Production guides · Meme Plugins · 2 November 2026
Diagnose recording noise before choosing a repair. Hum often has a stable tonal pattern with related harmonics; hiss spreads energy more broadly. Listen to a noise-only passage and inspect its spectrum, then test the capture path before applying removal to the voice.
1. Listen to a representative quiet region
Keep the original and find a passage without the wanted performance. Compare several gaps because a fan, cable fault or moving background can change over time. Do not raise monitoring sharply and then forget to lower it before the voice returns.
Audacity’s Plot Spectrum documentation explains its frequency-versus-amplitude display. A narrow peak helps identify a candidate tone, but it does not by itself prove which physical device caused it.
Illustrative synthetic signal recipes
These are authored mathematical examples; no audio files or measurements are supplied. They illustrate the distinction rather than reproducing a particular recording.
- Tonal example: combine a 50 Hz sine with quieter 100 Hz and 150 Hz sines. The recipe contains a fundamental and related harmonics. A 60/120/180 Hz recipe is another illustration.
- Broadband example: use a low-level white-noise signal. It distributes energy across a wide band rather than supplying one stable note.
- Mixed example: combine the tonal recipe with the noise bed. Removing one narrow tone will not remove the whole broadband component.
Real hum can have more harmonics or changing levels, and real hiss is not always ideal white noise. Matching a mains-related frequency is a clue, not proof of a ground loop.
2. Isolate the capture path
Make short tests changing one connection or source at a time. Compare the same microphone and gain with a known suitable cable or another supported input. Note whether the noise changes with the source, device or room.
Follow manufacturer-approved connection checks and keep protective earth connected. If the noise is already in a source file, changing the playback buffer will not remove it; use the playback-versus-source diagnosis when the fault is intermittent.
3. Match the repair to the pattern
A narrow notch can target a stable tone, but it can also remove wanted musical energy at that frequency. A broadband reduction process needs a suitable noise profile and careful auditioning of the voice.
Audacity’s Noise Reduction manual describes profiling a noise-only region for relatively constant noise and warns about artifacts and unsuitable irregular sounds. Use a light candidate first, checking consonants and low-level endings.
4. Compare damage as well as noise
Match listening level and compare untreated and treated phrases in the song. If speech becomes watery or the instrument loses body, reduce the correction. A little steady background may be less distracting than heavy processing.
Render and reopen the result, preserving the source and the repair settings. For future takes, improve the recording setup before relying on removal.