a voice track” — not “how to clean voice in [a brand].” Cleaning is several jobs that get sold as one slider: broadband noise, mouth clicks, filler words, dead air, and loudness. The right method depends on which of those you actually have. Re-recording in a quieter room beats any plugin. Dedicated audio tools such as Cleanvoice sit next to NLE mixers and speech pipelines. Tolik voice-only is a speech cleanup pass when the picture is already done or there never was picture. It is not iZotope RX as a career.
Name the problem before you pick a tool
Fridge hum and computer fans are noise. “Um” is filler (remove filler on camera if there is also a face). A two-second hole is silence (remove silence). A sentence 8 dB quieter than the one before it is a level problem. Mouth clicks are a different pass. If you run a noise reducer on filler, you still have filler. If you gate on loudness, you chew the ends of words.
When you should re-record instead
Clipping, a rival conversation in the room, a mic in a bathroom, a take where every clause is a stumble. Cleanup recovers a usable web VO. It does not recover a performance you did not give. Move the mic, turn off the fridge, record 10 seconds of room tone.
Noise reduction, honestly
Spectral tools (RX and cousins) let you paint out a fridge. Consumer denoisers guess and leave underwater voice if you push. Room tone is not the enemy — a dead-silent hole after a denoise sounds fake. Keep a bed of the room under the VO when you can.
Filler and silence on audio-only
Transcript delete (Descript) or a speech cleaner that drops ums and pauses. Then listen: which breath was character. Two-person mixes are harder; bounce stems if you can. vs Cleanvoice if audio is the only product you will ever buy.
Loudness for the web
YouTube and many podcasts sit near the platform’s loudness target; creators often mix around −14 LUFS integrated as a web starting point, then check on phone speakers. Broadcast delivery specs are a different contract — measure them if someone wrote them into a brief. A “leveler” that crushes every sentence will make you sound like a radio ad. After any automatic pass, listen on earbuds.
Where a speech pipeline fits
Picture locked in Premiere, VO still embarrassing: export audio, clean, lay back. Or never leave the NLE. If you still need a talking-head MP4, do not use an audio-only pass and hope a camera appears — use talking head. If the next step is B-roll you have not cut, montage in one run beats ping-pong. Tokens in Tolik pay for the run; download does not — pricing.
| Audio-only cleanup | Use a video path instead |
|---|---|
| Wav/m4a is the deliverable | Need jump cuts of a face |
| Picture locked elsewhere | Need B-roll laid to lines |
| Noise, ums, level | Need generated picture |
Questions.
How do I remove background noise from a voiceover?
Prefer a quieter record. Then a spectral denoise or a dedicated cleaner. Do not confuse noise with ums.
How do I de-um audio without a video editor?
A transcript editor or a speech cleaner. Then restore any um that was a beat.
Should I remove breaths?
Usually no, not all of them. Dead-silent VO sounds processed. Thin the worst mouths, keep the rest.
Is a noise gate enough?
Rarely. Gates chew word endings and miss voiced filler.
What loudness should YouTube VO be?
Match the platform; many web mixes start near −14 LUFS and are checked on phones. If a client specified a number, measure that number.
Cleanvoice vs Descript vs a video pipeline?
Cleanvoice: audio specialist. Descript: transcript as the room. Video pipeline: cleanup on the way to an MP4. Pick the room you live in.
Two speakers on one track?
Harder. Cleanup will treat the mix as one voice. Stems are better.
Will cleanup replace a mastering engineer?
No. It is a first clean for web VO.
Picture already finished?
Clean the VO and lay it back, or stay in the NLE.
Do I need captions on audio-only?
Not burned into a video. For a video file, caption after the cut.