How to Clean Up Interview Audio Without Losing the Voice

Listen before you touch anything

Most damage to a voice happens in the first five minutes, when someone reaches for a noise reduction tool before they know what they are removing. I ask learners to start with headphones on, the full interview played through once, and a notepad beside the keyboard. Write down what you hear and roughly where it sits in the recording: a constant hum, a fridge that cycles on and off, traffic that rises and falls, a chair that creaks, a mouth click on every other sentence.

Different problems need different tools, so a list made by ear saves you from applying one heavy treatment to everything. It also tells you which problems you can live with. A little room tone under an interview is natural and often helps the edit feel continuous. The aim of dialogue cleanup is a voice that sits comfortably in the finished programme, not a recording that sounds as if it was made in a vacuum.

• Play the whole file once without editing
• Note each problem and where it occurs
• Decide which faults the viewer will actually notice
• Keep an untouched copy of the original

Removing hum and constant noise

Mains hum is the easiest fault to treat because it is steady and sits at fixed frequencies. A narrow notch filter, or a dedicated hum removal tool, can take out the base tone and its harmonics. Work with small cuts and keep checking against the original. If the voice starts to sound hollow or thin, you have cut too deeply or too wide, and it is better to leave a faint trace of hum than to lose warmth from the speaker.

For broadband noise such as air conditioning or distant traffic, use a noise profile taken from a few seconds where the speaker is silent. Apply modest reduction first, perhaps a few decibels, and listen to the gaps between words as well as the words themselves. If you hear watery or bubbling artefacts in the tails of sentences, the reduction is too strong. Two gentle passes usually sound more natural than one aggressive one.

• Notch hum narrowly and compare with the original
• Take the noise profile from genuine silence
• Start with a few decibels of reduction
• Listen to sentence endings for watery artefacts

Taming harsh sibilance

Sibilance is the sharp hiss on s and sh sounds, and it tends to show up in close-microphone interviews and in recordings made on small devices. A de-esser is the usual remedy. Set the frequency range to where the harshness lives, often somewhere between five and eight kilohertz, then lower the threshold until the worst sounds soften. Check the result on words such as sister, session and sensible, which expose over-processing quickly.

If the de-esser makes the speaker sound as if they have a slight lisp, back off. Some voices are simply bright, and a small reduction is kinder than a heavy one. For isolated bad sibilants, it can be better to lower the level of that single consonant by hand, using clip gain, than to process the whole recording. This takes longer, but it leaves the rest of the performance untouched and is worth the effort on a key line.

• Find the harsh band by sweeping while listening
• Test on words with several s sounds
• Prefer clip gain for one or two stubborn consonants

A sensible order of work

The order in which you apply fixes matters, because each step changes what the next one hears. I suggest removing clicks, bumps and obvious mouth noise first, then hum, then broadband noise, then sibilance, and only afterwards any tonal shaping or level work. Treating sibilance before noise reduction can lead the noise tool to react to hiss that you were about to remove anyway.

Keep each treatment as a separate stage or plugin slot so you can bypass it and compare. Switch each one off in turn and ask whether the voice sounds better or merely different. When a fix cannot be justified in that comparison, remove it. Save a version after each stage, with a clear name, so that a producer who asks for a softer touch can be answered without starting again from the original file.

• Clicks and bumps, then hum, then broadband noise
• Sibilance after noise reduction, not before
• Bypass each stage to confirm it helps
• Save named versions as you go

Checking on more than one system

Cleanup that sounds fine on studio headphones can reveal problems elsewhere. Play the treated interview through laptop speakers and a pair of ordinary earbuds, since most viewers will hear it that way. Small speakers tend to exaggerate sibilance and hide low-level hum, while earbuds expose breaths and residual noise in the gaps. A few minutes of this saves a revision round later.

Listen also in the context of the programme, not just on its own. An interview that has been cleaned thoroughly may feel strangely empty when it cuts against a location sequence with natural ambience. Leave a little of the original room sound in the gaps, or lay a soft bed of room tone beneath, so that the voice stays anchored in a real space. The treated dialogue is then ready to go forward to level balancing.

• Test on laptop speakers and earbuds
• Check the voice against neighbouring scenes
• Keep a little room tone in the gaps

When to stop

There is always one more thing you could improve, and that is the danger. Decide beforehand what finished looks like for this recording: no audible hum, noise that does not distract, sibilance that is comfortable, and a voice that still sounds like the person who spoke. When those four are met, stop and move on.

Some recordings cannot be fully rescued. If a speaker was recorded very close to a loud source, or the file is clipped, no amount of processing will restore it. In those cases it is more honest to tell the producer what is possible and, where a retake is feasible, to suggest one. Learners often find this the hardest lesson, but it is a practical one, and it protects both the voice and the schedule.