RMS & Peak Metering

A good level meter shows two numbers because it's answering two different questions. One tells you how loud your voice will feel to a listener; the other tells you how close you are to the hard ceiling where digital audio breaks. Confuse the two and you'll either sound faint or you'll clip — usually in the same week.

Two numbers, two questions

The peak reading answers a purely technical question: what is the highest instantaneous value the signal has hit? In a digital system that number matters because there is an absolute ceiling — 0 dBFS, decibels relative to full scale — above which the waveform simply cannot be represented. A single sample touching the ceiling is a fact, not an opinion, and the peak meter reports it.

The RMS reading answers a perceptual question: how loud does this signal feel? Human hearing doesn't respond to instantaneous peaks; it integrates acoustic power over time. A short spike barely registers, while a sustained tone at the same peak level sounds far louder. RMS — root mean square — is the classic engineering estimate of that sustained power, which is why it tracks perceived loudness far better than peak does. A voice with healthy RMS and controlled peaks sounds strong; a voice with high peaks and low RMS sounds thin and distant, even though "the meter hit the top."

From swinging needles to LUFS

The split between loudness metering and peak metering is as old as broadcasting. In 1939, engineers from Bell Telephone Laboratories, CBS and NBC standardized the VU meter — the "volume unit" indicator with the famous cream face and swinging needle. Its defining feature was its ballistics: the needle was deliberately damped, taking roughly 300 milliseconds to settle. That sluggishness was the point. The VU meter ignored momentary spikes and showed the average "volume" of program material, which is what an operator riding a fader actually needed to match one announcer to the next.

European broadcasters went the other way. The BBC and, later, German and Nordic broadcasters adopted the peak programme meter (PPM), a fast-attack, slow-decay instrument designed to catch brief peaks that would over-modulate a transmitter — exactly the events a VU needle glides straight past. The two meter families, eventually codified in IEC standards, embody the same division of labour modern meters still have: one instrument for loudness, one for overload.

Digital audio sharpened the peak side of the question. Analog tape and transmitters overload gradually; a digital converter does not. That gave us the dBFS scale, where 0 is the last representable value and everything useful lives in negative numbers. And in the 2000s the loudness side finally got a rigorous successor to the VU meter: ITU-R BS.1770 (2006) defined an algorithm for measuring perceived loudness in LUFS, and the EBU's R 128 recommendation (2010) built Europe's loudness-normalization practice on top of it. Streaming platforms adopted the same idea, which is why music services today turn tracks down to a common loudness rather than letting the loudest master win. RMS metering is the direct ancestor of all of it — LUFS is essentially a frequency-weighted, gated refinement of the same mean-power measurement.

The math, in one paragraph

Root mean square is exactly what the name says, applied in reverse order: take a window of samples, square each one (making everything positive and proportional to power), take the mean of those squares, then take the square root to get back to amplitude units. For a sine wave the result is about 3 dB below the peak; for a square wave, RMS and peak are equal, because the signal spends all its time at full amplitude. The gap between the two values is the interesting part: it tells you how "spiky" a signal is.

Crest factor: why speech needs headroom

That gap has a name — crest factor, the ratio of peak level to RMS level. Speech has a famously large one: depending on the talker and how you measure, the peaks of conversational speech sit roughly 12–20 dB above its RMS level. Plosives ("p", "b", "t") and stressed syllables produce brief spikes carrying little sustained energy, while vowels carry the power that determines loudness.

This is the practical reason you cannot set a speaking level by peaks alone. If you push speech peaks to just under 0 dBFS, its RMS may sit near −15 to −20 dBFS — and one slightly more emphatic consonant clips. Working with deliberate headroom — speech peaks around −12 dBFS is a common studio target — leaves the crest factor room to breathe. It's also why compression is so effective on voice: by trimming the tallest peaks, it lowers the crest factor, letting the average level rise several dB without moving the peaks any closer to the ceiling.

Reading a meter on a call

For live speech into a call, the routine is simple. Watch the RMS reading while you talk normally — not your "testing, testing" voice — and treat it as your loudness anchor; it should sit well up the scale and move with your phrasing. Watch the peak marker for the ceiling: if it keeps flirting with the top of the scale, turn input gain down at the source rather than hoping the spikes stop. And take any latched clip indicator seriously even if you didn't hear anything wrong — clipping on consonants is often inaudible in your own monitoring but becomes harsh after a platform's codec has chewed on it. A meter you glance at every few minutes prevents the two classic call failures: the whisper nobody can hear, and the boom that distorts.

One caveat applies to any metering: it only tells the truth about the point in the chain where it's measuring. A meter reading the raw microphone says nothing about what your processing — gain, EQ, compression — did afterwards. If the meter and the far end of the call disagree, the meter is measuring the wrong spot.

In DeskBroadcast

DeskBroadcast's level meter shows both numbers at once: a continuous RMS bar with a separate peak marker, both calibrated in dBFS on a −60…0 scale. A CLIP flag latches when the signal reaches −0.5 dBFS, so a single overload can't slip past between glances. Crucially, the meters read the processed signal — after noise reduction, EQ and compression — which means they show literally what Zoom, Meet or Teams receives from the virtual mic, not what the hardware capture looked like. The processing chain itself targets speech peaks around −12 dBFS, the broadcast-style delivery level that keeps crest-factor spikes clear of the ceiling; if the peak marker rides near −12 and the CLIP flag stays dark, your level is right. The same processed signal feeds the spectrum analyzer, and mic calibration uses these meters to set your input gain in the first place.