What is phase in audio?

Phase in plain English

A sound wave pushes the air, pulls it back, then pushes again. One push and one pull make a cycle, and phase says how far through that cycle the wave is at a given moment. It's measured in degrees, like a trip round a circle: 0° is the start, 90° is the top of the push, 180° is halfway, where the wave crosses back through zero, and 360° brings it back to where it began.

On its own, the phase of a sound tells you almost nothing. Your speaker starts somewhere in a cycle, and you couldn't hear the difference if it started somewhere else. Phase becomes audible when there are two versions of the same sound: two microphones on one drum, the left and right channels of a stereo guitar, or a track and a copy of it that plays slightly late. What counts then is the phase of one compared with the other.

If the two line up, they add together and get louder. If one is half a cycle behind, every push from one meets a pull from the other and they cancel. Real recordings usually land somewhere in between, and that's where the trouble starts, because the amount of cancellation changes with frequency.

Degrees, cycles and milliseconds

A delay is measured in milliseconds, but what it does to phase depends on the frequency, because a low note's cycle takes longer than a high note's. A 100 Hz wave takes 10 ms to complete one cycle, so a 1 ms delay leaves it 36° out of step. The same 1 ms is half a cycle (180°) at 500 Hz and a whole cycle at 1 kHz, where the delayed copy lines up again.

So a delay shifts each frequency by a different number of degrees, which cancels some and reinforces others. The figure shows that happening to a wave made of two frequencies.

The first frequency to cancel is 1 divided by twice the delay. So 1 ms cancels 500 Hz, 2 ms cancels 250 Hz and 10 ms cancels 50 Hz, and the next cancellations sit at three, five and seven times that frequency, carrying on all the way up the spectrum.

Phase vs polarity

Polarity gets called phase all the time, and it's a different thing. Flipping polarity turns a waveform upside down: every push becomes a pull at exactly the same moment. Nothing moves in time, so every frequency is inverted at once, the same as a 180° shift at all of them together.

A delay can only put one frequency, and its odd multiples, at 180°. You can see the difference in the lab. Flip the polarity of a sound that's identical in both channels and its mono sum is silence; delay one side instead and the mono sum is a row of notches with plenty of sound left between them.

Consoles and DAWs often label the polarity switch as phase, or with the Ø symbol, and Logic's Gain plug-in calls it Phase Invert. Whatever the label says, it flips polarity. The phase vs polarity lesson covers the difference in more depth.

What is phase cancellation?

Phase cancellation is what happens when two versions of a sound are added while out of step. Where one pushes as the other pulls, they subtract. At the same level and exactly opposite, the result is silence; if one is quieter, or the timing is only partly off, you get a dip instead.

It turns up whenever the same sound arrives by two routes. Two microphones on one instrument hear it at slightly different times. A kick sample layered on a recorded kick rarely starts at precisely the same point. And when a stereo mix plays in mono, its left and right channels are added together, so anything out of phase between them cancels.

That last case catches people out, because a mix can sound fine in stereo and fall apart on a mono speaker. In stereo each speaker plays its own channel and nothing gets added electrically. Sum the two and the cancellation is real.

What is comb filtering?

When a sound is added to a delayed copy of itself, the cancellations land at evenly spaced frequencies. Plot the result and you get a row of deep notches like the teeth of a comb, which is where comb filtering gets its name.

Take a 1 ms delay. The first notch is at 500 Hz, where the copy is half a cycle late. At 1 kHz it's a whole cycle late and adds back up; at 1.5 kHz it's one and a half cycles late and cancels again, and the pattern repeats every 1 kHz. So the notches sit at 500 Hz, 1.5 kHz, 2.5 kHz, 3.5 kHz and on up. Make the delay longer and they move down and crowd together: at 10 ms they start at 50 Hz and sit 100 Hz apart.

Comb filtering sounds hollow and boxy, a bit like talking through a cardboard tube. When the delay changes over time the notches sweep up and down, which is exactly how a flanger works. In the lab, the wide screen labelled “What mono does to the tone” draws this response from your settings and marks each notch in Hz.

