Noise Reduce Strategies for Urban Buildings

By Priya Nair · February 9, 2026

Noise Reduce Strategies for Urban Buildings

1) Introduction: What you’ll learn and why it matters

Recording or editing audio in urban buildings means dealing with a predictable set of problems: HVAC rumble, elevator motors, traffic wash, distant sirens, hallway chatter, electrical buzz, and room tone that changes every time the building “cycles.” This tutorial teaches a practical workflow for reducing noise without destroying the voice or ambience you actually want. You’ll learn how to identify noise types, choose the right tools (filtering, gating/expansion, spectral repair, and multiband control), and apply them in an order that minimizes artifacts. The goal is not “silent at any cost”—it’s intelligible, natural audio that holds up in a mix, a podcast, a corporate video, or a film dialogue edit.

2) Prerequisites / setup requirements

Audio source: Dialogue, narration, or ambience recorded in an urban building (24-bit preferred). If you only have 16-bit, work carefully to avoid quantization and processing artifacts.
Monitoring: Closed-back headphones plus speakers if possible. Set a consistent monitoring level (e.g., 79–83 dB SPL for speakers in a small room) so you don’t over-process.
Software tools: Any DAW plus at least one noise-reduction option:
- Spectral editor/denoiser (iZotope RX, Acon Digital, Steinberg SpectraLayers, or DAW equivalents)
- EQ with high-pass/low-pass filters and narrow notches
- Expander or gate (downward expander preferred)
- Compressor (optional but common in dialogue chains)
File prep: Duplicate your track (or commit to a new version) before heavy processing. Keep an untouched original for A/B and rescue.
Room tone clip: Find 2–5 seconds of “silence” from the same recording. This is critical for training and matching noise profiles.

3) Step-by-step strategies (in a proven order)

Step 1: Identify the noise type and map it to a tool

Action: Listen and label what you’re hearing before touching any plugins.

Why: Each noise category responds best to a different technique. Using a denoiser on tonal hum often leaves warbling; using a notch on broadband HVAC won’t solve the problem.

What to do: Loop a problem section and classify:
- Low rumble (building vibration, trucks, HVAC duct rumble): mostly below 80 Hz
- Tonal hum (electrical): 50/60 Hz plus harmonics at 100/120, 150/180, etc.
- Broadband noise (HVAC hiss, distant traffic): wide spectrum, often 200 Hz–8 kHz
- Intermittent events (elevator clunks, door slams, sirens): best handled with spectral repair and automation
Technique: Use a spectrum analyzer or spectrogram view. On a spectrogram, hum appears as steady horizontal lines; HVAC is a fuzzy band; sirens sweep diagonally.

Common pitfalls: Treating everything as “noise” and over-reducing. If the “noise” includes room tone you want to keep for continuity, aim for reduction, not removal.
Step 2: Do conservative high-pass filtering to remove subsonic baggage

Action: Add an EQ high-pass filter (HPF) before any denoising.

Why: Subsonic and low-frequency rumble eats headroom, triggers compressors, and makes denoisers work harder than they should. Removing it early improves every downstream step.

Settings to try (dialogue):
- Male voice: HPF at 70–90 Hz, slope 12 dB/oct (start at 80 Hz)
- Female voice: HPF at 90–120 Hz, slope 12 dB/oct (start at 100 Hz)
- If the building is very rumbly: consider 18 dB/oct, but keep the cutoff conservative
Technique: Sweep the cutoff upward until the voice starts thinning, then back off by 10–15 Hz. Compare bypassed vs engaged at equal loudness.

Common pitfalls: Over-high-pass filtering that removes chest resonance and makes speech brittle. Also, doing steep HPF after compression can exaggerate pumping—place it early.

Troubleshooting: If you hear “hollowness,” lower the cutoff or reduce slope. If rumble remains but you can’t raise HPF without damage, plan to use multiband expansion later focused below 120 Hz.
Step 3: Remove tonal hum with notches (before broadband denoise)

Action: Identify hum fundamentals and notch them with a narrow EQ.

