
Reduce steady air-conditioner or fan noise in a narration file, choose between four cleanup levels, and avoid the watery over-processed sound.
The take was good: steady pacing, no stumbles, the right emphasis on the product name. Then you put on headphones and hear it, a low roar underneath every sentence that disappears only when you stop talking. In a home office that is usually the air conditioner or a furnace fan. Background Noise Remover is built for exactly this kind of constant noise: hum, hiss and fan sound that stays at roughly the same level.
The clip below shows the upload, level choice and download.
It uses spectral subtraction: it measures the noise in your own file and subtracts that pattern from the whole recording. That works well on a steady noise floor and poorly on a busy, changing scene such as a crowded room or a conversation with traffic and dishes. A separate fourth level, Voice, applies a small noise-suppression model for speech, which copes better when the noise changes. More on that below.
Be honest with yourself about the source. Cleanup reduces noise; it cannot rebuild a voice recorded far from the microphone, and strong settings take some of the voice's body along with the noise.
For each file the tool first listens for the quietest stretches, averaging the quietest quarter of short 50 millisecond windows, and treats that as the noise floor. Suppose your narration measures -52 dB. The three standard levels then work like this:
So the same file is tuned differently from a quieter room's file, because the target moves with the measured floor. Higher levels clean more but risk a dull, underwater voice.
Say you recorded a three-minute voiceover with the air conditioning cycling and a steady hiss beneath it. Open the tool, drop the file and wait for the file name to appear. If a size notice shows, the file is over your plan's limit and Remove Noise stays disabled; trim it first with Audio Trimmer.
When the process is done, a line under the player reports roughly how much quieter the pauses became, for example about 18 dB in the quietest parts. It measures the gaps, not the speech, so treat it as an indication, not a promise. If it reports little noise to remove, the recording was already clean or the noise is not steady.
Voice is meant for speech with changing background sound: a fan that cycles, traffic passing, a keyboard being tapped. It loads a small voice model from this site on first use, so the first run takes a moment longer. It is not suitable for music, where it can damage the sound, so do not run a song or a musical bed through it.
Many voiceover artists try Standard first, since it is gentler on tone, then fall back to Voice if the noise moves. If both leave a watery quality, the better answer is a re-record in a quieter space, even with a blanket over the microphone, rather than a harder setting.
If your take is inside a video file, Extract Audio pulls the sound out first so you can clean it separately. After cleanup, trim the pauses and the false starts, then normalise. The processing runs in your browser, so the recording is not uploaded, although the audio engine has to load first. For wider cleanup ideas there is the earlier post on removing hum from voice memos, and the audio tools category lists the other steps.
What does your recording space sound like most often: air conditioning, computer fans, traffic or something else? Tell us on the contact page.