How Loud Is a Subway Train? Decibel Levels Underground

The strongest dataset is a study of 243 measurements taken across the New York City transit system in 2007–08, published in the American Journal of Public Health. Researchers wore dosimeters while riding and waiting, which captures what a passenger receives rather than what a fixed microphone at a set distance would record.

Read the averages and the maxima together. An average equivalent level in the low 80s dBA is comparable to a busy road; a maximum of 102.1 dBA on the loudest platform is a different order of experience, and it is the arrivals and braking that produce it.

The New York data separates local stations from major hubs and finds a 3.2 dB difference in mean level (79.0 versus 82.2 dBA). More trains, more platforms and more people all contribute, and 3 dB is a doubling of sound energy.

Track geometry matters too. A study in The Laryngoscope found that stations on curved track measured higher than those on straight track — the mechanism being wheel-rail curve squeal. We could not extract that study's numeric values, so the finding is reported here as a direction, not a figure.

A survey of all 15 Seoul metro lines used a Class 1 sound level meter near passenger seating at peak commuter times, and reported unweighted (Z-weighted) levels: mean 72.78 ± 3.62 dB LZeq across lines, ranging from 62.46 to 78.34 dB, with maximum levels averaging 87.26 ± 3.73 dB and the loudest reaching 91.94 dB.

It is tempting to place these numbers alongside the New York figures. Do not. Z-weighting includes low-frequency energy that A-weighting discounts heavily, and rail vehicles have plenty of it, so unweighted levels are systematically different. Two honest studies, two metrics, no direct comparison.

The New York study framed its results against occupational criteria, which is the right instinct but requires care: an occupational limit assumes an 8-hour day, and a commute is typically under an hour each way.

For reference only: NIOSH recommends 85 dBA over 8 hours with a 3 dB exchange rate; OSHA's permissible exposure limit is 90 dBA over 8 hours with a 5 dB exchange rate and an 85 dBA action level. A mean platform level of 83.5 dBA sits below the NIOSH reference, while the loudest platform's 90.2 dBA average does not. We do not calculate personal exposure from any of these figures.

Three mechanisms dominate. Wheel-rail interaction, especially curve squeal, produces the sharp high-frequency peaks that drive maximum levels. Braking adds another transient. And the tunnel itself, with hard surfaces on all sides, provides very little absorption, so sound arrives multiple times.

That last point explains why underground platforms measure higher than the all-platform average in the New York data: an open-air platform loses energy upward that a tunnel returns to you.

The published studies used dosimeters and Class 1 meters. A phone or browser is not that, and train arrivals are exactly the kind of transient that consumer microphone chains handle worst.

Our meter is useful here for relative questions on one device: does this platform read higher than that one, does moving away from the platform edge help, is the train interior quieter than the platform. Uncalibrated it shows A-weighted signal level in dBFS rather than sound pressure; calibrated once against a trusted meter it gives an estimated dBA figure.