Set Noise Triage: Fix Generators, HVAC & Traffic
Executive Summary
Set noise is diagnosed, not denoised. Every mitigation on set is cheaper, cleaner, and more transparent than any iZotope RX pass in post. This guide covers the five noise categories that eat indie dialogue tracks (generators, HVAC, refrigerators, traffic, crew) with source-level fixes: distance and load management for gennies, mechanical isolation for HVAC and fridges, mic proximity and take-timing for traffic, and a rehearsed quiet-set protocol for the crew. It closes with a staged post workflow (de-hum, de-click, voice denoise, de-reverb, de-ess) built around A/B comparison, so cleanup never becomes more audible than the noise it removes.
Key rules of thumb:
Table of Contents
For the full sound workflow, start with the Production Sound Definitive Guide: Set Recording to Editorial Handoff.
Source Identification and Triage: What Kind of Noise Is It?
Before any mitigation begins, diagnose accurately. Not all noise is created equal, and the appropriate fix depends entirely on the source. Is it a steady tonal hum, a broadband whoosh, an intermittent compressor cycle, or a transient car horn? Current production sound practice classifies noise before choosing a remedy. A steady tonal hum calls for a surgical notch filter. Broadband environmental noise (wind, room tone) requires a combination of mic placement, physical isolation, and light spectral denoising.
The process starts with listening in two places: at the recorder through closed-back headphones (Sony MDR-7506 or Sennheiser HD 25 remain the location standards), and at the boom operator''s position where the microphone actually lives. Dual monitoring exposes the gap between what the mic captures and what the ear perceives across the room.
Classify the noise into one of six buckets, because each drives a different response:
| Noise Type | Typical Frequency | Character | First Move |
|---|---|---|---|
| Electrical hum | 50/60 Hz + harmonics | Steady tonal | Notch filter, ground check |
| HVAC rush | 100 Hz to 4 kHz broadband | Steady broadband | Turn off; if not, HPF at 80 Hz |
| Fridge compressor | Cyclic, 60 to 200 Hz tonal | Intermittent tonal | Unplug; tag with tape |
| Generator | 40 to 500 Hz + harmonics | Steady, load-dependent | Distance + blankets |
| Traffic | Transient, wideband | Event-based | Time takes, mic proximity |
| Crew | Random, wideband | Event-based | Protocol, quiet-set call |
Two common early mistakes: treating every noise problem with a generic broadband denoiser, and ignoring intermittent sources because they are quieter than a steady hum. A fridge compressor kicking on mid-take is worse than a constant HVAC hum, because the ear locks onto onset transients. Distinguish steady bed noise (reducible transparently) from event noise (usually requiring surgical intervention or prevention).
If noise only appears in dialogue pauses, careful manual editing or a light gate is more transparent than aggressive reduction. iZotope RX remains the industry standard for dialogue repair, with dedicated modules for De-hum, Voice De-noise, De-click, and Spectral Repair. DaVinci Resolve Fairlight includes solid noise reduction, a de-hummer, and dynamic gating inside the NLE. Audacity handles basic noise profiling for budget workflows.
💡 Pro Tip: Record a 30-second wild track of room tone at every major setup, slated verbally with scene and take. It is the single most valuable insurance policy in a sound bag.
Generators: Power Generation as a Sound Problem, Not Just an Electrical One
Generators reach the set through two paths: airborne noise (engine, alternator whine) and structure-borne vibration (ground coupling, cable-transmitted rattle). Simply moving the gennie farther away addresses only half the problem. Effective mitigation combines distance, physical barriers, mechanical isolation, and stable load management.
Distance first. Sound pressure follows the inverse square law in free field: every doubling of distance drops the level about 6 dB. A tow-plant generator that reads 75 dB SPL at 3 m reads roughly 69 dB at 6 m, 63 dB at 12 m, and 57 dB at 24 m. Real locations rarely behave like free field (buildings reflect, terrain absorbs), but the math sets expectations. For dialogue-critical interiors, 20 to 30 m (65 to 100 ft) of separation is a reasonable target for a mid-size diesel plant, extended by cable length limits.
Physical masking. Sound blankets (moving blankets or purpose-built acoustic blankets), plywood baffles, or natural terrain features (a building corner, a berm, a parked truck) attenuate airborne noise. Blankets are most effective on the direct line of sight between the generator''s exhaust and the recording area. A generator tucked behind a solid wall is dramatically quieter than one behind the same wall with a corner open to the set.
