How to read a correlation meter

What is a correlation meter?

A correlation meter is a needle or bar that answers one question about a stereo signal: how much do the left and right channels have in common? You'll also see it called a phase meter or a phase correlation meter. It usually sits on the master bus, though it works on any stereo track, and it reads from −1 at one end to +1 at the other.

The reason to watch it is mono. Most single Bluetooth speakers play in mono, and so do plenty of club sound systems, which means they play left and right added together. Whatever the two channels share survives that addition, and whatever is opposite in them cancels. The correlation meter shows which way your mix leans while you're still listening in stereo.

It's the number version of a goniometer, the scope that plots left against right as a cloud of dots, and most metering plug-ins show the two side by side. This page goes through reading both, what a reading predicts about mono, and the problems one number can't show you. If phase itself is new to you, start with what is phase in audio?

The sum behind the needle

At every sample, the meter multiplies the left channel's value by the right channel's. When both speakers are being pushed the same way at that instant, the two values have the same sign and the product is positive; when one pushes while the other pulls, it's negative. The meter adds those products up over a short window and divides by the level of each channel. Written out, that's Σ(L × R) ÷ √(ΣL² × ΣR²), Pearson's correlation coefficient, which statisticians use to compare two columns of figures, with the step that subtracts each column's average left out because audio averages zero anyway.

Dividing by the levels makes the meter ignore volume. A sound that's 6 dB quieter in the right channel than the left still reads +1, because both channels have the same shape, and pulling the master fader down doesn't move the needle at all.

The length of the window decides how jumpy the needle is. The lab's needle follows the last 4,096 samples, which is under a tenth of a second, and is smoothed so you can follow it by eye. Other meters choose their own timing, so the same mix can look calmer on one than another.

Reading the lab

The correlation meter is the screen labelled Correlation. Its bar runs from −1 on the left to +1 on the right, shaded pink below zero, amber up to about +0.3 and teal above that. The white needle follows the loop as it plays, and the small triangle under the bar marks the reading for the whole loop. Under the bar, a strip shows the reading in 240 slices across the loop, so you can see the quiet moments and the dips without waiting for them. Hover over either for the figures.

The goniometer above it draws the same thing as a picture. Next to it, three meters give the left, right and mono sum levels, and the screen headed “What mono does to the tone” measures the mono version's frequency response by sending a click through your settings. That last screen is the one to check whenever the needle can't tell you enough.

On this page the lab has Flip R, Delay R, Width, Pan and the Mono button. The copy controls belong to the comb filtering lesson and are greyed out. You can also upload your own mix, and the Right now panel will give its correlation and how much it loses in mono before you change anything.

The readings from +1 to −1

+1: the same in both speakers

The needle reaches +1 when the left and right channels have exactly the same shape, which makes the signal mono. A single vocal mic does this, and so does the nylon guitar loop in the lab, which was recorded in mono and reads +1.00. Set the width to 0% and the whole mix does the same, with the goniometer shrinking to a vertical line.

A finished mix that sits at +1 all the time has no stereo in it. If that isn't what you meant, look for a mono switch or a utility plug-in left in mono somewhere on the master.

Between 0 and +1: a centre with width around it

Most mixes spend their time here. The full mix loop reads +0.64 overall: the kick, bass and vocal are shared by both channels, while panned parts and reverb make the two sides differ. Watch it play and the needle wanders with the arrangement, but of the strip's 240 slices only four dip below zero, each for a few hundredths of a second.

0: two unrelated channels

At 0 the channels have nothing in common on average. Two separate recordings, one in each speaker, read close to it, and so does a mix whose width has been pushed hard: at 200% width the full mix reads +0.06 and comes out 2.8 dB quieter in mono than in stereo. Press Mono at that setting, though, and it sounds exactly as it did at 100%. Mid/side width only turns up the part of the signal that mono throws away, so the stereo version has grown while the mono one hasn't changed.

