People forgive shaky footage long before they forgive bad sound. A hum under every sentence, a voice you have to turn up to hear, a clip that whispers and then shouts. Each one makes viewers leave, and each one needs a different fix. You can repair most of them in a video you already shot, once you know which one you have. Below we sort the common problems by how fixable they are, with the fix for each. Two of them really need a new take, and it is better to learn that before you spend an evening on them.
Bad audio is usually one of five problems, and the fix for one does little for another. Before you change anything, play the clip once on headphones. Laptop speakers hide the hiss and hum that your viewers will hear on earbuds.
If you mostly hear the last two, fix what you can and plan a new take of anything that matters. The rest of this guide covers the first three, because those are the ones software handles well.
A constant sound has the same character from the first second to the last, which makes it the easiest noise to remove. A noise reducer estimates the steady layer under the voice and subtracts it from the whole clip. In Supercut, drop the clip into the editor and ask in plain words, like "remove the background noise" or "clean up the audio". The filter runs on your device, and you preview the result before you export. If the voice itself sounds thin or muffled, also ask to "make my voice clearer", which cuts the rumble below speech and lifts the frequencies that carry consonants. Our guide to removing background noise from a video has the step-by-step version. One warning applies to every noise reducer. More is not better. Push the reduction too far and the voice starts to sound watery or robotic, which viewers find more distracting than a little hiss. Compare the cleaned version with the original before you commit, and keep whichever one a stranger would find easier to listen to.
Traffic, keyboard clicks, a door, a dog. Irregular noise is harder. Supercut’s noise reducer is the classic kind, which estimates one steady noise and subtracts it, and a passing truck has no steady profile. The reducer still takes out the constant layer under the truck, which helps, but the truck stays. For the video itself, do what helps most. Run the cleanup anyway for the steady part. Trim the worst moments if the edit allows it. Then add captions, so a sentence that a horn drowned out still gets read. Supercut transcribes on your device, so the audio does not travel anywhere to become text. An audio-only copy is a different case. If the recording will also go out as sound alone, like a podcast cut of an interview or the audio of a talk, extract the audio as an MP3 and clean that file with a speech denoiser. Speech denoisers are neural networks trained on voices. They keep what sounds like speech and suppress the rest, including noise that comes and goes. ihateaudio, a free set of browser audio tools, has one in its background noise remover, and it processes the file on your device too. Work on a copy and keep the original next to it, since no denoiser gets every file right.
Three different fixes all get called "make it louder", and they are not interchangeable. Volume, or gain, moves everything up or down by the same amount. Use it when the whole clip is evenly too quiet. In Supercut you can ask for an exact amount, like "increase the volume by 6 dB". Its limit is the loudest moment in the clip. Raise the gain too far and those peaks distort. Compression narrows the gap between the quiet and loud parts. A speaker who leans away from the microphone and back stays at an even level, and that evenness is most of what makes a voice sound steady and present. Ask Supercut for "podcast quality" and its podcast preset does the whole chain in one step. It trims the rumble, reduces steady noise, compresses the voice, and normalizes the result. Normalization measures how loud the clip is overall and moves it to a target. The useful kind measures loudness the way streaming platforms do, in LUFS, instead of looking at the single highest peak. Ask Supercut to "normalize the audio" and it aims at -16 LUFS with a true-peak ceiling of -1.5 dBTP, so the loudest word does not distort on the way up. For most talking videos, compression followed by normalization is the pair that fixes "hard to hear" without creating "suddenly too loud".
Platforms measure loudness in LUFS, short for loudness units relative to full scale. The scale follows how loud audio sounds to people rather than how high its peaks reach, and it runs in negative numbers, so -14 is louder than -16. Most big streaming platforms turn loud uploads down to a house level. At the time of writing, YouTube plays videos back at about -14 LUFS, and it turns loud ones down without turning quiet ones up. A video mixed too quietly simply stays quiet next to everything around it. The published targets differ by platform, and ihateaudio keeps a table of loudness targets by platform that covers music streaming, podcasts, and broadcast. TikTok and Instagram do not publish an official figure at the time of writing. For those, aim where the big platforms overlap, which is about -16 to -14 LUFS for a spoken video with peaks below -1 dBTP. Supercut’s normalize step lands inside that range at -16 LUFS. Make it the last audio step, after noise reduction and after any music goes in, because each of those changes the overall loudness.
When a video sounds bad, work through it in this order.
Clean up muddy audio and reduce background noise in a video, in your browser. No upload, your footage and audio stay on your device.
Make a quiet video louder or tame loud audio in your browser. Adjust volume in dB without uploading your clip.
Rip audio from any video and save as MP3 in your browser. Convert video to audio without uploading, your recording stays private.
Add auto-captions to any video in your browser. On-device transcription with viral caption styles, no upload, your footage and audio stay private.
Remove background noise from a video right in your browser. Type a plain-English prompt, preview, and export. Footage never uploads. Steps, tips, and FAQ.
Add subtitles or captions to any video by typing a plain-English prompt. On-device transcription times each word to your speech, and nothing uploads.
Turn podcast episodes into captioned vertical clips for social. Transcribe and reframe in your browser without uploading raw audio or video.
Trim lessons, add captions for accessibility, clean up audio, and cut promo clips, all in your browser. Unreleased course content never leaves your device.
Often, yes. Steady background noise, low volume, and uneven levels respond well to noise reduction, compression, and normalization. Two problems are part of the recording itself. Clipping is the harsh distortion from recording too loud, and heavy echo comes from the room. Software can soften both but not undo them.
Normalize it instead of only turning it up. Normalization measures the overall loudness and moves it to a target while keeping peaks under a ceiling, so the loudest words do not clip. In Supercut, ask to normalize the audio. It aims at -16 LUFS with a true-peak ceiling of -1.5 dBTP, and it runs in your browser.
At the time of writing, YouTube plays videos back at about -14 LUFS and turns louder ones down without turning quieter ones up. A mix between about -16 and -14 LUFS with peaks below -1 dBTP is a safe target. Anything louder than -14 gets turned down, and anything much quieter stays quiet.
Echo comes from sound bouncing off hard walls before it reaches a microphone that is too far away. Software can reduce it a little, but the reliable fix is on the next recording. Move the microphone closer to your mouth, and record in a room with soft things in it, like curtains, a rug, or a sofa.
No. Supercut runs its audio fixes in your browser, so the video stays on your device. If you clean an audio-only copy with a separate tool, pick one that also runs locally, and the recording never leaves your computer.