Guide · Explainer
What Noise Can (and Can't) Be Removed from a Recording
"Can you remove the background noise?" The honest answer is: it depends entirely on the kind of noise. Some comes out so cleanly it feels like magic; some barely moves, no matter which tool you use. The difference isn't the software — it's the noise. Here's how to tell which you have before you spend time on it.
Noise removal is excellent on steady noise — fan, AC, mains hum, hiss, constant room tone — because it's predictable. It's unreliable on complex noise — background voices, echo, music — because that overlaps and mimics speech. Listen to the gaps between words: a constant hum removes well; other voices or echo are the hard cases.
The one distinction that matters: steady vs complex
Almost everything about what's removable comes down to whether the noise is steady or complex:
✓ Removes reliably
- Fans, laptops and air conditioning
- Mains hum and electrical buzz
- Tape hiss and broadband hiss
- Constant room tone and rumble
~ Results vary
- Background voices and café chatter
- Echo and room reverb
- Music bleed and mixed vocals
- Very damaged or distorted audio
Why steady noise is easy to remove
Steady noise has a constant, predictable signature — a fan hum sounds the same from second to second. That makes it easy for a model to learn the noise and subtract it while leaving the voice intact. You can see it in a recording: before cleanup there's a noise floor everywhere, even between words; after, those gaps drop to near-silence and the voice sits out front.
Why the hard cases are hard
- Background voices look like your voice — same frequencies, same rhythms — so a speech model can't cleanly tell which voice to keep.
- Echo and reverb are your own voice, smeared and repeated in time. There's no separate "noise" to subtract.
- Music is structured and overlaps speech across the spectrum, so removing it tends to take chunks of the voice with it.
We measured it — including the weak case
Most tools show you only their best examples. Soundproof publishes the numbers across noise types, including the one it's worst at. This is objective signal-to-noise improvement (SI-SDR gain) at a hard 5 dB starting point — higher means more noise removed:
Hum, hiss and fan noise post big, consistent gains. Babble — background voices — hovers near zero and can even come out slightly worse. That's not a flaw in the chart; it's the honest shape of the problem, and it's why we show it.
Play your recording and listen to a gap between sentences. If you hear a constant hum, whir or hiss, it'll clean up well. If you hear other people talking or a distinct echo, that's the hard case — you can improve it, but don't expect a studio result.
So what should you expect?
If your problem is a fan, an air conditioner, mains hum or hiss, you're in the sweet spot — expect a clearly cleaner recording. If it's a café full of chatter, a boomy echoey room, or music under the voice, temper expectations: it may improve, but no tool reliably removes those. The best move is to try it on your actual file — a free download that cleans your own recordings tells you in two seconds whether yours is a good fit.
Find out how your recording cleans up
Soundproof is excellent on steady noise and honest about the rest. Download it free, clean your own file, and hear the before/after before you pay a cent.
macOS 13.4+ · Apple Silicon & Intel · Works offline · One-time purchase, no subscription
Frequently asked questions
Can you remove background voices from a recording?
Not reliably. Other voices share the same frequencies and rhythms as your voice, so a speech denoiser can't cleanly separate them. Results vary — some improvement, but often still audible.
Can noise removal fix echo or reverb?
Only partly. Echo is your own voice repeated and smeared in time rather than a separate noise, so a denoiser isn't the right tool. It may sound easier to listen to if there's steady noise underneath, but the echo can remain.
Why does babble sometimes come out worse?
Because the model can mistake background speech for the main voice and remove the wrong parts. That's exactly why we publish the babble result instead of hiding it.
How do I know if my recording will clean up well?
Listen to the gaps between words. A constant hum or hiss removes cleanly; other voices or echo are hard. Or just try it — the free download cleans your own file so you can hear the result first.
What noise is Soundproof best at?
Steady background noise behind a voice: fans, air conditioning, mains hum, and hiss. Those are the cases with the biggest, most consistent improvement.