Dialogue Cleanup Basics: Noise, EQ, De-Essing and Levels

Viewers forgive slightly soft picture far more easily than they forgive hard-to-hear speech. Clean dialogue is the single biggest improvement most small video projects can make. The good news is that the core cleanup steps are simple, repeatable and do not require a studio. This guide walks through them in the order that gives the best result with the least damage to the voice.

Start with the recording, not the plugins

No processing repairs a recording that was made badly. Distance from the microphone, room reflections and background hum all set the ceiling on what cleanup can achieve. Closer microphone placement improves the ratio between voice and room. A quiet room with soft furnishings reduces echo. A thirty-second recording of silence in the same room, called room tone, gives you a clean sample of the background for noise work later.

If you have room tone, record it before you leave the location. Keep the same microphone, gain and position. Room tone is the most useful tool in the whole cleanup chain, because it tells the software exactly what to remove and nothing else.

Remove steady noise first

Steady noise includes fan hum, air conditioning, refrigerator buzz and electrical hum at 50 or 60 hertz with its harmonics. Noise reduction works best on this kind of sound because it stays the same over time. Train the reduction on your room tone, then apply it at a moderate amount, usually well below the maximum.

Push too far and the voice starts to sound watery or robotic. Artifacts of this kind are the usual reason people abandon noise reduction altogether. A gentle setting that removes most of the hum is better than an aggressive setting that removes everything, including the warmth of the voice.

Irregular noise, such as a dog barking or a passing car, is harder. Cut it out with clip edits or fades instead of trying to filter it. A short gap that is faded properly is less distracting than a filter that partly smears the sound.

Cut the rumble with a high-pass filter

Human speech carries most of its useful energy between about 100 hertz and 8 kilohertz. Below roughly 80 to 100 hertz, a voice contains very little information, while the low end often collects rumble from wind, handling and air conditioning. A high-pass filter at that range removes the rumble without making a normal speaking voice sound thin.

Male voices have more energy in the low end than female voices, so set the cutoff by listening rather than by a fixed number. Sweep the filter upward until the voice starts to lose body, then back off a little. The goal is to remove mud that the viewer cannot hear as speech.

Shape tone with gentle EQ

Equalization changes the balance of frequencies. For voice, three ranges matter most. The low-mid region, roughly 200 to 400 hertz, can sound boomy or muddy when it is too strong. The presence region, roughly 2 to 5 kilohertz, helps the words stand out and improves clarity on small speakers. The air region above about 10 kilohertz adds sheen when it is gently lifted.

Make small moves, such as one to three decibels, and compare them with the original by toggling the effect on and off. A large boost usually means the microphone or the room needs a different solution. EQ is the tool for polishing a good recording, not for rescuing a bad one.

De-ess sharp sibilance

Sibilance is the hissing energy in sounds such as s, sh, ch and z. Microphones and close speaking can exaggerate it, especially in the range from about 5 to 8 kilohertz. A de-esser is a compressor that reacts only to that band, turning it down when a sibilant sound arrives.

Set the de-esser so it engages on the hard sounds and leaves ordinary words alone. You should be able to hear it working only when you listen very closely. If the speech starts to lisp or sound dull, the de-esser is set too hard, and the fix is to raise its threshold.

Control dynamics with compression

Speech varies a lot in level. A soft word and a loud word can differ by many decibels within one sentence. A compressor reduces the difference by turning down the loud parts. For dialogue, a ratio of about 2 to 1 to 4 to 1 is a common starting point, with a moderate attack and a release that lets the voice breathe.

Think of compression as evening out the path, not changing the sound. If you can hear the compressor pumping, with the background rising and falling under each word, reduce the amount of gain reduction or lengthen the release. The ideal result is a voice that is steady enough to hear on a phone speaker and still sounds natural on a large screen.

Match levels across clips

Once each clip sounds clean, balance the levels between clips. Interviews are often recorded on different days, in different rooms and with different microphone settings. A viewer will notice a jump in volume between two cuts even when each clip is clean on its own.

Clip gain is the best tool for this step because it changes the level before any processing runs. Listen to each clip in sequence, starting from the first, and adjust the gain until the voice sits at a similar apparent level from one cut to the next. Use short comparisons, a few seconds at a time, rather than trying to judge a whole scene at once.

Check the full chain at the end

After all the steps, play the whole dialogue track with your eyes closed. Listen for artifacts such as watery noise, sharp edges on sibilants, unnatural thinness or pumping. Then listen on a laptop speaker, on earbuds and on a larger monitor. Problems that hide on one device often stand out on another.

Keep the original recording untouched while you work. Apply each cleanup step to a copy or to a separate effect chain, so you can disable any step and compare it with the source. A cleanup that sounds good in isolation can sound wrong in the final mix, and the ability to switch steps off is the fastest way to find the one that is causing the problem.

Finally, measure the result against your loudness target. Dialogue cleanup and loudness work together, so a clean track that is far below the target should be raised with clip gain, not with a limiter that squeezes the voice.

Key takeaways

Good dialogue starts with a good recording and a room tone sample. Remove steady noise gently, cut rumble with a high-pass filter, shape tone with small EQ moves, and tame sibilance with a de-esser. Use compression to control variation without pumping, and balance levels with clip gain before the final loudness pass. Work in that order and the voice will sound clear, consistent and natural from one cut to the next.