The Order to Clean Up Bad Audio

When presented with a dialogue recording that contains background hiss, room resonance, and harsh sibilance, the instinct is to start applying effects randomly until it sounds better. You add an equaliser to make the voice brighter, a compressor to make it louder, and then a noise reduction tool to remove the hiss that just got louder.

This approach destroys audio. Audio processing happens in a linear signal chain. Each plugin receives the audio output from the plugin directly above it. If your first step is compression, you are compressing the background noise, the low-end rumble, and the harsh 'S' sounds right along with the voice. You have permanently embedded those flaws into the signal, and no plugin further down the chain can cleanly remove them.

To achieve a professional, broadcast-quality voice track, you must apply processing in a specific, subtractive-to-additive order. You remove the problems first, control the dynamics second, and enhance the tone last.

The Standard Signal Chain

The following order of operations is the industry standard for dialogue processing. While specific plugins vary by editor, the categories and sequence remain identical whether you are working in Premiere Pro, DaVinci Resolve, or a dedicated Digital Audio Workstation (DAW) like Pro Tools.

1. Noise Reduction

The first step in any chain is to clean the raw signal. Noise reduction tools identify continuous background frequencies—such as air conditioning hum, refrigerator drone, or microphone preamp hiss—and phase-cancel them.

If you apply noise reduction after a compressor, you force the tool to work harder. The compressor will have already brought the noise floor up, making the hiss louder and more prominent. By placing noise reduction first, you isolate the interference while it is still quiet.

Do not try to remove all of the background noise. Pushing a noise reduction plugin too hard produces watery, chirping artefacts and makes the voice sound like it was recorded inside a tin can. Reduce the noise until it stops drawing attention to itself, then stop — a faint, steady room tone is far easier to listen to than a clean voice surrounded by digital debris.

2. Subtractive EQ

Once the continuous noise is gone, you must address frequency imbalances using an Equaliser (EQ). In this stage, you are strictly cutting frequencies, not boosting them.

First, apply a High-Pass Filter (also known as a Low-Cut Filter). Set this around 60Hz to 80Hz. Human voices do not produce meaningful frequencies below this point. What resides there is low-end rumble: traffic outside, microphone stand vibrations, and plosive thumps from 'P' and 'B' consonants. Cutting these frequencies frees up massive amounts of headroom in your mix.

Next, sweep for resonant frequencies. These are specific, narrow frequency bands that ring or sound muddy, usually caused by the acoustic reflections of an untreated room. Use a narrow EQ band, boost it significantly, and sweep it across the lower midrange (200Hz to 500Hz). When you hear a terrible, boxy ringing sound, you have found the resonance. Pull that specific band down by -3dB to -5dB.

3. De-Essing

Sibilance refers to the harsh, piercing high frequencies generated by 'S', 'Sh', and 'T' sounds. If a speaker is too close to a sensitive microphone, these frequencies can be physically painful to listen to.

A De-Esser is essentially a highly specific compressor that only activates when it detects those harsh high frequencies (typically between 5kHz and 8kHz). It ducks the volume for a fraction of a second when an 'S' occurs, and leaves the rest of the audio untouched.

You must de-ess before your main compressor. If a harsh 'S' hits your main compressor, the compressor will react to that massive spike in high-frequency energy and duck the volume of the entire voice, causing noticeable pumping artifacts. By taming the 'S' first, you provide the main compressor with a much smoother signal.

4. Compression

Now that you have removed the noise, the rumble, and the harshness, you are left with a clean voice. However, human speech is highly dynamic. A person might whisper one sentence and shout the next. If you turn the track up so the whisper is audible, the shout will clip the master bus.

A compressor reduces dynamic range. When the audio exceeds a specific volume threshold, the compressor turns it down by a specified ratio.

For standard dialogue, set a ratio of 3:1 or 4:1. Set the threshold so that the compressor only activates (shows gain reduction) on the loudest words or sentences, pulling them down by -3dB to -6dB. With the loud peaks controlled, you can now use the compressor's "Makeup Gain" to turn the entire track up safely. The quiet words become louder, the loud words remain controlled, and the voice sits consistently in the mix.

5. Additive EQ

With the dynamics controlled, the voice might sound a bit flat or dull. This is the time to add presence.

Because you are applying this EQ after the compressor, any boosts you make will remain consistent. If you had boosted the high frequencies before the compressor, the compressor would fight you, pulling those highs back down every time it activated.

For additive EQ, use wide, gentle curves. A subtle boost of +2dB around 3kHz to 5kHz will increase articulation and intelligibility, helping the voice cut through background music. A "high shelf" boost of +1dB above 10kHz will add "air" and expensive-sounding clarity to the recording. Do not boost the low end; if the voice lacks bass, it is usually better to use a tool like a sub-harmonic generator, or simply accept the tone of the microphone.

6. Limiting

The final plugin on your dialogue track—or more commonly, on your master output bus—must be a Limiter. A limiter is a compressor with an infinite ratio; it acts as a brick wall that audio cannot pass.

Even with a compressor active, a sudden laugh or a pop could occasionally slip through and cause digital clipping (exceeding 0 dBFS), which results in aggressive distortion.

Set your limiter's "True Peak" ceiling to -1.0 dB. This guarantees that no matter what happens in the mix, the audio will never clip. You can then gently push the input gain of the limiter until your overall mix hits the standard loudness target for web video, which is -14 LUFS.

Where This Chain Sits in the Edit

The signal chain is the last thing you build, not the first. Everything above assumes the dialogue track is already assembled — takes chosen, stumbles removed, joins patched — because every cut you make afterwards lands inside processed audio and has to be re-checked against it.

That ordering has a practical consequence. Do the speech editing first, including the gap-filling that keeps the noise floor continuous; the technique for that is in removing filler words without making speech sound chopped. A discontinuous noise floor fed into a noise reduction plugin produces uneven results, because the plugin is learning a profile from a track whose background keeps changing.

The chain also depends on how the recording was made, and there is a hard limit to what it can recover. A voice captured too far from the microphone brings the room with it in a way no plugin fully removes, which is why the cheapest improvement in audio quality is nearly always at capture rather than in post — recording a clean reference clip covers what that actually takes.

Finally, keep track structure separate from processing. Applying this chain to a track that also carries music and effects means compressing all three together, which is why dialogue, effects and music each need their own track from the start, as described in timeline editing fundamentals.

Next Steps

If you are working in a non-linear video editor like Premiere Pro or DaVinci Resolve, you do not need third-party audio plugins to build this chain. Every professional editor includes native tools for EQ, compression, and limiting in their audio effects panel.

Your next step is to create a custom effects preset. Open your editor, apply the six effects listed above to an empty audio track in the correct order, and configure the baseline settings (e.g. the 80Hz high-pass filter and the -1.0dB limiter ceiling). Save this chain as a track preset. The next time you begin an edit, apply the preset and adjust the noise reduction and compressor thresholds to fit the specific recording.

Build it once and resist rebuilding it per project. The values that change between recordings are the noise profile and the compressor threshold; everything else is a constant, and treating it as a constant is what turns an hour of fiddling into five minutes of adjustment. The rest of the post-production craft is collected under editing craft and workflow.