Decibel Meter

How the Decibel Meter works

A decibel is a ratio, not a quantity, and that single fact explains why browser sound meters are so often misunderstood. What a microphone hands a web page is a stream of sample values between −1 and +1. The only honest reading a page can give from that is dBFS, decibels relative to full scale, where 0 dB is the loudest signal the input can carry without clipping. Every real measurement is therefore a negative number, and that is what this meter shows by default.

Why the scale is logarithmic. Human hearing spans a pressure range of roughly a million to one, from the threshold of hearing to the threshold of pain. A logarithmic scale squeezes that range into 0–120 dB SPL. It also brings a few rules of thumb worth knowing. +6 dB doubles the sample amplitude. +3 dB doubles the power, so two identical fans are 3 dB louder than one, not twice as loud. +10 dB is heard as roughly twice as loud.

Where levels matter. Workplace noise rules are written in decibels. The UK and EU Noise at Work rules require action at a daily exposure of 80 dB(A), hearing protection at 85, and set an absolute limit of 87. OSHA in the US uses 90 dBA over eight hours with a 5 dB exchange rate, while NIOSH recommends 85 with a 3 dB exchange, meaning that every 3 dB louder halves the safe time. The WHO's guideline for night-time bedroom noise is around 30 dB(A). Broadcasters normalise loudness to −23 LUFS (EBU R128) or −24 (ATSC A/85), and streaming services to around −14. For recording, engineers aim for speech or instruments averaging about −18 dBFS with peaks under −6, which leaves room for surprises.

RMS versus peak. The RMS figure is the root mean square of the samples in the current analysis window. It tracks how loud something sounds, rising and falling with the energy of the signal rather than with individual spikes. The peak figure is the largest single sample in the window, and it is what matters for clipping. A snare hit or a slammed door can peak 15 dB above the RMS level without changing the apparent loudness much at all.

The meter response. Real sound level meters specify a time weighting because the answer depends on it. Fast (125 ms) follows transients closely. Use it for setting a recording level or hunting down an intermittent rattle. Slow (1 s) averages over about a second, which suits a room's steady background level, where a fast meter jitters too much to read. Both are applied as smoothing to the same underlying measurement, so you can switch mid-reading and watch the same signal settle differently.

How to read the decibel meter

• Level is a bar for glancing at from across the room.
• RMS is the smoothed average level in dBFS. Quiet rooms on a laptop microphone often sit around −60 to −50. Normal speech at arm's length is usually around −35 to −25.
• Peak now is the loudest sample in the current window.
• Peak held keeps the loudest value since you pressed Play. Stop and start again to reset it.
• Headroom is how far the held peak sits below 0 dBFS. At 0.0 the input clipped, and anything recorded at that moment is distorted. Under 3 dB is too close for a recording.
• Estimated SPL stays switched off until you enter a calibration offset.
• Sample rate confirms what the audio device is running at, usually 44,100 or 48,000 Hz.

Why SPL needs calibration. Turning dBFS into dB SPL, the absolute pressure level that regulations and safety guidance use, requires the sensitivity of the microphone and the gain of everything between it and the browser. A page cannot discover any of that. To calibrate, measure a steady source with a real meter or a calibrated phone app, note the difference from the RMS figure here, and enter it as the calibration offset. From then on the estimate is only as good as the reference you used, and it holds only for that device at that input gain. The reading is unweighted, so it will read higher than an A-weighted dB(A) meter on rumbly sources such as traffic or ventilation, because A-weighting discounts low frequencies the way the ear does.

Tips and pitfalls.

• Keep the microphone in the same spot when comparing, because level falls about 6 dB every time you double the distance from a small source.
• Laptop and phone microphones roll off deep bass heavily and will under-read a subwoofer or a lorry idling outside.
• Bluetooth headsets switch to a low-quality call mode when their microphone is in use, which changes the level. Use the built-in microphone instead.
• The page asks the browser to disable echo cancellation, noise suppression and automatic gain control, because all three rewrite the level, which is the one thing a meter must leave alone. Some systems enforce their own processing anyway; if quiet rooms and loud rooms read almost the same, that is the cause.

What this is and is not. It is a genuinely useful relative meter. You can compare two rooms, find which appliance dominates, check that a recording input is not clipping, or confirm that a fan got quieter after a fix. It is not a Class 1 or Class 2 instrument to IEC 61672, and it should not be used for any measurement with a legal or occupational-health consequence.

There is no export, because a live meter has nothing to save. Stopping the meter releases the microphone immediately; no buffer is kept and no file is written. To clean up a recording you have measured, see how to clean up a voice recording.

Related tools