Audio Engineering
Background Noise Remover
Remove steady background noise (fan hum, air conditioning, computer noise) from any audio file. Uses FFT-based spectral subtraction. Runs entirely in your browser.
Knowledge Base
How to Use the Background Noise Remover
The Background Noise Remover uses spectral analysis to identify and attenuate constant background noise (air conditioning, traffic, computer hum) without affecting voice or instruments. Upload a noisy recording, select noise profile, and download cleaned audio.
How Noise Reduction Works
The tool samples a short silence (noise-only section) from your recording, learns the frequency pattern, and subtracts that pattern from the entire file. This works best for steady, consistent noise, like HVAC hum, electrical buzz, or distant traffic.
When Noise Reduction Helps
- HVAC/Fan Noise: Air conditioning, computer fans, furnace rumble. Excellent results.
- Electrical Hum: 50/60 Hz hum from power sources. Highly removable.
- Traffic/Ambient: Distant traffic or office noise. Partial removal (some artifacts).
- White Noise: Constant white/brown noise. Moderate removal.
When It Struggles
Intermittent noise (dogs barking, phone ringing) is hard to remove without distorting voice. Noise similar in frequency to voice is difficult, high-frequency hiss removal can make audio sound thin. Recording in a quiet space is always better than fixing in post.
Best Practices
Use the strongest reduction setting that doesn't degrade your voice/content. Listen critically to the result, artifacts (comb filtering, reverb-like quality) mean you overdid it. Start conservative and increase strength until you find the balance.
FAQ
Q: Will noise reduction make my audio sound weird?
A: Some artifacts are possible (slightly robotic, thinned-out sound) depending on noise type and reduction strength. Subtle settings are best.
Q: How much noise can I remove?
A: Steady noise at -30 dB or quieter relative to voice can be mostly removed. Loud noise (like traffic) leaves artifacts.
Getting a clean result without wrecking the voice
Spectral subtraction works by measuring what the steady noise looks like across the frequency range, then removing that shape from the whole recording. Because it removes a fixed profile, it handles constant sounds well: air conditioning, a computer fan, mains hum, road noise through a window. It handles nothing that changes. A door slam, a chair creak or a passing siren stays exactly where it is.
The most common mistake is asking for too much. Every decibel of reduction also removes part of the voice, because speech and noise share the same frequencies. Past a certain point you hear the result as a thin, watery quality with faint tones warbling in the gaps between words. That artefact is called musical noise, and once it is in a recording no further processing removes it. Reduce until the noise stops being distracting, then stop.
Work in this order. Trim the recording to the parts you actually need first, so the tool measures noise from a representative section. Apply the least reduction that helps. Listen to the quiet passages rather than the loud ones, because that is where damage shows. Then do any compression or loudness work afterwards, since compression raises the noise floor and will amplify whatever you left behind.
If the noise is louder than about a quarter of the voice level, no repair will sound clean. Re-recording in a quieter moment, or moving the microphone closer to the speaker, fixes in seconds what an hour of processing cannot. Getting the microphone close is the single most effective noise reduction available, because it raises the voice without raising the room.
Vocabulary
Noise floor. The constant background level present in a recording when nobody is speaking. Measured in decibels below the loudest peak. A low noise floor gives you room to raise quiet passages later without the background rising with them.
Spectral subtraction. A method that measures the frequency shape of steady background noise and subtracts it from the whole recording. Effective on constant sounds and useless on sudden ones, because the measured profile no longer matches.
Musical noise. The watery, warbling artefact left behind when noise reduction is pushed too hard. Faint tones appear in the gaps between words. It cannot be removed afterwards, so the only defence is using less reduction in the first place.
Signal to noise ratio. The gap between the level of the thing you want and the level of everything else, in decibels. A large gap makes repair easy. A small gap means no processing will produce a clean result and the recording needs doing again.