How to reduce traffic and café noise in a recording
Start with a short section containing speech and a pause. Decide whether the problem is low rumble, changing street sound or another voice, then compare one cleanup pass with the original. For the next take, test a quieter position and bring the microphone closer.
Street interview: save the sentence, not just the pause
Before a long answer, record a short question and answer at the intended position. Play back the words during a passing vehicle—not just the quiet moment after it. If a name or number disappears, move or wait before recording the full interview.
While recording
Keep a short list of interrupted sentences. After the car passes, repeat the full sentence from its beginning. Leave a little room sound before and after the retake so the edit has space.
When editing
A car between sentences can often be shortened or replaced with nearby room tone and a short crossfade. A car over a word needs a usable alternate take or careful cleanup. Muting that moment would mute the word too.
Café conversation: make each voice easier to capture
Start with the seat, not a stronger cleanup setting. Compare a table away from the loudspeaker, queue and coffee machine. Keep the person you are recording close to the microphone rather than placing the recorder halfway toward the loudest background.
| Situation | Recording choice | Check before continuing |
|---|---|---|
| One speaker | Place the mic for that person and keep the speaking distance steady. | Can you follow a quiet sentence over the nearby conversation? |
| Two people sharing one mic | Test both seats. The best position for one may leave the other too distant. | Compare both answers at normal playback volume. Reposition before a long interview. |
| Two available microphones | Record each person on a separate track if the recorder supports it. | Check each track alone for clothing noise, overload and background. Separate tracks permit individual edits; they do not prevent acoustic spill. |
Background conversation is especially awkward when it covers the same syllables as the main speaker. A quieter pause after processing does not establish that those syllables were recovered. Keep a note of unclear names, dates and numbers, and repeat them while still at the location.
Find the part that needs fixing
- Save an untouched copy. Choose 10–20 seconds with a complete sentence, a natural pause and the loudest nearby interruption.
- Listen once for the background, then again for the words. Note the exact word covered by a car, cup or nearby conversation.
- Try Noise Removal on that section. Compare at a similar speaking volume; a quieter overall file can seem cleaner simply because everything is quieter.
- Keep the result only if the sentence remains clear. Check beginnings, quiet endings and the moment when the interruption happens before processing the full recording.
Match the sound to the next step
| What you hear | What to try | Check afterwards |
|---|---|---|
| A low, fairly steady road rumble | Test a gentle high-pass filter in your recorder or editor, or compare a noise-removal pass. | Lower voice tones should still have body. |
| Cars swell in and out during a sentence | Compare cleanup at the loudest pass-by, not just in a quiet pause. | Every word should remain understandable through that change. |
| Cups, dishes or a sudden clatter | If it falls between words, try a small local edit. If it covers a word, compare cleanup with a replacement take. | The edit should not clip a consonant or create a sharp jump in ambience. |
| Another person speaks over your subject | Test the overlap itself. Move or repeat the sentence if the voices remain hard to separate. | Do not judge success from silence between sentences. |
A high-pass filter can reduce low rumble, but raising its cutoff can also thin the voice. A noise gate mainly lowers quiet sections; it is not a way to remove a passing car while the gate is open for speech. When the background stays constant rather than changing with passing traffic or conversation, distinguish fan or air-conditioning noise from a pitched electrical hum before choosing a repair.
Compare traffic masking with competing café voices
These controlled comparisons combine real narration with separately recorded street or café sounds. Each after clip is the actual Noise Removal result from the same mixture. Timing and voice levels are matched for comparison; the narrators were not recorded in those locations.
In the charts, time runs left to right and frequency runs bottom to top. Yellow shows stronger sound, which can be voice or noise. Both panels use the same scales. Follow the same moment in each panel and listen before deciding whether less colour means a useful improvement.
1. Nolan Fout — Passing cars behind narration.
Inspect the before-and-after spectrogram
2. number6 — Louder traffic behind another voice.
Inspect the before-and-after spectrogram
3. Rachel Nelson-Smith — Café chatter and table sounds.
Inspect the before-and-after spectrogram
4. Heather Hogan — Louder café background.
Inspect the before-and-after spectrogram
The opening pauses became quieter in all four examples. That measurement describes these short pauses only; it does not prove that all background sound disappeared during speech or that every voice sounds better. Compare the words and the background together.
Improve the next recording before editing
Make two short takes with the same sentence. First use your current setup; then move the microphone closer while keeping speech comfortable and avoiding overload. Compare at similar playback volume. Placement near the intended voice and away from unwanted sources is a useful starting point.
- Near a road: compare a room farther from the street, a closed window or a quieter time. Record the same pause as well as the same sentence.
- In a café: test a seat away from the coffee machine, dish station and nearby conversations before starting the interview. Keep the microphone close to the person speaking.
- For an interruption: pause, let it pass and repeat the whole phrase. A complete alternate phrase gives you more room to make a natural edit.
- Keep consistent microphone distance when turning to another person. Check the actual selected input if moving the external microphone seems to change nothing.
Know when cleanup is enough
Stop when the speech is easy to follow and the remaining background is less distracting than the changes caused by more processing. Listen for watery texture, broken word endings or background sound that rises and falls with every syllable. Compare against the untouched original, not just the previous processed version.
If you use a traditional noise-profile effect in an editor, follow its intended use. Audacity recommends its Noise Reduction effect for fairly constant noise and describes limits with irregular traffic and audience sounds. A profile taken from one quiet moment is not evidence that every changing sound can be removed.
If a key word is still impossible to understand, use another take or re-record the sentence when possible. Turning down a file or repeatedly cleaning it is not proof that the missing detail has returned.
Recording credits
Narration: Nolan Fout, number6, Rachel Nelson-Smith and Heather Hogan, from LibriSpeech, © 2014 Vassil Panayotov, CC BY 4.0. The excerpts were resampled, level-adjusted and mixed with the backgrounds below; the after versions were processed and aligned.
Backgrounds: “00 - Car_across.wav” by semenov_nick and “Cafe ambient sound.mp3” by evsecrets, both CC0. Excerpts use the public MP3 previews, adjusted in level and mixed for these comparisons.