A stereo mix is built on two assumptions. The first is that your right ear hears a delayed and quieter version of the left channel through the air. The second is that the room you mix in has a roughly known frequency response. Headphones break both. The left transducer fires into the left ear with effectively zero crosstalk, and the "room" is now a sealed cavity a few millimetres deep. That is why headphone mixes so often sound exciting at home and brittle in the car.
This is not an argument against headphone mixing. It is an attempt to be precise about what actually changes when you put cans on, what you lose, what you gain, and the order of corrections that work.
What speakers give your ears that headphones do not
Sit in a stereo triangle and the left speaker reaches both ears. The contralateral signal arrives later and quieter than the ipsilateral one, and it is filtered by your head, torso and outer ear on the way around. The brain uses those differences to externalise the sound and place it in space. The two relevant cues are the interaural time difference (ITD) and the interaural level difference (ILD), plus the high-frequency rolloff from head shadow above roughly 1.5 kHz.
For a 60 degree stereo triangle the contralateral delay sits in the hundreds of microseconds and the level drop is several decibels. Sonarworks cites figures in the 270 to 300 microsecond range for the delay and 6 to 9 dB for the attenuation, which is consistent with textbook ITD and ILD values for that geometry (Sonarworks). On headphones, all of this disappears. You get a clean, isolated signal in each ear.
The audible consequences are predictable. Phantom-centre material sits inside your head instead of in front of it. Hard-panned elements feel detached. Reverb tails sound bigger than they really are because there is no room reflection arriving from the wrong side to mask them. Stereo width is, in a word, lying to you.
What headphones do better
The flip side is real and worth using. Because each transducer is a sealed signal path with no room in between, headphones expose everything speakers smear over. Edit clicks, pops, mouth noise, low-level reverb tails, tiny phase issues, and any left or right asymmetry become obvious. This is why mastering engineers routinely do a final QC pass on headphones even when the mix decisions were made on speakers (LANDR, HEDD Audio).
Headphones are also room-independent. A pair of HD650s sounds the same in a hotel room, on a train, and in a treated studio, give or take leakage and background noise. That is a significant advantage for anyone working without a controlled space.
So the practical division is roughly: balance and low-end on speakers, surgical work and translation checks on headphones. If headphones are the only option, you need to close the gap with tools.
Crossfeed, where it comes from
The idea of feeding part of each channel into the opposite ear, delayed and filtered, is not new. Benjamin Bauer of CBS Laboratories published "Stereophonic Earphones and Binaural Loudspeakers" in JAES in April 1961. He described two passive networks that lift the high frequencies on the direct channel, apply a complementary cut on the crossfed channel, and delay the crossfed signal by roughly 0.4 ms below 1 kHz (The Well-Tempered Computer, AES e-Library). Every modern crossfeed plugin descends from this basic shape, including the free open-source BS2B library (bs2b.sourceforge.net).
Modern implementations swap the passive network for IIR or FIR filters and add a speaker-angle control. Goodhertz CanOpener Studio lets you pick a crossfeed amount and an angle: smaller angles below 30 degrees place the image more in front, larger angles above 45 degrees feel more immersive, and the factory Mix Engineer preset sits at 100 percent crossfeed and 60 degrees (Goodhertz manual). Sonarworks SoundID Reference includes a virtual monitoring section with similar controls.
Two things to keep in mind. Crossfeed belongs on the listening chain, not on the master bus. Bouncing a crossfeed-processed mix is a category error. And crossfeed assumes the headphone you are listening on is already neutral, which leads to the next problem.
Calibration, and why it usually matters more
Real headphones do not have flat frequency responses. They are tuned to a target curve, typically Harman or diffuse-field, with peaks of several dB in the 3 to 9 kHz region driven by the target and by driver-specific resonances. A pair of HD650s tuned to a diffuse-field target and a pair of Beyerdynamic DT 770s tuned closer to a free-field target can differ by 4 to 6 dB at some frequencies. Two engineers using "studio reference" headphones can therefore make opposite EQ decisions on the same source. Crossfeed does nothing about this.
Calibration software measures the headphone model and applies an inverse-EQ correction in real time. Sonarworks SoundID Reference uses its patent-pending PAPFR (Perceived Acoustic Power Frequency Response) methodology, which measures multiple physical units of each model and averages them, with all measurements referenced to 83 dB SPL (Sonarworks PAPFR). Community measurement projects like AutoEq and oratory1990 publish free EQ profiles for popular models. There is no published head-to-head comparison between the two approaches, so claims about which is more accurate should be read carefully.
The order that works in practice is calibration first, then crossfeed or virtual monitoring on top. If the headphone's response is wrong, the spatial layer just builds on a coloured foundation.
Panning, level, reality checks
A few practical points that follow from the physics:
- Back off extreme hard pans. If something feels right at 100 percent left or right in headphones, it will often feel timid on speakers. Sonarworks suggests around 70 percent as a more translatable maximum; the exact number depends on the source (Sonarworks panning). The instinct is the opposite of what you want.
- Keep the level moderate. Sustained loud listening shifts your perceived loudness curves and pushes you into bad mid-balance decisions. The 83 dB SPL reference used by calibration tools is a defensible target.
- Check on something else before bouncing. Laptop speaker, phone speaker, car, mono-summed Bluetooth speaker. The point is not the quality, it is the difference.
- Closed-back versus open-back. Closed-back designs trap pressure behind the driver and tend to emphasise sub and low-mid. Low-end EQ decisions made on closed-back cans without calibration often translate poorly. Open-back designs vent that pressure and behave more linearly down low (Sweetwater).
The least obvious part
If you are going to add one tool to a headphone mixing chain, it is probably not crossfeed. It is calibration. The reasoning is geometric: tonal accuracy is a precondition for every other decision. Pan a snare with the wrong 4 kHz response and the panning was never the problem. Crossfeed makes the experience more pleasant and more like speakers, but it does not fix the spectrum sitting underneath it.
Treat headphones as one instrument in the chain, not as a substitute for the chain. They are excellent at exposing problems, weak at confirming balance, and entirely dependent on the model in question. The fixes exist. The order is what tends to get reversed.