how to widen stereo without phase issues

the mix that disappears

you spend an hour widening your mix. the pads fill the stereo field. the guitars breathe. the reverb tails shimmer across the speakers. you bounce the file, send it to your phone, and hit play. the pads are gone. the guitars sound thin. the vocal you did not touch sounds like it is coming through a telephone. you have been here. i have been here.

this is what phase issues sound like. your stereo widening created differences between the left and right channels. on speakers or headphones, those differences sound like width. on a phone speaker, which sums left and right into a single driver, those differences cancel. the wider you made it, the more it cancels.

the problem is not stereo widening itself. the problem is widening without understanding what survives mono playback and what does not. every widening technique introduces phase relationships between the channels. some of those relationships are benign: they cancel into mild spectral ripple that is barely audible. others are destructive: they cancel into deep notches that hollow out your mix.[^1]

this guide is about the difference between the two, and how to stay on the safe side.

what causes phase cancellation

when two signals are summed, identical content adds together and opposite content cancels. this is physics, not opinion. if the left channel has a 1 kHz sine wave at +1 and the right channel has the same wave at -1, the mono sum is zero. the sound disappears.

stereo widening works by making the left and right channels different. Haas delay offsets one channel in time, which means certain frequencies arrive in phase (they add) and others arrive out of phase (they cancel). the result on mono fold-down is comb filtering: deep notches at regular frequency intervals, spaced at 1/delay Hz apart. a 10 ms delay creates notches every 100 Hz. a 20 ms delay creates notches every 50 Hz.

allpass decorrelation shifts the phase of different frequencies by different amounts without changing the magnitude. the mono sum produces partial cancellation and partial reinforcement, but because the phase shifts are spread irregularly across the spectrum, the result is mild spectral ripple rather than deep periodic notches. typically less than 1 to 2 dB of variation.

M/S processing boosts the Side channel, which is the difference between left and right. on mono fold-down, the Side content cancels by definition (Mid survives, Side vanishes). moderate M/S boost is safe because the Side is usually a fraction of the total energy. excessive Side boost makes the Side louder than the Mid, which means mono playback loses more energy than it keeps.

reading the correlation meter

the correlation meter is your primary tool for monitoring mono safety. it measures how similar the left and right channels are, on a scale from -1 to +1.[^2]

+1.0: the channels are identical. the signal is perfectly mono. maximum mono compatibility, zero stereo width.

+0.3 to +1.0: safe zone. the channels share enough common content that mono summing preserves the overall balance. most of the energy survives. this is where a well-mixed stereo master should live.

0 to +0.3: caution zone. the channels are significantly different. mono playback will lose noticeable energy and spatial impression. individual tracks (reverb returns, wide pads) can live here. the master bus should not.

below 0: danger zone. the channels are more different than similar. mono summing causes net cancellation. the mix will sound thin, hollow, or broken on any mono device. if your master bus reads below 0, something is wrong.

a single number does not tell the whole story. the correlation meter shows the average across the entire spectrum. a mix can read +0.5 overall while having severe phase cancellation in a narrow frequency band. this is why you need to listen in mono, not just watch the meter.

three rules for safe widening

rule 1: mono the bass

frequencies below 100 to 200 Hz should be mono. always. your ears cannot localize bass, so stereo bass adds no perceptual width. what it adds is phase differences that cancel on mono playback, costing you low-end power on every phone speaker, bluetooth device, and club system with a mono subwoofer.

apply a bass crossover before any stereo widening. collapse everything below 100 to 200 Hz to mono, then widen above. the exact frequency depends on the material: electronic music often uses 100 to 150 Hz, pop and rock use 80 to 120 Hz. KERN WIDE’s FOCUS parameter does exactly this: it sets a crossover that monos the bass and applies widening only above it.

rule 2: choose the right technique

not all widening techniques are equal in mono safety. the hierarchy, from safest to most dangerous:

1. allpass decorrelation: preserves magnitude spectrum, mild spectral ripple on mono fold-down (1 to 2 dB). the safest option for most production work
2. M/S boost: safe at moderate levels (1 to 3 dB of Side boost). dangerous at extreme levels where Side energy exceeds Mid
3. Haas delay: dramatic width, terrible mono compatibility. deep comb filter notches on every mono device. use only when you have verified the mono fold-down is acceptable
4. pitch-shift widening: time-varying comb filtering, audible artifacts on mono and stereo. use sparingly

the right technique depends on the source. stage 5 covered source-specific decisions in detail. the short version: allpass decorrelation for mono sources, M/S for already-stereo content, Haas only when the mono fold-down has been verified.

rule 3: check before you ship

the mono check is not optional. it is the last step before export. fold the mix to mono and listen for 30 seconds. you are listening for three things:

1. disappearing elements: if a sound drops significantly in level, its widening is creating too much cancellation
2. tonal changes: if a sound becomes nasal, hollow, or thin, comb filtering is affecting its frequency balance
3. overall level drop: some level loss is normal (stereo content loses its Side energy). more than 1 to 2 dB of overall loss suggests excessive widening across the mix

if any element fails these checks, back off its width or switch to a more mono-safe technique.

