Haas effect and allpass decorrelation

the two ways to make width

you have a mono vocal. it sits dead center in your mix. you want it to feel wider, more spacious, more present in the stereo field. panning will not help because there is only one signal. M/S processing will not help because the Side channel is silent. you need to create a difference between left and right that did not exist before.

there are two fundamental approaches: delay one copy of the signal, or change the phase of one copy. the Haas effect uses delay. allpass decorrelation uses phase. both create width. the trade-offs are very different. (this is the choice that determines whether your width survives on a phone speaker.)

understanding these two techniques matters because every stereo widener you will ever use is built on one of them, or a combination of both. the Haas effect has been used since the 1950s. allpass decorrelation is the approach behind most modern psychoacoustic wideners, including KERN WIDE’s STEREO mode.

the Haas effect

in 1949, Helmut Haas demonstrated something that Wallach, Newman, and Rosenzweig had described the same year as the “precedence effect.” when two copies of a sound arrive within a short time window, your brain fuses them into a single perceived event located at the earlier source. the delayed copy is not heard as a separate echo. instead, it contributes to a sense of spaciousness and width.[^1]

this is the principle behind Haas-based stereo widening: send the original signal to one channel and a delayed copy (5 to 30 ms) to the other. your brain hears one sound but perceives it as wider than the original.

the timing matters. below about 1 ms, the delay is too short for your brain to detect a directional difference. between 5 and 30 ms is the sweet spot for music: width without obvious echo. above 30 to 40 ms, the delayed copy starts to separate perceptually and you hear a distinct repeat, a slap-back echo rather than spaciousness.[^2]

why it works so well

the Haas effect produces dramatic, immediately obvious width. a 15 ms delay between left and right makes a mono vocal feel like it fills the entire stereo field. the effect is intuitive and easy to set up: duplicate the track, pan the copies left and right, and add a short delay to one side.

this is why Haas delay has been a studio trick for decades. it works on any source, it is simple to implement, and the width it creates is bold and unmistakable.

the mono problem

here is the cost. when the left and right channels are summed to mono (a phone speaker, a bluetooth device, a club PA system), the delayed copy interferes with the original. the result is comb filtering: a series of deep notches in the frequency response where the two copies cancel each other. the Haas effect is an auditory illusion. your brain is being tricked, and on stereo speakers that trick works beautifully. on mono, the trick falls apart.

the math is straightforward. for a delay of d milliseconds, the first cancellation occurs at f = 1000/(2d) Hz, and subsequent cancellations repeat at every 1000/d Hz. a 10 ms delay produces its first null at 50 Hz and repeats every 100 Hz. a 20 ms delay produces its first null at 25 Hz and repeats every 50 Hz.

these are not subtle artifacts. comb filtering from Haas delay can thin out the bass, add a hollow, metallic quality to the midrange, and make vocals sound nasal on every device that plays mono audio. and in 2026, mono playback is not rare. phone speakers, smart speakers, many bluetooth devices, and club systems all sum stereo to mono at some point.

allpass decorrelation

allpass decorrelation takes a completely different approach. instead of delaying the entire signal by a fixed amount, it shifts the phase of different frequencies by different amounts. the result is a signal that sounds nearly identical to the original on its own, but is mathematically different in a way your brain interprets as spatial separation.

what allpass filters do

an allpass filter passes all frequencies at the same level (hence the name) but changes the phase of each frequency by a different amount. the magnitude spectrum is unchanged: what goes in comes out with the same loudness at every frequency. only the phase relationship between frequencies changes.[^3]

this is a subtle but important distinction. your ears are largely insensitive to absolute phase. a single allpass filter on a mono signal sounds essentially identical to the original. but when you place the original in one channel and the allpass-filtered version in the other, your brain detects the phase differences and perceives width.

cascaded allpass: frequency-dependent group delay

a single allpass filter produces a smooth, gradual phase shift. it creates only mild decorrelation. the real power comes from cascading many allpass filters in series, each tuned to a different frequency. the combined effect creates a complex, frequency-dependent group delay: different frequencies are delayed by different amounts.

this is where the psychoacoustics get interesting. research has shown that decorrelation is not equally effective at all frequencies. the mid-frequency range (roughly 400 to 2000 Hz) is where your auditory system is most sensitive to phase differences between the ears. decorrelation in this range produces the strongest perception of width. at low frequencies, your head is too small relative to the wavelength for phase differences to matter much. at very high frequencies, phase sensitivity drops off again.[^4]

ERB spacing

the spacing of allpass filters matters. linear spacing (equal frequency intervals) gives too much resolution at high frequencies and not enough at low frequencies. this is because your auditory system does not resolve frequencies linearly. it uses something closer to the ERB (equivalent rectangular bandwidth) scale, derived by Glasberg and Moore in 1990, where each “band” corresponds to a fixed width on your basilar membrane.[^5]

spacing allpass filters on the ERB scale means more filters where your ears can detect finer differences (low to mid frequencies) and fewer where perception is coarser (high frequencies). the result is decorrelation that sounds natural and perceptually even across the spectrum, rather than heavy-handed in one range and weak in another.

the mono advantage

here is the key difference. when you sum an allpass-decorrelated stereo signal to mono, you do not get the deep, evenly-spaced notches of Haas comb filtering. the phase shifts partially cancel and partially reinforce, but the effect is mild. the mono sum sounds close to the original, with only slight frequency-response variations (typically less than 1 to 2 dB).