Reading the lab

The lab plays a stereo loop through a chain you control. The sound passes through an optional second copy added to each channel, then a delay and a polarity switch on the right channel, then width, pan and finally the Mono button. Every screen is measured from that same chain.

The goniometer, top left, plots the left channel against the right, turned 45° so that a sound identical in both speakers draws a vertical line. A wide stereo sound makes a fuzzy ball, and a sound in one speaker only leans along that speaker's diagonal. When the shape tips over towards horizontal, the two sides are moving in opposite directions, and that cancels in mono.

Beside it, three meters show the left and right levels and the level of the mono sum. If SUM drops well below L and R, something is cancelling.

The wide screen shows what mono does to the tone. The lab sends a click through your settings, adds left and right together and measures the frequency response. A flat line at 0 dB means mono leaves the tone alone; every dip is a cancellation, labelled with its frequency.

Underneath, the correlation meter reads from −1 to +1, and the strip below it shows the same reading across the whole loop. Next to that, two cycles of a 110 Hz tone sent through the chain show the left channel in teal and the right in pink, so you can watch the right one slide along as you delay it or turn upside down as you flip it. The panel on the right explains what your settings are doing, with numbers measured from the loop.

You can use your own sound too. Upload a file, or record eight seconds with the Record button, and it loops through the lab in place of ours without leaving your device.

The controls

Flip R

Turns the right channel upside down. On a sound that's almost the same in both channels, like the acoustic guitar loop, the goniometer swings from vertical to horizontal and the correlation goes from nearly +1 to nearly −1. In stereo it sounds wide and oddly placeless, with weaker bass, because two speakers playing opposite low frequencies cancel in the room as well. Press Mono and it all but vanishes.

Delay R

Plays the right channel up to 20 ms late. Under about 1 ms the sound slides towards the left speaker, because your brain uses tiny timing differences between your ears to work out where a sound is. From a few milliseconds up to about 30 ms it stays on the earlier side but seems wider and more spacious, which is the Haas effect.

Whatever it does in stereo, in mono the delay becomes a comb filter with its first notch at 1 ÷ (2 × delay). At 0.5 ms that's 1 kHz. At 15 ms it's 33 Hz, with another notch every 67 Hz above.

Copy level and copy delay

These add a second copy of the sound to each channel, the way a second microphone on the same source would. Copy level sets how loud it is and copy delay sets how late. At 100% the copy matches the original and its notches go all the way down; at 50% they're about 10 dB deep. Because this happens inside each channel, you hear the comb in stereo as well as in mono.

Width

Width works on mid and side. The mid is what the two channels share, (L + R) ÷ 2, and the side is how they differ, (L − R) ÷ 2. The width control turns the side up or down and leaves the mid alone, so 0% throws the side away (which is mono), 100% leaves the sound as it was and 200% doubles the difference.

Mono is the mid on its own. That means mid/side width leaves the mono version exactly as it was at any setting, and the only thing it changes is how much of the stereo picture goes missing when the sides are summed. Watch the mono response stay flat whatever you do with the width knob. Stereo wideners that use delays or phase shifts are a different story, and those can comb filter.

Pan

Moves the sound towards one speaker. On a stereo sound it folds the far channel into the near one, so hard left plays both channels from the left speaker. Panning only changes levels, without delaying anything, so it can't cause a comb filter.

Mono

Adds left and right together and plays the result, halved, through both speakers, so a sound that was identical in both channels stays at the same level. That's what you get from a single Bluetooth speaker and from a lot of venue PA systems. It's the button to keep pressing while you work through this page.

Phase on multi-miked drums

A drum kit is the usual place to run into a phase problem. A snare might have a mic on top and another underneath, and it also turns up in the overheads and the room mics. Each of those mics sits a different distance from the snare, so each hears the hit at a slightly different time, and when you mix them they comb filter against each other.

Sound travels at about 343 metres a second at room temperature, which is roughly 34 cm every millisecond. An overhead 1 m further from the snare than the top mic hears the hit about 2.9 ms later, which puts the first notch near 170 Hz, right in the body of the drum.

The snare bottom mic has a different problem. When the stick hits, the top head moves away from the top mic while the bottom head moves towards the bottom mic, so the two start each hit with opposite polarity. That's why flipping the bottom mic's polarity is the normal first move. A kick with one mic inside and one outside needs the same check, and so does a bass recorded through a DI and an amp mic at once.