Cable discipline. Longer runs of heavier gauge cable (4/0 stage cable or 2/0 depending on load) preserve voltage over distance. Route cabling so the generator sits where its noise is masked, not where the cable run is shortest. Check that a "convenient" parking spot is not directly against a thin exterior wall of a quiet interior location, which negates the distance you just paid for.
Load planning. Generators produce different noise profiles under different loads. Unstable loads cause the engine governor to hunt, creating audible RPM changes during takes. Aim for a stable, consistent load: keep HMIs, hair and makeup trailer HVAC, and craft-services equipment either fully on or fully off across a scene, not surging. When the generator noise is predictable, post can match room tone across cuts. When it changes mid-scene, no denoiser will hide the transitions.
Modern quiet options. Whisper-series tow plants (Multiquip WhisperWatt, Wacker Neuson) are rated in the 63 to 68 dB(A) range at 7 m under load, versus 75 to 80 dB(A) for standard construction plants. Battery-plus-inverter systems (Portable Electric VOLTstack, Instagrid) run genuinely silent for lower-draw setups and are increasingly used for dialogue-only interior days. Even when a gas plant runs the lights, switching sound-adjacent power to a battery cart eliminates a common ground-hum path.
Common mistakes: parking too close for convenience; ignoring the difference between engine drone and alternator whine (they sit in different frequency bands and need different treatment); running unstable loads; and forgetting that cable-borne vibration can rattle grip stands or a dolly track. Isolate the plant on rubber mats or a wheeled cart if structure-borne coupling is audible in the room.
If generator noise is unavoidable, schedule dialogue coverage against the quietest engine states and capture a wild line of the loudest line of dialogue immediately after take-lock, so post has a matched-state fallback.
HVAC and Refrigerators: The Steady-Noise Enemies of Dialogue
HVAC and refrigerators produce continuous, insidious noise that the ear tunes out in the room and hears immediately in the cut. Their hums, whirs, and compressor cycles mask dialogue in quiet scenes and are notoriously hard to remove transparently once baked in.
Turn it off. The simplest and most effective solution is to kill nearby HVAC and refrigerator compressors during dialogue takes. Assign fridge-power responsibility to a specific crew member (typically the utility sound tech or a set PA coordinating with the electric department) with two obligations: unplug at the start of the setup, and plug back in with a strip of blue tape flagging the plug at wrap. Fridges get forgotten. Melted craft services is a career-teaching moment nobody wants twice.
Isolation for perishables. If the fridge contains items that cannot be moved (practical cold plate, product placement, catering), swap it to a smaller cooler for the day, or wrap the compressor housing in sound blankets and tent the whole appliance. In a pinch, schedule takes between compressor cycles: most residential fridges run 8 to 12 minute cycles with 4 to 6 minute rest periods, predictable enough to hide behind.
HVAC hierarchy. Central HVAC on a thermostat can usually be switched off at the panel for the duration of a scene. Split systems and PTAC units need to be turned off at the unit. In summer or winter shoots, negotiate with the location owner in advance: 20 minutes of no cooling is a small ask; four hours in a July attic is not, and it changes the schedule. Restore HVAC between setups to keep talent and crew stable.
Location scouting is a sound decision. A location with a modern quiet HVAC system, soft furnishings (carpet, upholstery, curtains), and thick walls will save hours in post. Hard, reflective surfaces (tile, glass, concrete) exacerbate every noise problem by adding reverb. Sound-scout locations before the camera department signs off. See How to Scout Locations for Sound Before Camera Dept Cares for a full checklist.
When you cannot turn it off. Ensure the noise is consistent. Record a full minute of matched room tone with the HVAC running so post has a stable ambient bed. A high-pass filter at 80 to 100 Hz removes low-frequency HVAC rumble without touching dialogue. A narrow notch at the fundamental (often 60 or 120 Hz for compressor tonal peaks) removes hum surgically. iZotope RX Voice De-noise can reduce steady broadband noise conservatively, but is not a rescue tool for aggressive interference.
💡 Pro Tip: In every interior location, record two full minutes of room tone: one with HVAC and fridge running (matched to takes where they were on), and one with them off (matched to protected takes). This gives post maximum flexibility for blending and consistent ambient beds across the cut.