A reading of zero has a second cause that's no problem at all. Pan a mono sound hard to one side and the other channel goes silent, so the meter has nothing to compare and reads 0.00. Nothing cancels in mono, and the flat line on the mono screen shows it; the level reading drops 3 dB only because the lab's mono plays half the sum through each speaker. Pan the same guitar halfway and the reading goes straight back to +1.00, because both channels have the same shape again, just at different levels.

Below 0: the sides pulling against each other

Below zero, the speakers are mostly moving in opposite directions: when one pushes, the other pulls. Flip the polarity of one channel and every reading is mirrored, so the full mix's +0.64 becomes −0.64 and the goniometer turns on its side. This is where mono goes badly wrong. The mix comes out 7.4 dB down, and the kick and bass drop furthest because they sat almost entirely in the middle.

Real material can read below zero without anyone flipping anything. The lead synth loop reads −0.18 as supplied and spends more than half of its eight seconds below zero; in mono it's 3.8 dB quieter and much narrower. Wide synth patches and some stereo wideners can do this too. A needle that flicks under zero for a moment is normal, but one that settles there for whole sections is the warning.

How much does a reading lose in mono?

Correlation ties straight into mid and side. The mid is what the two channels share, (L + R) ÷ 2, and the side is how they differ, (L − R) ÷ 2. With both channels about equally loud, the reading is (mid power − side power) ÷ (mid power + side power). So +1 means there's no side at all, 0 means the side is as loud as the mid, and anything negative means the side is louder.

Mono keeps the mid and drops the side, so the same reading tells you what mono costs in level. With c standing for the reading, the change is 10 × log₁₀((1 + c) ÷ 2) dB. The full mix's +0.64 works out at −0.86 dB, which the lab rounds to −0.9, and flipped to −0.64 it's −7.4 dB.

Most of what you need is in the top half. Between +1 and +0.6 mono costs a decibel or less, which is a fair price for width. From 0 downwards the losses grow fast, and by −0.6 the mono version is 7 dB down, enough for a part to vanish behind everything around it.

The maths assumes the two channels are about equally loud. That holds for most mixes and fails for a single part panned hard to one side, which is why the panned guitar reads 0 without losing anything to cancellation.

Reading the goniometer alongside it

A goniometer, also called a vectorscope or phase scope, plots each moment of the sound as one dot, placed by how far it pushes the left speaker against how far it pushes the right. The plot is turned 45° so that mono stands upright, with the mid running up the screen and the side running across it. The correlation meter is a one-number summary of that cloud, and the shapes are quick to learn:

• A thin vertical line is mono, reading +1.
• Taller than it is wide, a fuzzy column means a solid centre with stereo around it, the usual shape for a mix.
• When the cloud is a round ball, the two channels are unrelated and the reading is near 0.
• A line lying along the L or R diagonal is a sound in one speaker only.
• Wider than it is tall means more side than mid, and a negative reading.
• One channel upside down against the other draws a flat horizontal line, at −1.

The goniometer shows two things the needle can't. When one channel is louder, the shape leans towards that side, which the correlation meter ignores entirely. And because it draws every moment, you can watch a single part, like a wide crash cymbal or a stereo delay throw, spread the cloud sideways as it comes in.

Blind spots of a single number

Which frequencies are wide

The meter counts each frequency by how much energy it carries, and in most mixes the low end carries the most. In the full mix loop, the band below 150 Hz is 7.6 dB stronger than the band above 2 kHz, so the needle mostly follows the kick and bass. A top end that's far wider can hide behind them. These are the lab's loops measured across the whole range, then low-passed at 150 Hz and high-passed at 2 kHz, along with the share of each loop's 240 slices that reads below zero:

The full mix reads +0.80 in the low end and +0.33 at the top, and its whole-range +0.64 lands between the two. The drums go the other way: both ends read higher than the whole, so most of their width sits between 150 Hz and 2 kHz. The bass loop deserves a second look. Its +0.47 overall seems fine, but its low end reads only +0.56, and for almost a quarter of the loop the needle is below zero, which means mono costs it more than 3 dB in those moments. A wide bass costs far more than a wide hi-hat, because the bass is where the energy is, and the mono compatibility lesson covers keeping it centred.

