First, Identify the Noise
Background noise is not one problem. It is several problems with different fixes.
- Constant broadband noise: fan, AC hum, computer whir, mic hiss. This is the easiest for AI noise removal because the model learns a steady profile and subtracts it.
- Electrical hum: low 50 Hz or 60 Hz drone. A notch or high-pass filter helps; AI can often remove it without thinning the voice.
- Intermittent noise: keyboard clicks, notifications, dog barks, traffic. AI reduces some, but hard clicks and barks usually need manual editing.
- Room reverb: echo from bare walls. Noise removal does not fix reverb. You need de-reverb, treatment, or a different mic position.
- Clipping: if audio is already distorted, denoising will not restore it. It can make it worse.
Quick Fixes Before AI
Work from the best source
Transfer the original phone recording, not a compressed messaging copy. Use WAV or FLAC if possible. If you only have MP4, extract audio to WAV at 48 kHz or 44.1 kHz. Lossy files already contain artifacts, and aggressive denoising makes them louder.
Back up the original
Always work on a copy. You may want to compare settings or start over.
Trim and isolate
Delete long silences. If mic and game audio are separate tracks, process only the mic. A mixed file gives you less control.
Set a baseline level
Normalize the loudest peak to about -3 dB. For podcasts and voiceovers, aim for -16 LUFS average. For YouTube, around -14 LUFS.
Try a high-pass filter
Most voice content has little useful energy below 80 Hz. Set a high-pass at 80-100 Hz for male voices, 100-120 Hz for female voices. If you hear 50/60 Hz hum, add a narrow notch there. Too high and the voice sounds thin.
Use a gentle noise gate only if needed
Set threshold just above the noise floor, ratio around 2:1, attack 5-10 ms, release 150-300 ms, range -6 to -12 dB instead of full mute. This reduces room tone without chopping word beginnings.
How AI Removal Fits In
AI noise removal separates speech from noise. It handles complex, changing noise better than EQ alone. Some video editors include it, but a free online option is often faster for a single file. ClearVoice offers AI background noise removal for video and audio: upload the recording, let it process, and download the cleaned file. It works on podcasts, Zoom calls, voice memos, and gaming clips without installing an editor.
Use AI as one step, not the whole workflow. It removes noise, but it does not fix levels, reverb, or clipped audio.
Step-by-Step: Remove Noise from a Video Clip
- Detach the audio. In most editors, right-click the clip and choose Detach Audio or Extract Audio. Online tools often accept the video directly.
- Test on 20-30 seconds. Pick a section with speech and the worst noise. Run the AI denoiser and listen with headphones.
- Set the reduction amount. Start at 50-60%. Go to 70-80% only if noise is still obvious. Back off if the voice sounds watery, robotic, or hollow. For one-click tools, compare processed and original before committing.
- Patch remaining noises manually. Use spectral repair, a short crossfade, or cut-and-delete for clicks, coughs, and lip smacks.
- Shape with EQ. Add a high-pass at 80-100 Hz. If dull, boost presence around 2-5 kHz. If sibilance is harsh, use a de-esser.
- Compress and limit. Use 2:1 to 3:1 ratio, 10 ms attack, 80-150 ms release, 3-6 dB gain reduction. Set limiter ceiling at -1 dBTP.
- Export and test. Listen on phone, laptop, and earbuds. Adjust and re-export if needed.
Faster version: high-pass filter, AI pass, then a light compressor. That handles most voice memos and Zoom recordings in minutes.
Settings Cheat Sheet
- Podcast/voice memo: high-pass 80-100 Hz, AI denoise 50-70%, gate -35 to -45 dB with soft range, target -16 LUFS.
- Zoom/Meet: platform may already suppress noise. Use lighter AI reduction, 30-50%, high-pass 100 Hz. Avoid heavy gating; it cuts first syllables.
- Phone recording: high-pass 100-120 Hz, AI denoise 60-75%, manual edits for wind blasts. Low-frequency wind rumble is easier to fix than clipping.
- Gaming clip: if mic and game audio are separate, denoise only the mic. If mixed, AI speech isolation may remove game effects or music. Use EQ and a gate on the mic before mixing, or re-record with separate tracks.
Common Mistakes and Tradeoffs
- Cranking reduction to 100%. Fastest way to robotic, underwater audio. Use the least amount that solves the problem.
- Using a gate as the only fix. Gates create unnatural silence and can chop words. Use after denoising.
- Processing the original. Always work on a copy.
- Expecting AI to fix reverb or clipping. It cannot. Reverb needs de-reverb or better acoustics. Clipping needs a re-record.
- Ignoring the noise floor. No tool makes a noisy room sound like a studio. Move the mic closer, turn off the fan, and close the window next time.
A Repeatable Workflow
- Back up the original.
- High-pass at 80-100 Hz.
- AI background noise removal at 50-70%.
- Manual cleanup for clicks and coughs.
- EQ, compression, and limiting.
- Export and test on phone, laptop, and earbuds.
You do not need a studio to get clean audio. Match the tool to the noise, use AI for the heavy lifting, and stop before the voice sounds processed. If you want a quick free option for a video or audio file, try ClearVoice for AI background noise removal and see how it handles your recording.