Noise removal
Removing what should not be there without removing what should. The second half is the hard part.
The short version
Is this for you? For the voice scrubbed until it sounds robotic, and equally for the one nobody dares touch. The skill on this page is knowing when to stop as much as how far to go. Not if the problem is the room rather than the noise; reflections are made of your own voice and no removal reaches them cleanly, which is a different fix and we will tell you so.
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 honest call when a recording is better re-recorded than rescued.
What starts working. A recording that sounds like a person in a room, not a person in a plugin, and still legible at phone volume on a train, where most of your audience actually is.
The smallest way in. A free audit: three fields, and a written reading back, usually within about three working days. It commits you to nothing.
What to read next. The five levels to see where this fits, or the casework to see it done.
The room stays in, on purpose
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
Automatic and AI-driven removal will clean up a recording, and left unsupervised it will do it by trading away quality and legibility. We use those tools for the thing they are genuinely good at: finding where the noise lives and how much of it there is. What comes out, and how far, is a surgical decision taken by ear, one problem at a time, chosen by the fault in front of it and judged against the bypassed original rather than against silence, and stopped the moment more cleanliness would start costing the voice its body.
A microphone pointed away from the source, on purpose
Where there is a spare input, one microphone captures nothing but the room. That capture is a frequency portrait of the space, which makes removal accurate where it would otherwise be approximate, and it is the cleanest way to give a little of that space back at the end.
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
Where the noise comes from
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 actually stands.
Microphone placement decides how much can be saved
Too close and you buy plosives, harsh sibilance and a hot signal that clips. All of that is fixable afterwards and every one of them is cheaper to avoid. Too far and reflections start competing with the direct sound, which is the one thing removal cannot work without: clean removal needs enough direct source to hold on to.
A room is part of a voice. The job is not to remove the noise; it is to know what to leave.
Anyone can take a sound out; knowing when to stop, judged against the original and not against silence, is the part that is hard to buy.
What changes for you
A recording that sounds like a person in a room rather than a person in a plugin. Words that stay legible at phone volume on a train, which is where most of your audience actually is.
Related: Audio mixing · Multi-microphone setups · Microphone setup · Mastering for platform
Audio problems hide in plain sight because viewers feel them without naming them. The audit listens through recent videos and puts in writing what the sound is costing and what would fix it first.
Request your free audit.
Three fields, and a written reading back: a fit snapshot, two or three prioritised opportunities, and a recommended next step.
A person reads the channel and writes the audit by hand: a considered read typically takes three working days. That is the usual shape, not a promised turnaround. We use these details only to reply to you: no lists, no lurking.
What you will get
A fit snapshot: where your channel stands, and whether we are a match.
Two to three opportunities: specific, prioritised, yours to keep.
A recommended next step, even if that step is not us.
The audit is free and commits you to nothing: nobody follows up with a call you did not ask for.
The casework shows all of this done.
The case studies are the same work with the numbers, the sources and the dates attached.
See the casework →