How to Reduce Background Noise in a Voiceover Recording

Background noise in a voiceover recording is a career problem, not just a technical one. Clients notice it immediately. Platforms reject submissions because of it. And unlike some audio issues that are fixable in post, noise that's baked into a recording can only be reduced — not removed — without side effects.

The right approach is two-stage: eliminate as much noise as possible at the source, then handle what remains in post-processing. Here's how to do both.

Stage 1: Eliminate noise at the source

Post-processing is for the noise you couldn't prevent. Treating it as the primary defence is a mistake — heavy noise reduction always introduces artifacts that experienced listeners can detect, and some types of noise (intermittent noise, impulse noise, noise that varies in character) don't respond well to algorithmic treatment.

Identify your noise sources

Record 30 seconds of silence in your recording space at normal gain settings and listen back on headphones at full volume. You'll hear your actual noise floor. Common sources to identify:

• HVAC/air conditioning — a consistent broadband hiss that's the most common failure point for ACX submissions
• Electrical hum — a steady 50Hz (UK/Europe) or 60Hz (US) tone, sometimes with harmonics at 100Hz or 120Hz
• Computer fan noise — a mid-frequency drone, particularly from gaming rigs or older machines under load
• Traffic/environment — variable, unpredictable, worst in lower floors of urban buildings
• Neighbour/household noise — voices, footsteps, appliances

Fix HVAC noise

Turn off your air conditioning, heating fan, or ventilation system during recording sessions. This is the single most effective noise reduction step available to a home studio. Yes, it means recording in shorter sessions during summer. Yes, it's worth it — HVAC noise is extremely difficult to remove cleanly in post because it's broadband and sits in the same frequency range as voice.

Fix electrical hum

Electrical hum usually comes from a ground loop in your recording chain — typically when the computer, interface, and power supply share a ground path that creates a small current loop. Solutions:

• Use a balanced XLR cable from microphone to interface — balanced cables reject common-mode noise, including hum
• Plug everything into the same power strip / the same circuit
• Try a different USB port or cable for your interface
• A DI box with a ground lift switch solves stubborn hum cases

Reduce computer fan noise

Position your computer as far from the microphone as cable length allows. Move it under a desk or behind a physical barrier. For high-performance setups, consider running processing-intensive tasks (renders, updates) after recording rather than during.

Stage 2: Noise reduction in post-processing

After minimising noise at the source, your recording will still have some residual noise floor. Post-processing handles this. The key is using the right type of noise reduction at the right strength — aggressive processing always introduces the "underwater" artifact (a gurgling, phasey quality on voice consonants) that sounds worse than the noise it removed.

Types of noise reduction

• Static noise reduction — samples a region of silence to build a noise profile, then subtracts that profile from the entire recording. Works well for consistent, steady noise (HVAC, electrical hum). The standard approach in Audacity and most DAW noise reduction plugins.
• Dynamic noise reduction — adjusts in real time as noise levels change. More sophisticated than static, handles variable noise better, but still limited by the amount of reduction you can apply before artifacts appear.
• Speech isolation / AI-based noise reduction — separates voice from non-voice using a model trained on speech. The most effective approach for professional results. Doesn't subtract a noise profile — it identifies and preserves voice, discarding everything else. This is the method used in CleanCut VO's Full Polish and tools like Adobe Speech Enhance.

How much noise reduction to apply

For static noise reduction: 6–12 dB of reduction is usually appropriate. Above 12 dB, artifacts become audible. If you need more than 12 dB of noise reduction to reach your noise floor target, the problem is in the recording environment — post-processing can't compensate for fundamentally noisy source material.

For speech isolation models: 60–75% strength is a practical starting point for home studio recordings. Higher than 80% starts to affect voice consonants — you lose the crisp 's' and 't' sounds that give voice clarity and presence.

Hum reduction: a separate step

Electrical hum at 50Hz or 60Hz responds poorly to broadband noise reduction because it's a tonal signal at a specific frequency, not broadband noise. It needs notch filtering — a narrow filter that cuts specifically at 50Hz (and its harmonics at 100Hz, 150Hz, 200Hz) without touching the rest of the frequency range.

Most noise reduction plugins include a hum reduction option. CleanCut VO's Full Polish has a single toggle for 50Hz/60Hz hum elimination. It's fast and effective — turn it on if you're in the UK or Europe, 60Hz if you're in the US.

What noise reduction can't fix

Be realistic about what post-processing can achieve:

• Intermittent noise (dog barking, door slamming, passing traffic) — cannot be removed algorithmically. Re-record affected takes.
• Reverb and room sound — reverb is not noise. It's the acoustic fingerprint of your space baked into every sample. Noise reduction does nothing to it. Only room treatment prevents it.
• Clipping and distortion — cannot be repaired. Clipped audio is permanently damaged. Set your gain correctly before recording.
• Very high noise floors (above −45 dBFS) — if your noise floor is louder than −45 dBFS, post-processing will produce audible artifacts before the noise floor is sufficiently reduced. The room environment needs addressing first.