To read the low end on its own, put a low-pass filter at about 150 Hz in front of the meter for a moment, or use a meter that splits the reading into bands. Voxengo's free Correlometer does that.

Comb filtering

A small delay between the channels is the case the meter handles worst. Play the right channel 0.5 ms late and the full mix still reads +0.44, a number that passes any quick glance. In mono, that half millisecond makes a comb filter with a complete notch at 1 kHz and another every 2 kHz above it. Low frequencies barely notice 0.5 ms (at 100 Hz it's a twentieth of a cycle), so they stay correlated and hold the needle up while the mids and highs cancel.

The mono response screen in the lab shows every one of those notches, and pressing Mono lets you hear them.

Problems inside each channel

The meter only ever compares left with right. Two mics on one snare that comb filter against each other, or a doubled part nudged a few milliseconds late on the same track, do the same damage to both channels, so the two sides still match and the needle stays where it was. Those need your ears, a solo button and the drum phase alignment lesson.

Anything after a mono switch

Put the meter after a mono switch and it reads +1 whatever the mix is doing, because both channels are now the same. Press Mono in the lab with the right channel flipped and the needle jumps to +1 while the SUM meter shows 7.4 dB going missing. Read the meter in stereo, and keep the mono button for listening.

How to check a mix with a correlation meter

1. Put the meter last on the masterAfter every other plug-in, so it sees exactly what you'll export. If you use a mono-check plug-in, place it after the meter or switch it off while you read.
2. Play the busiest sectionThe chorus or the drop, where the most parts play at once.
3. Watch for a few barsNotice where the needle spends most of its time and how far it dips. A flick towards zero on a reverb tail or a wide fill is fine.
4. Find whatever pulls it below zeroIf it settles under zero, solo groups or mute tracks one at a time until it comes back up. The usual causes are a flipped channel, a stereo widener, a part copied and nudged to fake a double, or spaced stereo mics.
5. Check the low end on its ownLow-pass the signal feeding the meter at about 150 Hz. In most styles the kick and bass should read close to +1 there.
6. Listen in monoThe meter tells you where to look and the mono button tells you whether it matters. The vocal, kick, snare and bass should keep their level and tone.

Correlation meters in your DAW

Logic Pro has a Correlation Meter plug-in, and its MultiMeter shows correlation next to a goniometer. If your DAW has nothing built in, Voxengo SPAN is free and has a stereo correlation meter alongside its spectrum analyser, and Voxengo Correlometer gives a reading per frequency band. Whichever you use, check that it reads +1 on a mono sound and −1 on that sound with one channel flipped. Then you know which end is which and that it's seeing both channels.

Common correlation meter mistakes

Treating zero as a fault

A hard-panned part on its own or a spaced pair of room mics can read near zero without anything being broken. Zero tells you the channels are unrelated. It's the readings below zero that predict cancellation.

Chasing +1

A mix pinned between +0.95 and +1 is close to mono. Width always costs some level in mono, about 1 dB at +0.6, and for everything except the low end that's normally a trade worth making.

Trusting a good number

A reading of +0.44 can hide a notch at 1 kHz, as the half-millisecond delay showed, so press Mono before you trust any reading.

Reading one moment

The needle at any single instant depends on which notes happen to be playing. Judge it over a few bars of the loudest section, or look at a strip or history view if your meter has one.

Train your ears for mono

The quiz under the lab asks one question each round: would this loop survive mono? It covers the screens and hides the settings, and some rounds only change the width, which mono doesn't mind, while others hide a flipped channel or a delay of a millisecond or so. The Mono button disappears too, so you judge from the stereo sound alone. Each answer comes with the reason, and your best score is kept in your browser.

Questions people ask

Keep going