Noise removal
Removing what should not be there without removing what should. The second half is the hard part.
The short version
What you get. A clearer recording that preserves the voice’s character, with limits on repair explained early.
Is this for you? For hum, hiss or other unwanted sound that makes a recording harder to follow. We assess whether repair, a setup change or a retake is the better answer. Skip it if every trace of the room has to go; stripped that far, a voice sounds thin.
What leaves your plate. The room kept on purpose, the tools used only to find the noise while the decision to cut it stays a human one, and the call when a recording is better re-recorded than rescued.
Start with the Free Audit.
How much to take out
We leave the room in
The easiest way to spot rushed noise removal is that the voice has nowhere to stand. Strip a recording of its room and you get something clean, thin and slightly dead, and a listener will feel the wrongness without being able to name it. A room is part of a voice. The work is to take out the hum, the hiss and the rumble while keeping the space the speaking happened in, and where a recording has already lost too much of it, to put some back.
The tools analyse; the decisions stay ours
We use automated and AI tools where they help, then compare the result with the original by ear. The decision is how much unwanted sound can be reduced before the voice loses body or clarity. Cleaner is not always better.
A spare microphone that records the room
A spare room recording can help us understand the unwanted sound and preserve a sense of space. It cannot guarantee exact removal, because each microphone captures a different balance.
The unstable noise is the one that matters
Steady noise is the tractable kind, because it sits still: a hum, a fan, a fridge, an electrical buzz keeps one shape, so a profile measured during a pause can be applied across the whole take. Even then the subtraction is an estimate rather than a clean lift, judged by ear and stopped early, because where the noise shares frequencies with the voice, reducing one takes a little of the other. It stops being routine altogether when the noise is nearly as loud as the voice, or its harmonics sit right inside the speech, and then it joins the harder kind. Noise that moves is that harder kind from the start: the newer adaptive and machine-learning tools reach some of it now, further than the old profile-based ones could, though not reliably enough to lean on, so we automate the steady, treat the moving noise by hand where the tools stop, and tell you how to stop it happening next time.
Where we begin, before the voice changes it
| Distance, non-shotgun | 15 to 30 cm |
|---|---|
| Angle | slightly off centre, out of the plosive path |
| Room microphone | one spare input, pointed away from the source |
| Steady noise | hum, fan, mains buzz, RFI: automated |
| Moving noise | treated by hand, or re-recorded |
Where the noise comes from, and where it gets caught
The usual sources
Even a properly insulated and acoustically treated room gives you something: low rumble through the building, hiss from the gain stage, radio interference, or a mains hum from a grounding fault nobody knew about. Air conditioning, a computer fan and a light fitting all have a signature, and every one of them is audible the moment the voice stops.
It begins with the recording, which is why we talk before the shoot
Almost none of this sounds like noise removal, and all of it is noise removal. A speaking level that is right and stays right. Real rests in the speech instead of run-on takes. A retake when the articulation wobbled, because an unstable delivery is far more expensive to repair than to record again. And a considered decision about where in the room a person stands.
Microphone placement decides how much can be saved
Placement and recording level affect plosives, sibilance, clipping and the balance of direct sound to reflections. Some faults can be reduced afterwards, but repair has limits. We listen to a test before committing the full shoot.
At the bench
The work, demonstrated
Try a decision from the craft: drag, tap or toggle, then read the trade-off.
These are illustrative models; no partner’s results are shown.
More demonstrations at the bench →A room is part of a voice. The job is knowing what to leave.
What changes for you
A clearer recording, with unwanted sound reduced where the material allows. If a take is better recorded again, we say so before building the edit around it.
Related: Audio mixing · Multi-microphone setups · Microphone setup · Mastering for platform
Send recent videos. We will listen for the main audio problems and suggest what to address first.
Request your Free Audit.
Three fields, and a written reading back.
Your audit usually arrives within three working days. We use your details only to reply to your enquiry.
What you will get
What is working, what could improve and whether we can help.
Two or three priorities, with the reason for each.
A recommended next step, even if that step is not us.
There is no obligation to continue. We only arrange a call if you ask for one.
The work, written up
The case studies keep their dates, sources and limits visible.
See the casework →