Basic Cleanup: Noise Reduction & Normalization

Audio Editing & Production Basics

Chapter 3 · Basic Cleanup: Noise Reduction & Normalization

Chapter 2 pointed out that a flat, near-silent stretch of waveform is exactly where noise reduction starts. This chapter puts that stretch to work — sampling it as a noise profile, removing that same background noise from the rest of the recording, then bringing the whole file up to a consistent, target volume.

Noise Reduction: Sampling a Profile

Every recording made outside a perfectly silent studio picks up some constant background noise — a fan, room hum, a faint hiss from the microphone itself. Noise reduction starts by selecting a stretch of the recording that contains only that background noise and nothing else — ideally a few seconds of the room tone or pause identified back in Chapter 2 — and using it to build a "noise profile," a fingerprint of exactly what that unwanted sound looks like.

Applying Noise Reduction

Once a profile exists, it's applied across the rest of the selection (often the entire track): Audacity looks for that same fingerprint throughout the recording and reduces it, leaving the wanted signal — the voice, the instrument — largely intact. This only works well because the profile was sampled from real silence in this specific recording; a generic, one-size-fits-all noise profile wouldn't match this room's own actual background sound.

Normalization: Setting a Target Level

Normalization raises or lowers a recording's overall volume so its loudest point sits at a specific target level, without changing the relative loudness of quieter and louder moments within the file. This matters for consistency: two podcast guests recorded on different microphones, or two episodes recorded on different days, can end up sounding evenly matched once both are normalized to the same target rather than one being noticeably quieter than the other.

Why Order Matters: Noise Reduction Before Normalization

Doing these two steps in the wrong order causes a real problem. Normalizing first raises the background noise right along with the wanted signal, since normalization can't tell the difference between the two — it only reads the overall waveform. Removing the noise first, then normalizing the now-cleaner recording, means the target volume is being set based on the signal that's actually wanted, not on a mix of that signal and the noise sitting underneath it.

StepWhat it doesDo this first?
Sample a noise profileSelects a silent stretch as a fingerprint of unwanted background noiseYes — first
Apply noise reductionRemoves that fingerprint from the rest of the recordingYes — second
NormalizeSets overall volume to a target level, without clippingNo — last, after cleanup
This chapter feeds Chapters 4 and 5
A recording that's already had its background noise removed and its overall level normalized is a much better starting point for the frequency-shaping work in Chapter 4 (Equalization) and the volume-evening work in Chapter 5 (Compression) — both build directly on top of this chapter's own cleanup pass, not on raw, unprocessed audio.
Aggressive noise reduction can sound worse than the noise it removed
It's tempting to push noise reduction as hard as possible, assuming more removal is always better. Pushed too far, noise reduction introduces its own artifacts — a warbly, "underwater" quality to voices, especially on quieter or breathier sounds. A moderate setting that leaves a faint trace of noise usually sounds far more natural than an aggressive setting that removes the noise completely but damages the voice along with it.

Hands-On Exercises

Exercise 1

A colleague normalizes their recording first, then tries to apply noise reduction, and is confused that the background hum still sounds noticeably loud. Using this chapter's own material, explain what went wrong.

📄 View solution
Exercise 2

A friend pushes their noise reduction setting to maximum strength "just to be safe," and now their narrator's voice sounds strange and warbly during quiet passages. Using this chapter's own warning box, explain what happened and what to do instead.

📄 View solution
Exercise 3

Explain why a noise profile sampled from one recording's own silent stretch works better on that recording than a generic, pre-made noise profile would, using this chapter's own material on how a profile is built.

📄 View solution

Chapter 3 Quick Reference

  • A noise profile is sampled from a silent stretch of the specific recording being cleaned up
  • Noise reduction uses that profile to remove the same background noise from the rest of the recording
  • Normalization sets overall volume to a target level without changing relative loudness within the file
  • Always reduce noise before normalizing — normalizing first raises the noise along with the signal
  • Overly aggressive noise reduction introduces artifacts — moderate settings usually sound more natural