Common mistakes: assuming a fridge is "quiet enough" because it does not stand out in the room; relying on post to fix HVAC without addressing the source (leading to phasey, metallic-sounding dialogue); and using a broad denoiser when a narrow notch would be transparent. If the offending sound is a narrow-band tone, a surgical notch is always more transparent than a broadband reduction.
Traffic and Exterior Ambient Noise: Controlling What You Cannot Turn Off
Traffic, HVAC compressors on adjacent buildings, distant construction, aircraft, and school bells present a different challenge: they are unpredictable and beyond direct control. The emphasis shifts from source control to mic proximity, take timing, and location choice. The cleaner the source audio, the less repair post needs to perform.
Microphone proximity is the master lever. Move the microphone as close to the actor''s mouth as framing and practical constraints allow. Proximity increases signal-to-noise ratio faster than any other single decision. A boom at 30 cm above frame captures dialogue 6 dB hotter than one at 60 cm, and 12 dB hotter than one at 1.2 m, before considering off-axis noise rejection. Lavs (Sanken COS-11D, DPA 4060 series) buried in wardrobe give even more proximity when boom placement is constrained. See Lav Strategy by Wardrobe Type: Suits, Dresses, Athletic Wear, Coats for placement tactics.
Mic pattern by environment. Indoors near traffic, a hypercardioid (Sennheiser MKH 50, DPA 4017C in short position) rejects off-axis room reflection better than a shotgun, which can develop comb-filtering in reverberant interiors. Outdoors, a shotgun (Sennheiser MKH 8060 or 416, DPA 4017B) with proper wind protection (a Rycote Cyclone or Cinela Pianissimo) leverages its directionality to reject flanking noise. Distance destroys the advantage of any directional mic: a shotgun at 1.5 m outperforms itself at 3 m by more than the pattern math suggests, because signal-to-noise falls with the inverse square law.
Take timing. Observe the location''s ambient pattern before rolling. Traffic-heavy streets often have measurable lulls tied to signal cycles a block away (typically 60 to 120 second cycles). Airports have flight paths that concentrate over specific windows. School zones have bell times. The sound mixer should walk the location for 15 minutes before crew call and note the patterns, then coordinate with the 1st AD to time takes accordingly.
Monitor every take. A passing siren, an unmuffled motorcycle, or a garbage truck must be caught in real time. Stopping and resetting is always cleaner than powering through and hoping post can rescue the line. If the noise is short and lands between words, a wild line of the affected line captured immediately (in the same mic position, same actor delivery) is a two-minute insurance policy that saves hours in ADR.
💡 Pro Tip: When shooting near traffic, brief the boom operator on "riding the traffic": adjusting swing timing and mic angles to minimize direct exposure to peak noise events, even if it requires slight compromises in framing coverage.
Common mistakes: relying on a shotgun mic from too far away and expecting post to isolate dialogue; recording usable dialogue but ignoring short transient traffic noises that ruin an otherwise perfect line; and forgetting that the first words after silence are the most vulnerable to aggressive denoising. Test cleanup settings specifically against the first syllables of a line, not the middle.
If traffic is intermittent, a well-timed pause and retake is almost always faster and cleaner than aggressive repair. If ambient state changes unexpectedly (a construction crew arrives across the street mid-scene), capture a fresh wild track of the new ambient. iZotope RX Dialogue Isolate and Fairlight''s Voice Isolate handle the extreme end of exterior cleanup, but their artifacts are audible on quiet syllables. Prevention beats software every time.
Crew Protocol and Set Discipline: The Cheapest Noise Fix Is Behavior
Many "equipment problems" on set are failures of crew protocol. The most effective noise reduction strategy costs nothing more than clear communication and adherence to quiet-set discipline. The best noise is the one that never got created.
Quiet-on-set is a real protocol, not a phrase. No walk-and-talk through active recording zones. No unnecessary chatter at video village. No moving of grip equipment, wardrobe racks, apple boxes, or C-stands during a take. Every department needs to understand its role. On sound-forward productions, this is cultural. On camera-first sets, it takes deliberate leadership from the director or 1st AD.
Assigned accountability. Designate specific noise-monitoring responsibilities:
The sound mixer maintains ultimate responsibility for calling "cut" on sound. Unwavering live monitoring is the sound department''s primary defense and is non-negotiable.