this is not perfect transparency. no widening technique is truly lossless in mono. but compared to Haas delay, where mono summing can destroy the low end and hollow out the midrange, allpass decorrelation is a dramatically safer option for modern production.

when to use which

the choice between Haas and allpass is not about which is “better.” it is about what you need.

use Haas delay when

• you want dramatic, obvious width. Haas delay creates a bold, wide sound that is immediately recognizable. if the effect needs to be a feature rather than invisible, Haas delivers
• mono compatibility is not critical. a sound effect, a background texture, or a doubled element that will not be featured on phone speakers
• you want directional bias. because one side arrives first, Haas delay creates a sense of direction. the sound feels like it leans toward the earlier channel. this is useful for creating spatial movement

use allpass decorrelation when

• mono compatibility matters. vocals, lead instruments, mix bus widening, anything that will play on a phone speaker or bluetooth device
• you want transparent, natural width. allpass decorrelation adds spaciousness without calling attention to itself. the source sounds like a wider version of the original rather than a stereo effect
• you need frequency-dependent control. ERB-spaced allpass cascades can target the mid-frequency range where widening is most effective, leaving the bass tight and the highs natural

the hybrid approach

some wideners combine both techniques. KERN WIDE’s HAAS mode, for example, uses a short delay for the primary width effect but constrains the output correlation to prevent the worst mono compatibility problems. the STEREO mode uses pure allpass decorrelation with a perceptually-weighted group delay profile.

the practical advice: start with allpass decorrelation for most production work. reach for Haas delay when you want something more dramatic and you have verified the mono fold-down.

practical tips

check the correlation meter

every widening technique pushes the stereo correlation downward. a correlation of +1.0 means the left and right channels are identical (mono). a correlation of 0 means they are completely unrelated. below 0, you have out-of-phase content that will cancel in mono.

for a full mix, keep the correlation above +0.3. individual tracks can dip lower, but the bus should stay safely positive. if your widening pushes the correlation below +0.3, back off the width or check the mono fold-down.

bass first, width second

low frequencies should almost always be mono or near-mono. bass energy that differs between left and right wastes headroom and creates phase cancellation on small speakers. apply a bass crossover (mono below 100 to 200 Hz) before adding width to ensure the low end stays tight and centered.

less is more

the first 20% of width adjustment does 80% of the perceptual work. a subtle widening that you notice when it is bypassed is better than an obvious effect that calls attention to itself. if the listener can hear that a widener is on, it is probably too much.

frequently asked questions

frequently asked questions

what is the Haas effect in stereo widening?

the Haas effect uses a short delay (typically 5 to 30 ms) between the left and right channels to create the perception of width. your brain hears both copies as a single sound but perceives it as coming from the earlier side. the delayed copy adds spaciousness without being heard as a distinct echo. the problem is that summing to mono causes comb filtering, which creates deep notches in the frequency response.

what is allpass decorrelation and how does it create width?

allpass decorrelation uses cascaded allpass filters to shift the phase of different frequencies by different amounts, creating a signal that sounds different from the original but has exactly the same frequency spectrum. when you place the original in one channel and the decorrelated copy in the other, your brain perceives the phase differences as spatial width. unlike the Haas effect, allpass decorrelation does not change the magnitude spectrum, so mono summing produces only mild coloration instead of deep comb filter notches.

is the Haas effect mono compatible?

no. Haas delay is one of the least mono-compatible stereo techniques. when you sum left and right, the delayed copy creates comb filtering: deep cancellations at evenly-spaced frequencies across the spectrum. a 10 ms delay produces nulls every 100 Hz. this can thin out the low end and create obvious tonal changes on phone speakers and club systems.

which is better for stereo widening, Haas or allpass?

it depends on your priorities. Haas delay is simple, dramatic, and immediately obvious, but it has serious mono compatibility problems. allpass decorrelation is more transparent, preserves the spectral balance, and sums to mono far more gracefully. for modern production where your mix will play on phone speakers and bluetooth devices, allpass decorrelation is generally the safer choice.

can you use both Haas and allpass decorrelation together?

you can, but be careful. stacking widening techniques adds complexity and increases the chance of mono compatibility issues. if you use Haas for obvious width on a specific track, check the correlation meter before adding allpass decorrelation on top. in most cases, one technique per track is enough.

references

a note from the developer

the Haas effect was one of the first stereo tricks i learned as a producer. duplicate a track, pan left and right, add a short delay to one side, and suddenly the vocal fills the room. it felt like magic. it also fell apart the moment i checked the mono fold-down.

that experience is what led me down the allpass rabbit hole. i spent months reading papers on decorrelation, building prototypes with different numbers of cascaded biquads, tweaking pole radii, and listening on every speaker i own. the goal was width that sounds natural, that does not change the tone of the source, and that survives mono playback without falling apart.

KERN WIDE’s STEREO mode uses 40 cascaded allpass biquads with center frequencies spaced on the ERB scale. the group delay profile follows a perceptually-weighted curve: strongest in the mid-frequency range where your ears are most sensitive to phase differences, tapering off at the extremes. a real-time correlation constraint prevents the output from ever pushing into dangerous territory.

full disclosure: i make KERN WIDE. the allpass techniques described in this guide are not proprietary. any developer can build a cascaded allpass widener. the difference is in the perceptual tuning: how the group delay profile is shaped, how the correlation constraint works, and how many hours of listening went into the parameter curves. try it on your mix and judge for yourself.

if you have a widening approach that works better for you, or if something here does not match your results, jonas@kernaudio.io. these guides improve with every conversation.