Noise Reduction or Manual Repair: Which Suits Interview Audio?

Two ways to treat a noisy interview

Most interview and report recordings that reach us carry some trouble: a fridge hum in a kitchen, traffic behind a doorway, a sharp edge on every s. There are two broad ways to deal with it. The first is broadband noise reduction, where a processor learns the unwanted sound and lowers it across the whole file. The second is manual repair, where the editor works through the recording, treating individual problems with narrow filters, short fades and careful cuts.

Neither is better in every case, and learners often arrive believing one of them is the whole craft. In the dialogue cleanup module we ask people to treat the same ten-minute recording both ways and then listen to the results side by side on headphones and on a small laptop speaker. The comparison teaches more than any rule, because the damage each method can do only becomes obvious when you hear it next to the alternative.

This article sets out what each option does well, where it goes wrong, and which one we recommend as a starting point for producers working to a deadline.

What automated noise reduction does well

Broadband reduction is fast and consistent. Give it a few seconds of room tone with no speech, and it will lower steady sounds such as air conditioning, hum from a lamp, or the hiss of a quiet recorder. On a long interview film with one speaker in one room, a single pass can save an hour of work, and the result is the same from the first minute to the last, which matters when edits are later rearranged.

It also suits material that cannot be re-recorded. A report taped outside a building, with wind and distant machinery, has too many small problems to treat by hand. A well-set reduction pass brings the whole file down to a level where the voice can be heard clearly, and the producer can judge the content rather than the noise.

The strength is evenness. The weakness is that the processor cannot tell a hum from a low vowel, so everything it removes comes from the voice as well as the room.

• Best on steady noise: hum, hiss, fans, distant traffic
• Needs a clean sample of room tone to learn from
• Gives an even result across a long recording

Where automated reduction damages the voice

The familiar fault is a watery, underwater quality on the ends of words, caused by removing too much. Speech has soft tails and breaths, and a heavy setting strips them, leaving a voice that sounds thin and slightly robotic. Listeners rarely identify the cause, but they tire quickly and the speaker seems less trustworthy, which is the opposite of what an interview film needs.

Sibilance is the other trouble. Harsh s sounds are not steady noise, so reduction ignores them, and a general brightening to compensate can make them worse. A pass that sounds clean in a quiet edit suite can sound brittle once the track is mixed with music and played on a phone.

The remedy is restraint. Reduce less than seems necessary, listen to what has been removed on its own, and stop when you start to hear syllables in that discarded sound.

What manual repair offers instead

Manual repair treats each problem where it occurs. A narrow cut at the hum frequency leaves the rest of the voice alone. A short fade removes a chair creak between sentences. A de-esser set to a gentle ratio calms a speaker whose sibilance is sharp only on certain words. Because the editor decides what to touch, the natural weight and breath of the speech stay intact.

The cost is time and attention. A forty-minute interview can take several hours, and the editor must keep checking that the treatment stays consistent, so that the room does not seem to change from one answer to the next. It also asks for judgement that comes only with practice, which is why we introduce it after learners have heard what over-processing sounds like.

For short pieces, such as a one-minute explainer voiceover, the extra care is usually worth it. For hours of material, it rarely is.

• Keeps breaths and word endings natural
• Suits short pieces and difficult sibilance
• Slow on long recordings and harder to keep consistent

Our recommendation for most producers

Start with a light automated pass, then repair by hand. Reduce steady noise gently, perhaps a third of what the processor suggests, and then spend your attention on the specific faults that remain: a hum that survived, a pair of harsh words, a door slam between answers. This order gives you the speed of the first method and the care of the second, and it keeps the voice as the thing you protect.

If you can only choose one, choose by the material. Long, uniform interviews with steady background noise suit reduction alone, used lightly. Short voiceovers and recordings with irregular problems suit manual repair. Mixed material, which is most of what producers actually bring us, suits the combination.

Whatever you choose, compare against the untouched file every few minutes. Switching back and forth is the quickest way to notice that something has gone missing, and it costs nothing.

A short check before you pass the file on

Before sending cleaned dialogue to the next stage, listen once at a low volume and once at a normal one. At low volume, thin or harsh voices stand out more, and any pumping from the processor becomes clear. At normal volume, check the gaps between sentences: they should not drop to digital silence, which sounds unnatural, but keep a little of the room.

Keep the original file and note which settings you used, in a line or two beside the project. When feedback arrives in the review round, you will want to be able to soften a treatment without starting again. This habit also helps the person who picks up the work after you, whether that is a picture editor or the mixer.

These are the same checks we ask of learners in the module, and they take under five minutes.

• Listen at low and normal volume
• Leave a little room tone in the gaps
• Keep the original and note your settings