How to Reduce HVAC Noise in Classrooms

By Marcus Chen · April 30, 2026

1) Project overview: what, where, who, and why

In late August, two weeks before the start of the fall term, SonusGearFlow was asked to troubleshoot a familiar problem at Brookside Middle School (a three-story, 1970s-era campus in northern New Jersey): newly upgraded classroom AV was “working,” but teachers and remote learners couldn’t clearly understand speech whenever the HVAC system kicked into higher airflow modes.

The district had just completed a summer refresh: new ceiling-mounted speakers, gooseneck podium microphones for hybrid instruction, and a small DSP-based audio system in each of 12 general-purpose classrooms in the B-wing. The integrator’s commissioning notes looked fine on paper, but faculty feedback was consistent: “It sounds like a tunnel,” “my voice disappears,” and “Zoom captions are wrong unless I shut off the AC.” The principal didn’t want to disable HVAC during instruction (thermal comfort and indoor air quality were non-negotiable), and the district’s facilities team had already received complaints about uneven temperatures when teachers manually reduced airflow.

The project team included an audio engineer (lead), the district’s AV project manager, the mechanical contractor who serviced the rooftop units (RTUs), and the controls vendor responsible for the building automation system. The objective was practical: reduce HVAC-related noise at the source where feasible, and then make the classroom audio chain more tolerant of the remaining noise without sacrificing speech naturalness or causing feedback.

2) Challenges and requirements at the outset

The first walkthrough made it clear this wasn’t a single-issue fix. The HVAC noise manifested in three distinct ways:

Broadband airflow noise from high static pressure and undersized return paths (a steady “whoosh” strongest at 250 Hz–2 kHz, exactly where consonant intelligibility lives).
Low-frequency rumble and structure-borne vibration when RTU supply fans ramped up (dominant energy below 125 Hz).
Intermittent tonal components (a narrow peak around 315 Hz in several rooms, likely duct resonance or diffuser whistling at certain damper positions).

Constraints and requirements were equally specific:

Schedule: 10 business days total, with only two after-hours windows (3 pm–11 pm) to open ceilings and adjust dampers without disrupting staff.
Budget: $38,000 not including any major RTU replacement. Minor mechanical modifications were possible, but not a re-ducting project.
Performance target: Improve speech intelligibility and reduce HVAC noise to a level consistent with classroom guidelines. Practically, we aimed for NC 30–35 during occupied HVAC modes and STI ≥ 0.60 at student seating positions with the teacher at the podium mic.
Existing AV: Each room had a small DSP amplifier (Q-SYS Core Nano in a wall rack), one gooseneck condenser microphone at the teacher station, and four ceiling speakers. No ceiling array mics were installed, and the district did not want to add bodypacks.
Operational reality: Teachers needed “push one button and teach.” No elaborate gain staging procedures, and no expectation that teachers would mute HVAC or change modes.

3) Approach and methodology chosen

We treated the problem as a combined mechanical-acoustical-DSP issue and used a simple decision framework:

Measure first: quantify baseline noise (SPL spectrum and NC), then quantify speech metrics (STI/ALcons) before changing anything.
Fix what the DSP cannot: remove tonal whistle, reduce turbulence, and address obvious vibration paths. DSP can help with residual steady noise, but it cannot “restore” consonants masked by excessive broadband noise without making the voice thin or fatiguing.
Optimize the signal chain last: once the room is quieter, tune mic EQ, gating behavior, and loudspeaker response for stable gain-before-feedback and intelligibility.

Instrumentation included a calibrated measurement mic (Earthworks M30), an audio interface, and Room EQ Wizard for spectral capture. For noise criteria, we used a portable SPL meter with 1/3-octave logging and NC curve overlay. STI measurements were performed using a handheld STI meter at three listener positions per room (front-center, mid-left, back-right).

4) Step-by-step execution narrative

Day 1–2: Baseline measurements and triage

We measured 12 rooms over two afternoons, capturing three HVAC states: fan low, fan high, and compressor on with fan high. In 9 of 12 rooms, the dominant issue was fan high. Baseline results were consistent:

Noise criteria: NC 42–48 in fan high; best rooms were NC 38.
SPL: 48–54 dBA steady-state in occupied mode.
STI: 0.44–0.55 (teacher at podium mic, typical speaking level). Remote transcript accuracy issues made sense at these values.

Two rooms were notably worse: Room B-214 and B-219 showed a sharp 315 Hz component 8–10 dB above adjacent bands, and audible “chuffing” at the diffuser when the VAV damper moved. Those were flagged for mechanical attention first because tonal noise is the easiest to perceive and the easiest to fix when caused by a specific restriction.

