Voice clean-up and audio levelling for iPhone video

Choose a voice treatment for the whole project

Voice treatment is a project-level choice, so you set it for the Cut instead of chasing a different clean-up setting on every clip. The panel gives you Studio, Natural, and Original. The treated choices use the voice chain; Original skips tonal and noise processing while still rendering a level-matched, peak-protected version.

That third option matters. A clean-up pass is not automatically better if you already like the recording. Original lets you keep its tone and noise profile while the loudness step gives the voice a consistent level and protects the peak. Compare the result with the same sentence, not by turning the treated version up and calling louder better.

What the voice chain actually does

The treated path uses six stages in a fixed order: high-pass, voice isolation, de-ess, compression, presence EQ, and peak limiter. Each stage has a narrow job. High-pass removes very low rumble. Voice isolation separates speech from surrounding sound. De-ess tames sharp sibilants. Compression narrows the gap between quiet and loud words. Presence EQ shapes the speech range, and the peak limiter catches overs that would otherwise make the mix jump or clip.

That is why the feature is closer to a voice clean-up tool than a one-tap volume slider. It works on the voice track first, then the final Cut is measured as a mix. You can use it on a talking-head take where the words are already there but the level and background make the line hard to sit beside the rest of the video.

• High-pass for very low rumble.
• Voice isolation for surrounding sound.
• De-ess, compression, and presence EQ for the voice treatment.
• Peak limiter for the final peaks in the chain.

Audio levelling uses a real loudness target

The voice track targets -18 LUFS. The final Cut mix targets -14 LUFS integrated with a -1.5 dBTP ceiling. Those are separate jobs: the voice needs a stable place in the mix, while the finished file needs a controlled overall level and a ceiling for peaks.

UGCut measures loudness with ITU-R BS.1770-5. The render measures after tone processing, applies gain and the limiter, and the export is measured again. That second pass matters because an encoded file is the file you deliver, not the meter reading from an earlier stage.

When you check an export, listen for words first, then compare the final level across clips. A number does not tell you whether a bad room sounds good, but it does give the Cut a repeatable level target.

What voice clean-up will not repair

Voice treatment cannot put detail back into clipped audio. Once a loud word has overloaded the recording, the harsh distortion is part of the source. It also does not turn room echo into a dry recording. Voice isolation can work on the surrounding sound the chain is designed to address, but a reflective room still changes the voice that reached the microphone.

Record a short test before the full take. If the room rings or the loudest word breaks up, move the phone, soften the room, lower the input level, or record the line again. Clean-up is a useful finishing pass. It is not permission to keep a take that already fails at the microphone.

Preview the original while the treated version renders

The voice chain runs on device, and voice isolation is device-only. While a treated version renders, the preview plays the original audio. That is intentional: you can keep working without mistaking the temporary source preview for the finished treatment.

When the treated version is ready, the Cut can use it for the selected treatment. If you choose Original, the app still renders the level-matched, peak-protected version described above. Export then re-measures the encoded file, so the final check follows the file that leaves the phone.

Keep the source and the finished version in mind during review. If the preview changes when the treated version becomes ready, judge the words, room, and peak behavior together, then choose the treatment that fits the recording rather than the one with the strongest effect.

Common questions

What voice treatments does UGCut have?

Three project-level choices: Studio, Natural, and Original. The treated choices use the voice chain; Original skips tonal and noise processing but still renders a level-matched, peak-protected version.

Does voice clean-up remove background noise?

Voice isolation is part of the treated chain and runs on device. It can address surrounding sound the chain is designed for, but it does not repair room echo or clipped recordings. If the source fails those tests, record the line again.

What loudness does UGCut target?

The voice track targets -18 LUFS. The final Cut mix targets -14 LUFS integrated with a -1.5 dBTP ceiling, measured with ITU-R BS.1770-5. Export is re-measured after encoding.

Does voice isolation upload my recording?

No. Voice isolation is device-only, and the voice chain runs on device. While the treated version renders, the preview plays the original audio.

Further reading

• How to fix bad audio in a video on iPhone

  Fix the room and microphone setup before you reach for clean-up.

• How to add captions to a video on iPhone

  Check the words after the voice is clear enough to caption.

• Why auto captions get words wrong

  The names and numbers worth checking before delivery.