the safe widening workflow

here is the practical workflow, in order.

step 1: set the bass crossover. before touching the width knob, collapse the low end to mono. 100 to 150 Hz is a safe starting point. listen in mono to confirm the bass stays intact.

step 2: widen in context. do not solo a track and widen it until it sounds impressive. widen it while the full mix plays. the width that sounds right in solo is almost always too much in context.

step 3: check the track correlation. after widening each source, check the correlation meter on that track. individual tracks can dip to +0.1 for ambient sources (pads, reverb). keep focused sources (vocals, guitars) above +0.3.

step 4: check the bus correlation. play the full mix and watch the master bus correlation. it should stay above +0.3, with brief dips into the caution zone being acceptable on transients or wide sections.

step 5: mono fold-down. the final test. fold to mono and listen. no element should disappear. the mix should lose depth and spatial impression but no instruments, no tonal balance, and no low end.

the five most common mistakes

1. widening everything. not every element needs width. lead vocals, kick, bass, and snare should anchor the center. wide elements only work when narrow elements give them contrast. if everything is wide, nothing sounds wide.

2. widening in solo. a pad that sounds impressively wide in solo pushes the bus correlation dangerously low when it plays with the rest of the mix. always widen in context. the right amount of width is always less than what sounds good in solo.

3. ignoring bass width. stereo bass wastes headroom and cancels on mono systems. even if you do not add width to your bass, check that your synth patches and samples do not arrive with stereo content in the low end. sum below 150 Hz.

4. stacking wideners. a stereo chorus into an allpass widener into an M/S boost compounds the decorrelation at each stage. the correlation drops multiplicatively, not additively. one widening technique per track, applied intentionally, is almost always enough.

5. skipping the mono check. the correlation meter helps during mixing. the mono fold-down catches what the meter misses. both are necessary. neither alone is sufficient. ten seconds of listening in mono can save you from a mix that falls apart on half the devices in the world.[^3]

frequently asked questions

frequently asked questions

why does my stereo widened mix sound thin on phone speakers?

phone speakers sum the left and right channels to mono. if your stereo width comes from phase differences between the channels, those differences cancel on summation. the result is partial or complete loss of the affected signal. techniques like Haas delay are especially vulnerable because the delayed copy creates comb filtering when summed. allpass decorrelation is dramatically safer because it preserves the magnitude spectrum even when the channels are combined.

what correlation value should I aim for in a stereo mix?

for a full mix, keep the correlation above +0.3. individual tracks can dip lower (down to +0.1 for wide pads or reverb returns), but the master bus should stay safely above +0.3. between 0 and +0.3 is a caution zone where mono playback will lose some spatial impression and possibly some tonal balance. below 0 means the left and right channels are more different than similar, and mono fold-down will cause audible cancellation.

how do I check mono compatibility before releasing a mix?

use a correlation meter on your master bus while mixing. before you export, fold the mix to mono using a utility plugin or your monitoring controller. listen for 30 seconds. if any instrument disappears, sounds thinner, or changes character, that element has phase issues from widening. back off the width on that track or switch to a more mono-safe technique like allpass decorrelation. also check on a phone speaker or single bluetooth speaker for real-world confirmation.

is allpass decorrelation completely mono safe?

no widening technique is 100% transparent in mono. allpass decorrelation preserves the magnitude spectrum, so the frequency balance survives mono fold-down with only mild spectral ripple (typically less than 1 to 2 dB). this is dramatically better than Haas delay, which creates deep comb filter notches of 10 to 18 dB. allpass is the safest option available, but you should still check your correlation meter and listen in mono before shipping.

should I apply stereo widening before or after reverb?

widen the dry signal first, then add reverb. reverb naturally adds stereo content because each channel gets different reflection patterns. if you widen an already-reverbed signal, you are decorrelating content that is already decorrelated, which pushes the correlation toward zero faster. the exception is M/S processing on a reverb return, where boosting the Side channel can widen the reverb tail without affecting the dry signal.

references

a note from the developer

the hardest lesson in building KERN WIDE was that the correlation meter is not a safety net. it is a guideline. i can build a signal that reads +0.5 on the correlation meter and sounds terrible in mono, and i can build one that reads +0.2 and translates perfectly. the meter measures average correlation across the spectrum. your ears hear the details the meter averages out.

that is why WIDE has a built-in correlation constraint that operates per-frequency-band, not just globally. the algorithm monitors the interaural cross-correlation in 40 ERB bands and prevents any individual band from going below a safe threshold. a global meter might read +0.4 while a narrow band at 2 kHz is at -0.3. the per-band constraint catches that. the global meter does not.

the three rules in this guide (mono the bass, choose allpass, check the fold-down) are the rules i follow in my own mixing. they are not WIDE-specific. they work with any tool. WIDE just makes the first two automatic: the FOCUS parameter handles the bass crossover, and the STEREO mode uses allpass decorrelation with the per-band correlation constraint built in.

if you have a mono-checking workflow that catches things i missed, jonas@kernaudio.io. in manufacturing, every part has tolerances. correlation is your tolerance spec. i am always looking for better ways to measure it.