Why: Broadband denoisers can leave chorusing artifacts when they try to suppress strong tonal components. Notches remove the “spine” of the noise cleanly.

Settings:
- Set a bell filter at 60 Hz (or 50 Hz depending on region)
- Use Q = 12–20 (narrow), start with -9 dB gain
- Add harmonics: 120/180/240 Hz (or 100/150/200 Hz) as needed, often at -3 to -6 dB
Technique: Solo the track, sweep a narrow boost (+12 dB) around 50–200 Hz to find the exact peaks, then convert boosts into cuts. Turn off the boost immediately after locating—this is a detection move, not a sound.

Common pitfalls: Cutting too wide (low Q) and damaging warmth. Also, forgetting harmonics: removing only 60 Hz often leaves an audible 120 Hz buzz.

Troubleshooting: If hum frequency drifts (common with elevator motors or dimmers), static notches may not hold. Use a spectral repair tool or dynamic EQ keyed to the hum region.
Step 4: Capture room tone and train a denoiser carefully

Action: Select 2–5 seconds of noise-only audio and train your denoiser (or create a noise profile).

Why: Urban building noise is often steady but complex. A trained profile lets the algorithm subtract what’s consistent while preserving transients and consonants.

Settings (starting points):
- Noise reduction amount: 3–6 dB on first pass
- Sensitivity/threshold: moderate (e.g., 2.5–4.0 on tools that use a 0–10 scale)
- Artifact control/smoothing: medium-high (to reduce musical noise)
- If available: enable “Dialogue” or “Voice” mode rather than “General”
Technique: Prefer two gentle passes (e.g., 4 dB + 4 dB) over one aggressive pass (e.g., 10–12 dB). Between passes, reassess; don’t assume more is better.

Common pitfalls: Over-reduction that creates watery, phasey “underwater” artifacts. Training on audio that contains quiet speech or breaths confuses the profile and removes life from the performance.

Troubleshooting: If denoise smears consonants (T, K, S), reduce reduction amount by 1–2 dB and increase smoothing. If noise “pumps” with speech, lower sensitivity or use a multiband expander instead of heavier denoise.
Step 5: Use spectral repair for intermittent building events

Action: Manually attenuate or replace discrete noises (door clicks, elevator dings, chair squeaks) using spectral editing or short crossfades.

Why: No global process can cleanly remove a single loud event without harming the whole track. Targeted repair preserves natural tone.

Techniques and values:
- Spectral Attenuate: select the event region; reduce by 6–15 dB rather than deleting
- Spectral Replace: if the event overlaps speech gaps, replace with surrounding room tone
- Fade strategy: use 5–20 ms fades around edits; longer (30–60 ms) for ambience-only segments
Common pitfalls: Hard deletions that create dead air. Over-attenuation that leaves a noticeable “hole” in the room tone.

Troubleshooting: If your repair sounds obvious, undo and reduce less (try 6–9 dB). Then layer room tone underneath to maintain continuity.
Step 6: Add a downward expander (not a hard gate) to control residual noise

Action: Use gentle expansion to lower noise between phrases while keeping natural decays.

Why: Urban room tone can be acceptable during speech but distracting in pauses. A hard gate often chatters and sounds amateur. Expansion is smoother and more professional.

Starting settings (dialogue):
- Ratio: 1.5:1 to 2.5:1 (start at 2:1)
- Threshold: set so it engages mainly in pauses (often -45 to -30 dBFS, depending on recording level)
- Attack: 5–15 ms (start 10 ms)
- Release: 120–250 ms (start 180 ms)
- Range (if available): limit reduction to 6–12 dB
Common pitfalls: Too fast release causing “breathing” or noise surging up at the end of words. Too much range causing unnatural dead silence.

Troubleshooting: If consonants get clipped, slow the attack (15–25 ms) or lower the threshold. If the noise audibly swells, increase release time or reduce ratio.
Step 7: Control harshness and hiss with dynamic EQ or multiband processing

Action: If HVAC hiss or traffic wash remains in the upper mids, use dynamic EQ instead of static cuts.