Walkie discipline. Program walkies to a channel matrix (typically Ch 1 production, Ch 2 sound, Ch 3 grip/electric) with clear rules about which channels are active during takes. Sound department walkies should be earpiece-only during takes. Any surface-audible squelch beep near the boom is a NG take.
Announce, listen, confirm. Build the habit into every take: the 1st AD announces the take, the sound mixer listens actively for room noise and confirms with the boom op, and only then gives the "sound speed" or "roll sound" cue. If an intermittent cycle is running (fridge, HVAC, distant train), the mixer holds until the cycle passes.
Alternative clean takes. Even when a take feels good, capture one alternative clean take of the key lines. It costs 90 seconds. It saves the picture editor from having to work around a stray noise months later. Coordinate with the script supervisor so the alt takes are logged in continuity.
💡 Pro Tip: Establish a silent signal (raised fist, closed-hand tap on chest) for crew to flag a noise issue to the mixer without adding a voice to the room. It preserves the take-in-progress until the mixer decides whether to call cut.
Common mistakes: allowing video village chatter, hair touch-ups, or grip work at full volume during dialogue recording; assuming the sound department alone can fix everything in post; failing to communicate when HVAC, generator, or fridge state has changed (creating continuity holes in room tone); and running through egregious noise instead of calling cut. If in doubt, call cut. A 60-second reset is cheaper than an hour of ADR.
See Production Sound Department Roles: Mixer vs Boom Op vs Utility for a full breakdown of who owns what on the sound cart.
Post-Wrap Triage and Cleanup Workflow: What to Do When the Take Is Already Noisy
Despite best on-set efforts, some noise will make it onto the tracks. The consensus among dialogue editors is consistent: start with the cleanest possible source, apply the lightest effective cleanup, and A/B compare at every step. The goal is transparency, not eradication.
Staged workflow. A modular approach outperforms one giant fix. Address noise types in sequence:
1. De-hum (iZotope RX De-hum or Fairlight De-hummer): Notch electrical fundamentals and harmonics first, because they are surgical and cheap. 60 Hz in North America, 50 Hz in Europe and most of the world, plus harmonics at 120/180/240 Hz or 100/150/200 Hz.
At every stage, A/B against the raw file. If the cleanup is more audible than the original noise, dial back. iZotope''s Ambience Match module (RX 10 and later) can rebuild a consistent ambient bed across cleaned regions when room tone is inconsistent.
Room tone beds. After individual line cleanup, fill silences between lines with matched room tone from the wild track. This is how professional dialogue tracks sound cohesive: not silence between lines, but consistent ambient. See Recording Metadata That Matters: Scene/Take, Track Names, Mic IDs for how proper metadata makes room-tone matching fast.
Document everything. Keep a per-scene log of module order, key parameters (noise-profile duration, reduction dB, module presets), and any decisions. Revisions and editor handoffs are dramatically faster with a written trail. If mix notes come back asking for a lighter touch, you can dial back specific stages instead of rebuilding the chain.
💡 Pro Tip: Before rendering cleaned dialogue, listen to it in context with music and sound effects at reference level. Aggressive noise reduction often sounds acceptable in isolation and unnaturally exposed in a full mix, where the missing room texture becomes obvious.
Common mistakes: overprocessing until dialogue sounds synthetic; using a broadband denoiser on a problem that is primarily hum, reverb, or sibilance; failing to A/B against the raw source; failing to document settings; and processing tracks before transcription (noise degrades automated transcription accuracy, so clean first if captions are downstream).
A note on rescue tools. CapCut, Descript Studio Sound, Adobe Enhance Speech, and ElevenLabs Voice Isolator produce impressive results on badly compromised audio, but they are re-synthesis tools: they reconstruct voice character from a model, not the recording. They are useful for social cutdowns and web content, and dangerous for theatrical mixes where the reconstructed voice may lack the room, timbre, and breath detail that establish presence. Use them for rescue, not as a default step.
Interface and Handoff Notes
What you receive (upstream inputs):
What you deliver (downstream outputs):
Top 3 failure modes for this specific topic:
Browse This Cluster
- Production Sound Definitive Guide: Set Recording to Editorial Handoff
Next Steps
Ready to see how this fits into the bigger picture? Start with the complete guide.
📚 Complete Guide: Production Sound Definitive Guide: Set Recording to Editorial Handoff
---
© 2026 BlockReel DAO. All rights reserved. Licensed under CC BY-NC-ND 4.0 • No AI Training.