Clean up the audio for a YouTube video

Watch how people behave on YouTube and the pattern is consistent: a slightly soft, badly-lit picture gets forgiven, and a viewer will happily sit through it if the content is worth it. Audio that hisses, hums or booms in a bare room gets a different response: the viewer stops watching, often within seconds, usually without being able to say exactly what was wrong.

That is not fussiness. Noisy audio makes speech genuinely harder to follow, and a viewer who has to concentrate to understand you is a viewer who leaves. Because watch time is what the platform actually rewards, a noisy soundtrack quietly costs you reach on top of the viewers it drives away directly.

The good news is that this is the cheapest problem in the whole production to fix. It takes one upload, and it happens after filming, so it works on footage you have already shot.

Almost every noisy upload comes down to the same short list, and CleanGrab handles all of it:

• Your computer. A laptop fan spinning up half a metre from the microphone is the single most common noise in screen recordings and tutorials.
• Air conditioning or a heater. A constant whoosh you stopped noticing in the room but which sits under every word on the recording.
• The street. Traffic rumble through a single-glazed window behind a talking-head shot.
• The microphone itself. Hiss and static from a cheap mic, or from a recording made too quietly and then turned up in editing.
• Mains hum. A low buzz from a ground loop or a bad cable, easy to miss on laptop speakers and painfully obvious in headphones.
• Typing. Keyboard clatter under a voiceover or a live screen recording.

Run CleanGrab on the finished video, before you add music and before you upload. Two reasons.

First, noise removal is designed to keep speech and suppress everything else, and a music bed is "everything else". Cleaning a video that already has music under the voice tends to leave the music sounding warbly and half-there. Clean the dialogue, then lay the music over the cleaned version in your editor.

Worth being precise about the job, too: if what you actually need is to boost a quiet track's volume, that's level, not noise, CleanGrab pulls the room out from behind your voice, it doesn't raise the volume of what's left. It's built to do that one job on purpose, rather than trying to be a full editor.

Second, YouTube will not do this for you. It re-encodes your upload and normalises loudness, and that normalisation actively works against you on a noisy file: quiet parts get pulled up, and the noise floor comes up with them. A recording that sounded acceptable on your laptop can sound noticeably hissier once the platform has evened it out.

Start on Balanced, play the before/after in the browser, and judge with headphones rather than laptop speakers: small speakers hide exactly the low hum and high hiss you are trying to catch.

If a layer of noise is still audible, move to Strong. If the voice has started to sound thin, hollow or slightly underwater, that is over-processing, and Gentle will keep more of the natural character of the voice. A little residual noise is less distracting than a voice that sounds like a phone call, so when in doubt, go gentler. CleanGrab is designed for speech, so birdsong and other nature sounds may be reduced as noise. There is no mode that isolates birdsong or keeps only a chosen nature sound. Changing the strength does not add that capability.

If you are publishing a series, keep the strength the same across episodes. A noise floor that shifts between clips draws attention to itself far more than one that stays quietly consistent.

Being straight about the limits saves you an upload:

• A very echoey room. Reverb is your own voice arriving late off bare walls, not a separate layer to subtract. Light echo is reduced; an empty room with hard floors is not solved in software. A rug, curtains, a bookshelf or just recording closer to the mic does more here than any tool.
• Clipped, distorted audio. If the level was far too high while recording, the peaks were flattened permanently at that moment. There is nothing left in the file to restore.
• A second person talking behind you. That is speech, and the model protects speech. It will not reliably keep only the presenter.
• Wind slamming into the mic. Steady wind rumble is reduced well; hard gusts that overload the capsule are distortion, and a foam or fur windshield prevents what software cannot repair.