Why: Static EQ cuts can make speech dull all the time. Dynamic EQ reduces the problem only when it’s prominent (often during pauses or softer words).

Settings to try:
- Band at 3.5–6 kHz for harsh noise edge; dynamic cut 1–3 dB, Q 1.0–2.0
- Band at 8–12 kHz for hiss; dynamic cut 1–4 dB, Q 0.7–1.2
- Attack 20–40 ms, release 120–200 ms for transparent movement
Common pitfalls: Over-de-essing with broad dynamic cuts that lisp the voice. Too fast timing that creates a “spitty” texture.

Troubleshooting: If the voice loses clarity, reduce the maximum cut or narrow the band. If “S” sounds jump out, use a dedicated de-esser centered around 6–8 kHz with 2–5 dB reduction.
Step 8: Loudness and consistency check (noise reacts to compression)

Action: If you compress dialogue after noise reduction, re-check noise audibility and adjust.

Why: Compression raises low-level details, including remaining noise. Many “my noise reduction didn’t work” complaints are actually “my compressor brought it back.”

Practical approach:
- Dialogue compression starting point: ratio 2:1 to 3:1, attack 15–30 ms, release 80–150 ms, gain reduction 3–6 dB
- After compression, revisit expander threshold (often lower by 2–4 dB)
- Check integrated loudness targets as relevant: podcasts often -16 LUFS stereo / -19 LUFS mono; broadcast varies by region
Common pitfalls: Heavy compression (10+ dB GR) that makes room tone dominate. Fixing this with more denoise usually creates artifacts; better to reduce compression or use automation.

Troubleshooting: If noise blooms between words after compression, increase expander release and reduce compressor makeup gain, then use clip gain to level phrases instead of relying solely on compression.

4) Before and after: what results to expect

Before: Voice competes with steady HVAC and traffic wash; pauses sound loud and “roomy”; occasional elevator tones and hallway bumps pull attention; compression makes everything worse.

After (realistic target):

Steady noise reduced by 6–12 dB without obvious warble
Pauses sit naturally with controlled room tone (not dead silence)
Tonal hum largely removed (fundamental and harmonics suppressed)
Intermittent events attenuated enough to stop distracting the listener
Dialogue remains articulate, with consonants intact and no metallic “swim”

Use level-matched A/B. If the processed version seems “better” only because it’s louder, you’re not evaluating noise reduction—you’re evaluating loudness. Match within 0.5 dB when comparing.

5) Pro tips for taking it further

Build a room tone library: Capture 30–60 seconds of room tone in each location (HVAC on/off if possible). Having clean tone makes repairs and continuity far easier.
Automate denoise amounts: Use less reduction during dense speech (e.g., 3–4 dB) and slightly more during sparse lines (5–7 dB). This reduces artifacts where they’re most noticeable.
Mid/Side strategy for stereo room recordings: Traffic often lives in the Side channel. Apply slightly stronger denoise or high-frequency control to Sides while keeping Mid (voice) cleaner and more natural.
Multiband expansion for rumble: If rumble is variable, expand only the 20–120 Hz band with a gentle ratio (around 1.6:1) and a range of 4–8 dB. This avoids thinning the whole voice.
Prevent the problem on the next recording: Move the mic closer (halving distance improves SNR dramatically), choose a tighter polar pattern, and record away from windows and mechanical rooms. A 10 cm move can matter more than any plugin.

6) Wrap-up: practice the workflow until it’s instinct

Urban building noise reduction is mostly about order and restraint: filter rumble, notch hum, denoise lightly, surgically repair events, then manage pauses with expansion and fine-tune with dynamic EQ. Do it a few times on different real recordings—HVAC-heavy office narration, lobby interviews with traffic wash, apartment dialogue with elevator bleed—and you’ll start hearing which tool to reach for immediately. Keep your originals, take notes on settings that worked, and trust your ears over the meter when the choice is “cleaner” versus “natural.”