Why street traffic noise is harder than it sounds
Street traffic is not one noise. It is a low engine rumble around 40–120 Hz, tire hiss that spreads from 500 Hz to 8 kHz, intermittent horns and sirens, and the sharp clatter of a truck bed or bus air brakes. Your voice lives in the same space: male fundamentals around 85–180 Hz, female fundamentals around 165–255 Hz, and consonants from 2–8 kHz. That overlap is why a simple noise gate or a single EQ notch rarely fixes the problem. A gate may silence gaps, but it does not remove the rumble under your words. A notch can cut a low hum, but it cannot chase a motorcycle passing mid-sentence. You need a layered cleanup, and you need to accept that some traffic sounds are easier to remove than others.
Step 0: Decide whether cleanup is worth it
Before you spend an hour on processing, listen on headphones and answer three questions. Is the traffic constant, like a highway hum, or intermittent, like a car door slam? Does it sit under your voice, or does it overlap and mask words? Is the recording for a client, a podcast, or a quick voice memo? Constant low-level traffic is usually treatable. A loud siren blaring over an important sentence is often not. If you can re-record the line, do that first.
Step 1: Grab a noise-only sample
If your recording has even two seconds where nobody is talking, use it. In Audacity, select that traffic-only section, then go to Effect > Noise Reduction and click Get Noise Profile. In Adobe Audition, select the sample and use Effects > Noise Reduction / Restoration > Capture Noise Print. If your file starts with room tone or a count-in, that can work too. The tradeoff: if your “noise only” sample contains a breath, a chair creak, or music, the profile will teach the tool the wrong thing. Find the cleanest slice possible.
Step 2: High-pass filter the low rumble
Most traffic rumble sits below 100 Hz. A high-pass filter is the safest first move because it removes energy your voice does not need. In Audacity, use Effect > High-Pass Filter. For a deep male voice, set 80–90 Hz with a 12 dB per octave slope. For a female or lighter voice, try 100–120 Hz. In a DAW, use a standard EQ with a high-pass at 100 Hz, 24 dB per octave. Listen carefully: if the voice starts to sound thin or telephone-like, lower the cutoff. This step alone can reduce bus engines, distant trucks, and air conditioning rumble without touching the voice.
Step 3: Apply gentle broadband noise reduction
Now use the noise profile you captured. In Audacity, open Effect > Noise Reduction. Start with Reduction at 6–9 dB, Sensitivity at 6.00, and Frequency smoothing at 3. Do not jump to 20 dB. A single aggressive pass creates metallic, underwater artifacts that are more distracting than traffic. If the traffic is still audible, run a second pass at 4–6 dB instead of one huge pass. In Audition, use the Noise Reduction process with 30–50% reduction and high FFT size for steady hum. In RX, use Spectral De-noise with adaptive mode. The key tradeoff: more reduction removes more traffic, but it also removes breath, room tone, and natural voice texture. Stop when the artifacts become noticeable.
Step 4: Use AI voice isolation for traffic that comes and goes
Traditional denoise is built for constant noises: fans, hiss, hum. Traffic is dynamic. A car passes, a motorcycle revs, a horn blares, then the street goes quiet. AI voice isolation is trained to separate speech from complex background noise, so it can handle those passing events with less manual automation. For a quick option, upload your file to a free online tool like ClearVoiceHQ (clearvoicehq.com). It is designed for specific background noises, including street traffic, and it works on voice memos, Zoom call recordings, podcast clips, and extracted video audio. Export a WAV at 48 kHz if you can, process the file, then A/B it against the original. If the result sounds over-processed, try a lower intensity setting or process only the noisy sections. Keep the original file untouched.
Step 5: Manually repair horns, sirens, and door slams
AI and denoise will not always remove a sharp horn or siren without side effects. For those moments, use spectral repair. In Audacity, select a tiny region around the horn—about 0.2 to 0.5 seconds—then use Effect > Spectral Delete. In RX, use Spectral Repair to attenuate or replace the event. If the noise happens in a pause, a simple fade or mute is cleaner than processing. If a siren overlaps a word, you have three options: re-record that word, cut around it, or accept a small artifact. Do not try to remove a loud siren from the middle of a sentence unless you are prepared to spend serious time in spectral editing.
Step 6: EQ and de-ess after cleanup
After denoising, your voice may sound dull or harsh. Use a gentle EQ to restore intelligibility. Cut 200–400 Hz by 1–2 dB if the voice sounds muddy. Add a subtle presence boost at 2–5 kHz, no more than 2–3 dB. If “s” and “sh” sounds become piercing, use a de-esser around 6–9 kHz. Be conservative. The goal is to make the voice sound natural, not to add a radio-style smile curve. Over-EQ after noise reduction often makes artifacts more obvious.
Step 7: Gate, compress, and limit
Compression does not remove noise, but it makes your voice more consistent. Use a light compressor: ratio 3:1, threshold around -18 dB, attack 5–10 ms, release 100–200 ms, and 3–6 dB of makeup gain. If traffic noise rises in the gaps between phrases, add a gentle expander or gate after the denoiser. Set the threshold just below your quietest voice, ratio 2:1 for an expander, or use a gate with 5–10 ms attack and 100–200 ms release. Avoid heavy gating before denoising because it chops room tone and makes the remaining traffic seem louder. Finish with a limiter at -1 dB true peak.
Settings cheat sheet by recording type
- Phone voice memo: high-pass 100 Hz, noise reduction 6–8 dB, then AI isolation if traffic is intermittent.
- Zoom call: use the original audio track if possible. High-pass 120 Hz, noise reduction 5–8 dB. Zoom’s built-in suppression can make voices choppy, so compare before committing.
- Podcast mic with street outside: high-pass 80 Hz, noise reduction 6 dB, subtle gate, EQ, and AI isolation for passing vehicles.
- Gaming clip: separate mic and game audio in OBS. Process only the mic track. High-pass 100 Hz, gate, and light denoise.
What not to do
- Do not crank noise reduction to 20+ dB. The watery artifacts will be worse than the traffic.
- Do not use an aggressive gate before denoising. It creates unnatural silence and highlights noise.
- Do not rely on one click. Layer high-pass, gentle denoise, AI isolation, and manual repair.
- Do not process the same file repeatedly and save over the original. Always keep a backup.
If the traffic still wins
Sometimes the recording is too far gone. If you can re-record, move closer to the mic, use a dynamic microphone, and record in a closet or under a blanket fort. Close windows, turn off fans, and try early morning or late night. For future recordings, enable a high-pass filter on your recorder, use a lavalier with a windshield, and record a backup track if possible. A free tool like ClearVoice can help, but it cannot rebuild words that were completely masked by a siren. Set realistic expectations, process in small steps, and always check the final file on phone speakers, laptop speakers, and earbuds before you publish.