Fix noisy dialogue without recording the take again

Two different noise problems, and Premiere has an answer for each

"My audio is noisy" usually means one of two genuinely different things. There's constant, mechanical noise — an AC hum, a fan, room tone, a fridge compressor — that sits at the same frequency the whole time a clip runs. And there's the kind of dullness you get from a laptop mic in an untreated room, where nothing is technically broken but the voice sounds thin, distant and slightly muddy. The first is a surgical problem: find the offending frequency and pull it down. The second is closer to what an AI model is actually good at: separating a voice from everything around it. Premiere has a dedicated tool for each, and they live in the same panel, which is exactly why people apply the wrong one and wonder why the result sounds hollow.

The Essential Sound panel: the surgical repair kit

Every audio-repair control in Premiere sits behind one classification step. Selecting a clip and telling Premiere it's dialogue is what unlocks the Repair section — skip it, and the sliders below simply aren't there.

1. Select the clip (or clips — multi-select works) on the timeline, then Window → Essential Sound.
2. Click Dialogue at the top of the panel to classify the clip. This is what exposes the Repair section below it.
3. Expand Repair and toggle on Reduce Noise for constant background hiss. Start the slider near 0 and raise it only until the noise fades — stop well before the voice starts to sound thin or wavery.
4. If there's a steady electrical hum, toggle DeHum and pick 50 Hz or 60 Hz depending on your region's mains frequency — the wrong one does nothing.
5. Reduce Rumble handles very low-frequency noise: an AC unit, a desk fan, traffic through a wall. It's a separate control from Reduce Noise because it targets a different frequency range.
6. DeEss tones down harsh "s" and "t" sounds a close condenser mic tends to exaggerate.
7. Reduce Reverb pulls back the roomy, bouncing-off-the-walls quality of dialogue recorded somewhere untreated — a spare room or a tiled kitchen, not a proper booth.

Enhance Speech: the AI pass, and where it actually helps

Enhance Speech is a separate button in the same panel, and it solves the other problem — a recording that isn't broken so much as thin. It runs an AI model locally on your machine rather than sending anything to a server, so it works offline and there's no per-clip usage limit to watch. What it's actually doing is closer to re-synthesizing a cleaner version of the voice than filtering the original, which is why it can rescue a recording that Reduce Noise alone would barely touch.

1. Select the dialogue clip and open Window → Essential Sound if it isn't already open.
2. In the Enhance section, click Enhance and let it process — this runs locally, so the wait is your machine's processing time, not an upload.
3. Once it finishes, use the Mix Amount slider to blend the enhanced result back with the original rather than committing to 100% automatically.

As a starting point: a lower Mix Amount, roughly in the 30–50 range, cleans up mildly noisy audio without it starting to sound processed. Save higher settings for audio that's genuinely rough. Enhance Speech isn't built for every situation — it doesn't do well with two people talking over each other, a loud music bed behind the dialogue, singing, or audio that's already badly degraded (old tape hiss, heavy compression artifacts). Those need the dedicated tools further down this post, not a higher Mix Amount.

Carrying the settings across the rest of the sequence

Both Reduce Noise and Enhance Speech add ordinary effects to the clip in Effect Controls behind the scenes — that's the useful part. Once one clip sounds right, select it, Edit → Copy, select every other clip from the same speaker in the same room, and Edit → Paste Attributes to carry the exact settings across instead of re-tuning each one from scratch.

Don't assume uniform is correct, though. A clip recorded right after someone switched off the AC doesn't have the same noise floor as one recorded while it was running, and pasting identical settings onto both can under-fix one and over-process the other. Paste as a starting point, then spot-check each cut rather than trusting the copy blindly.

The mistakes that make cleaned-up audio worse than the noise

• Pushing a slider to the maximum because a little bit worked. Every one of these controls has diminishing returns and an artifact threshold well before 100 — more is not a safer choice, it's a worse-sounding one.
• Judging the fix on studio headphones instead of on whatever most viewers will actually hear it through — earbuds, a phone speaker, a laptop. A subtly robotic edge that's obvious on good headphones can be inaudible on a phone, and the reverse is just as common.
• Applying one setting to a whole scene without re-checking each cut. Room tone shifts if a door opened, someone moved closer to the mic, or the fridge compressor in the next room kicked on mid-take.
• Only running the fix on the main dialogue track and forgetting a cutaway or reaction shot has its own room tone and its own hum — a clean main take next to an untouched insert shot is more noticeable than consistent noise throughout.
• Checking the cleaned dialogue in isolation and never against the full mix. A voice that sounds slightly processed on its own can sound fine once music and ambience are sitting underneath it — or the opposite, where a subtle artifact that was inaudible alone becomes distracting against a quiet music bed.

What to reach for when Premiere's own tools aren't enough

A dedicated repair tool for genuinely damaged audio

A single loud clank, a doorbell mid-sentence, a dog bark under one word — that's a spectral repair job, and it's a different category of problem from constant background noise. Dedicated audio-restoration tools built for exactly this (iZotope RX is the best-known one) can isolate and remove a single transient sound without touching the syllables around it, which Premiere's sliders aren't built to do. The cost is the round trip: export the clip, fix it externally, reimport, and re-sync it back onto the timeline — worth it for the one ruined line in an otherwise good take, overkill for routine background hiss.

Where this fits with a Backstage Cut edit

Backstage Cut doesn't clean up audio — that isn't one of its transcript-driven features, and everything above is the actual fix for it, on its own, free. What matters for a Backstage Cut edit specifically is the order: run this noise pass before transcribing the sequence, not after. Transcription is guessing at words from what it can hear, the same way a person would, and a hissy or humming track gives it worse material to work from. Clean the dialogue first, and the captions, chapters and AutoZoom timing that come off that one transcript are built against audio it actually understood correctly.

It also sets up Silence Cut better. A pause sitting inside a noisy track still has a floor of hiss in it, and ripple-deleting around that boundary is more likely to leave an audible pop where the cut lands. A pause in an already-cleaned track is closer to true silence on both sides of the cut, which is exactly the kind of boundary Silence Cut removes cleanly.

Backstage Cut starts with a free 7-day trial and no card — worth running once the dialogue track is already clean, since every caption, chapter and zoom it builds comes from one transcript, and a transcript is only ever as good as the audio it's reading.