What street noise is made of
A street is not one noise but several layers, and they do not behave the same way:
| Layer | What it sounds like | Removed? |
|---|---|---|
| City rumble | A low, continuous roar of engines and tyres, near and far | Yes |
| Passing vehicles | A noise that swells and fades over a few seconds; scooters and motorbikes add a rasp | Yes |
| Construction | A deep rumble, plus impacts from drills and hammers | The rumble, yes; loud impacts were not part of our test |
| Horns and sirens | Loud tones with a pitch | Not measured: compare with the Original/Cleaned switch |
| People and music | Chatter, calls, a shop’s radio | Not reliably |
What the app removes, and what it keeps
With Noise Remover, the rumble and the passing traffic are both treated as noise. The 80 Hz low-cut filter takes the deepest part of the roar, and the neural model, which keeps updating its estimate rather than learning the noise once, follows a car as it swells and fades instead of waiting for it to settle.
In our lab test at full intensity, traffic (a city rumble with passing cars) dropped by 24 to 37 dB in the speaker’s pauses, and a low rumble on its own by 25 to 34 dB. On traffic, the voice’s own level changed by just 0.1 to 0.3 dB, even with the street as loud as the speaker, which puts it among the gentlest noises on the voice that we measured. How we measured.
Part of the reason lies in where the noise sits. A city’s roar weighs most in the low frequencies, while the consonants that make words intelligible live higher, roughly between 1 and 4 kHz. With the two that far apart, the roar can come down a long way before the words start to blur.
| Noise | Noise removed during pauses, in dB | Change in voice level |
|---|---|---|
| Traffic | 24 to 37 dB | −0.1 to −0.3 dB |
| Low rumble | 25 to 34 dB | −0.3 to −1.0 dB |
What stays: people talking around you
Crowd chatter, a market, a café terrace: these are voices, and the model is built to keep voices. It lowers them only a little, and when they overlap your words it can turn your voice down a little too. In our lab test, four people talking at once in the background dropped by only 5 to 11 dB, and the speaker’s voice lost 0.3 to 2.8 dB. Music from a shop or a car radio is outside what the model is made for: it may come down, but not cleanly.
Settings for a street recording
- Start at the default 80%. Traffic costs the voice so little that going higher carries little risk: at 100%, passing cars fade further between sentences.
- Keep the low-cut filter on for the deepest part of the engine roar.
- Trim the start and end, where pressing record and stop often catches handling noise and the first passing car.
- Leave the ready-to-publish level on, so a voice filmed at arm’s length in a loud street is evened out toward −16 LUFS like the rest of your clips.
Clean your street recording on the phone, offline.
Free · Android 8.0 or later, 64-bit
Recording in the street: vlogs and interviews
Noise removal does more with a recording that starts out good. In a street, that comes down to distance and timing:
- Bring the microphone close. Hold the phone 15 to 20 cm from your mouth, or clip a lavalier to your chest: the voice then rises well above the traffic.
- Choose the quieter side. A side street, a courtyard or a square a block away can be far calmer than the avenue.
- Wait for the gap. Traffic comes in waves at lights and junctions; start your sentence after the bus has passed, not as it arrives.
- Stay clear of idling engines, generators and air vents, which blast noise straight at the microphone from close by.
- Turn your back to the road. Your body shields the phone’s microphone from part of the noise.
For a street interview, hold the phone near whoever is speaking and move it as the conversation turns: speakers taking turns clean well, two people talking at once do not. More in cleaning interviews.
Filming the street? Import the video and only its soundtrack is cleaned; remove noise from a video shows how to put the cleaned audio back under the picture.
With Pro, the cleaned soundtrack goes straight back into an MP4 of your clip, ready to post.