The fix is to listen for the fullest sound. Flip the polarity of one mic and keep whichever setting gives more low end and body. If it still sounds hollow, nudge the later track earlier a fraction of a millisecond at a time, or move the mic. The drum phase alignment lesson goes through a whole kit.

When you set up several mics, the 3:1 rule helps: keep each mic at least three times as far from the other mics as it is from its own source. The spill between them is then quiet enough that its comb filtering is hard to hear.

Stereo width and mono compatibility

A mix is mono compatible when it still sounds right with left and right added together. Some width always disappears in mono, which is fine. The vocal, kick, snare and bass shouldn't: they need to keep their level and tone.

Most stereo width comes from differences between the channels, and differences are exactly what mono removes. Width made by panning separate mono tracks to different places is safe, because each track is still a whole sound in one place. Width made with small delays between the channels, or with wideners that rely on delays, can turn into comb filtering when it's summed. The mono compatibility lesson covers testing a whole mix.

Low frequencies are where cancellation hurts most, since they carry most of a mix's energy. Your ears are also poor at telling where very low sounds come from, so there's little to gain from spreading them. That's why it's normal to keep everything below about 100 to 150 Hz in the centre.

Reading a correlation meter

A correlation meter compares the left and right channels and gives one number. Watch it while the loudest part of the song plays; how it moves tells you more than any single reading.

Short dips below zero on reverb tails and wide effects are normal; the warning sign is a needle that sits there. The correlation meter lesson goes into the goniometer as well, which shows the same thing as a picture.

How to check phase in your DAW

You don't need to buy anything. Most DAWs come with a way to sum to mono, and free correlation meters are easy to find.

1. Put a mono switch on the masterIn Ableton Live, add a Utility to the master and use its Mono switch. In Logic, the Gain plug-in has a Mono button. Keep it last in the chain.
2. Add a correlation meterLogic's MultiMeter includes a goniometer and a correlation meter, and plenty of free metering plug-ins have one.
3. Switch between stereo and mono during the chorusThe kick, snare, bass and vocal should keep their level and tone.
4. Find the pair that cancelsIf something drops, solo the tracks that make it up. Multi-miked instruments, layered samples and stereo effects are the usual suspects.
5. Try polarity firstFlip one of the pair and keep whichever setting sounds fuller in mono.
6. Then fix the timingIf polarity doesn't cure it, zoom in on the waveforms and slide one track a fraction of a millisecond at a time until the starts of the hits line up.
7. Check the meter againBrief dips are fine. A needle that stays below zero means something is still out of phase.

Common phase mistakes

Fixing a timing problem with the polarity button

Flipping polarity only helps when two signals are close to opposite at every frequency. If the real cause is a delay, flipping just moves the notches: with a 1 ms delay they jump from 500 Hz, 1.5 kHz and so on to 1 kHz, 2 kHz and up, plus a new one at the very bottom. Line the timing up instead.

Widening the bass

Wideners, chorus and Haas delays on a bass line sound huge in headphones and then thin out in mono, taking the low end with them. Keep the sub in the centre and widen only the upper part.

Trusting headphones

On headphones each ear hears one channel and the two never mix in the air, so many phase problems stay hidden. Check in mono, and on speakers if you can, before you call a mix finished.

Doubling a part by copying it

Copy a track, nudge it a few milliseconds and pan the two apart, and you get a wide sound that turns into a comb filter in mono. A second real take gives width without a fixed delay between the sides, so nothing cancels in a regular pattern.

Never pressing Mono

It takes a second. Press it every time you add a stereo effect or layer one sample over another.

Train your ears to hear phase

The quiz under the lab mixes two kinds of round. One plays a loop either normally or with the right channel flipped and asks whether it's in phase. The other hides the settings and asks whether the loop will survive mono, so you have to hear the stereo version and predict what summing will do to it. Every round is matched in loudness with the screens covered. Headphones make a flipped channel easier to spot, because it feels as though it's coming from inside your head with no clear centre.

Questions people ask

Keep going