Day 3–5: Mechanical fixes with the highest ROI

We worked with the mechanical contractor during after-hours. The most effective interventions were not exotic; they were targeted and verified by measurement after each change.

(a) Diffuser and damper corrections
In five rooms, diffuser face velocity was clearly too high. The VAV boxes were delivering more airflow than required for the actual classroom load, likely a legacy setting from pre-renovation occupancy assumptions. The controls vendor reduced minimum airflow setpoints during occupied mode by 15–25% (room-dependent) while ensuring CO₂ and temperature targets were still met. For two “whistling” rooms, we also corrected partially closed balancing dampers upstream that were acting like nozzles. After adjustments, the 315 Hz peak dropped by 6–9 dB.

(b) Return air path improvements
Several classrooms had return grilles installed but poor return paths above the lay-in ceiling due to cable trays and newly added backboxes. This increased static pressure and contributed to the broadband “whoosh.” We added two short return path baffles (simple lined sheet metal boots) to open up the plenum return in four rooms where the ceiling space was constricted. This did not require new duct runs—just re-establishing an unobstructed path. The audible change was immediate, and NC dropped by 3–4 points in those rooms.

(c) Vibration isolation touch-ups
One RTU serving the B-wing had a supply fan that transmitted low-frequency rumble into the structure. The mechanical team tightened loose hardware on the curb and replaced two deteriorated neoprene isolators. We also ensured flexible duct connectors weren’t overly taut (which can short-circuit isolation). This didn’t dramatically change dBA, but it reduced the felt rumble and cleaned up the sub-125 Hz band by ~3 dB in affected rooms.

Day 6–7: Acoustic and DSP tuning in representative rooms

With the worst mechanical offenders addressed, we moved to AV tuning. Rather than tune all 12 rooms blindly, we selected three representative rooms (quietest, median, noisiest post-fix) and developed a repeatable DSP template.

Microphone gain structure
The gooseneck mics were set too “hot” at the preamp stage, which made the DSP gate behavior erratic and increased the audibility of HVAC noise during pauses. We standardized input gain so that normal speech hit -18 to -12 dBFS on the DSP meter, leaving headroom for loud speech without hard limiting.

EQ strategy
We avoided aggressive high-pass filtering that makes speech thin. Final settings were modest and consistent:

High-pass filter at 100 Hz, 12 dB/oct to reduce rumble and handling noise.
Two narrow cuts: -3 dB at 315 Hz (Q≈6) in the two rooms where a remnant resonance persisted, and -2 dB at 630 Hz (Q≈4) where a boxy coloration appeared.
Presence shaping: +2 dB shelf starting at 2.5 kHz in rooms that sounded dull after mechanical noise reduction, applied carefully to avoid feedback.

Noise gating vs. expander
Hard gates were causing “pumping” as the HVAC noise opened the mic during teacher pauses. We replaced gates with a gentle downward expander: ratio 1.8:1, threshold set about 6 dB above the measured HVAC noise floor at the mic, attack 10 ms, release 250 ms. This reduced room noise without chopping word endings. In Q-SYS, we implemented this using a dynamics block with expander settings rather than a traditional gate.

Automixing decisions
Because each room had only one podium mic, automixing wasn’t a factor. However, we did add an adaptive AEC reference check for the conferencing feed: the AEC reference level was inconsistent, which made remote listeners hear more room noise. We standardized loudspeaker send levels into the AEC reference and verified ERLE improvements (not a lab-grade test, but enough to confirm stability). This reduced the “hollow” effect that teachers described on calls.

Day 8–9: Rollout to remaining rooms and verification

We cloned the DSP template to all rooms, then performed quick per-room verification:

Confirm input gain calibration with a teacher-voice playback reference and live voice check.
Run a 60-second HVAC-on measurement at the podium mic position.
Check for any remaining tonal peaks and apply only minimal additional EQ (no more than two filters per room beyond the template).

Finally, we repeated STI measurements in all 12 rooms with HVAC in occupied high mode, since that was the problem condition.

5) Technical decisions and trade-offs made

Mechanical reduction vs. DSP masking
The budget could have been spent entirely on “audio fixes” (more microphones, more processing), but we pushed for mechanical corrections first. The trade-off was coordination complexity: mechanical, controls, and AV teams had to agree on changes and verify that airflow requirements were still met. The payoff was that every downstream audio decision became easier once the noise floor dropped.

Lower airflow setpoints vs. IAQ safety margin
Reducing minimum airflow helped noise, but we could not compromise ventilation. We negotiated conservative reductions: we lowered minimums in rooms with historically cold complaints and verified CO₂ stayed below 900 ppm during a simulated occupied period (facilities had trend data). In two rooms with poor air mixing, we reduced less and accepted a slightly higher NC rather than risk comfort issues.

Expander instead of gate
A gate can make a room feel quieter between sentences, but it is unforgiving with HVAC noise. The expander approach was a deliberate trade-off: it doesn’t “black out” the room tone, but it preserves syllables and sounds more natural to both in-room students and remote listeners.

EQ restraint
There was pressure to “EQ out the HVAC.” We avoided broad midrange cuts that would have improved measured noise audibility at the cost of intelligibility. The guiding rule: use EQ to correct coloration and resonance, not to fight broadband noise.

6) Results and outcomes with specific details

Post-project measurements showed meaningful improvements.

Noise criteria: improved from NC 42–48 down to NC 33–38 in 10 rooms; two rooms landed at NC 39 due to higher required airflow and limited return path options without construction.
Steady-state level: dropped by 6–10 dBA in the median room during fan high mode.
Tonal component reduction: the 315 Hz peak in B-214 and B-219 fell by 8 dB after damper and diffuser corrections, eliminating the “whistle” complaint.
STI: improved from 0.44–0.55 to 0.61–0.71 depending on room and listener position. The back-right position, previously the worst, saw the largest gains because reduced noise improved effective SNR.
Conferencing quality: teachers reported fewer instances of remote participants asking for repeats. While subjective, the district’s helpdesk logged a drop from 8–10 weekly tickets related to “audio can’t hear teacher” to 2–3 in the first month.

Timeline and cost were kept within constraints. Mechanical and controls work totaled $21,400 (labor, return baffle materials, isolators, controls programming). AV engineering and commissioning across 12 rooms totaled $14,600. The remaining budget covered measurement time, after-hours access, and contingency.

7) Lessons learned and what could be done differently

1) Measure HVAC noise at the microphone, not just in the room
Room NC is useful, but the effective SNR at the mic is what drives conferencing and voice lift results. In two rooms, podium placement was directly under a supply diffuser; simply rotating the teacher station 6 feet reduced the noise at the mic by ~4 dB with no mechanical changes. We did this in one room late in the project and wished we had evaluated furniture/mic placement earlier.

2) Controls programming is part of the audio solution
The biggest single improvement came from adjusting minimum airflow and smoothing fan ramps. Audio teams sometimes stop at “that’s mechanical,” but VAV and RTU control logic has direct acoustic consequences. Future projects should include the controls vendor in the initial AV scope meeting.

3) Avoid commissioning DSP before the building settles
The integrator’s original settings were done when HVAC was still being balanced. As airflow changed, the noise floor changed, and the mic dynamics settings became inappropriate. In a perfect world, final audio commissioning happens after mechanical balancing is signed off, or at least with a scheduled return visit.

4) Some fixes require construction-level decisions
Two rooms couldn’t reach NC 35 without more significant return duct modifications. If the district renovates again, the highest-value architectural move would be to improve return sizing and reduce plenum bottlenecks, rather than adding more audio processing.

8) Takeaways applicable to other projects

Start with the noise source: Identify whether the problem is turbulence, resonance, or vibration. Each has a different fix, and only some are solvable in DSP.
Set targets you can verify: NC 30–35 and STI ≥ 0.60 are measurable and meaningful for classrooms. Use consistent HVAC states during testing so comparisons are valid.
Coordinate with facilities early: Airflow setpoints, damper positions, and fan ramp profiles can change noise by 5–10 dB. That’s often more impactful than swapping microphones.
Use gentle dynamics: In steady-noise environments, downward expansion usually yields more natural results than hard gating, especially for hybrid learning.
Keep EQ surgical: Cut resonances and correct coloration; don’t carve out midrange trying to “remove” HVAC. Intelligibility depends on that band.
Don’t ignore placement: A podium mic under a diffuser is a built-in SNR penalty. Sometimes the simplest fix is moving the mic and re-aiming a speaker zone.

By treating HVAC noise as a shared mechanical-and-audio responsibility—and by validating each change with measurements—Brookside’s classrooms ended up quieter, more intelligible, and easier for teachers to use. The most replicable insight from this project is that “HVAC noise” is rarely one problem; once you separate it into airflow, resonance, and vibration, the path to a predictable improvement becomes much clearer.