Why a Centered Vocal Sits Slightly to One Side

This page contains affiliate links. If you buy through one we may earn a commission, at no cost to you.

Usually it is not the song. The lead vocal is the only element in a normal mix whose position you have an exact expectation of, because it is almost always dead centre, which means it is the only one that can show you an imbalance that is affecting absolutely everything. The vocal is not off centre. The vocal is the ruler.

And the imbalance needed is tiny. Lateralisation saturates at roughly 10 decibels of difference between your two ears, so one decibel moves the image a tenth of the way across your head. One decibel is 12 percent more pressure in one ear. Ordinary manufacturing tolerance on two drivers supplies half of it before you have put them on.

The conversion that makes all of this easy

Your auditory system places a sound laterally using, among other things, the level difference between the two ears. Feed both ears the same thing and the image is in the middle of your head. Increase one side and the image moves toward it, roughly in proportion, until at about 10 decibels it is pinned at that ear and more difference changes nothing.

That gives a conversion worth memorising: multiply the decibel difference by ten to get the percent of the way from centre to one ear. A 0.5 decibel imbalance is 5 percent off centre. Two decibels is 20 percent. And the smallest interaural level difference most listeners can detect at all is somewhere around half a decibel, so the threshold of noticing and the threshold of mattering are nearly the same place.

Then look at what a decibel is in physical terms, because this is where the intuition fails. One decibel is 10 to the power of one twentieth in pressure, which is 1.122, and 10 to the power of a tenth in power, which is 1.259. So an image you would describe as very slightly left is one ear receiving 12 percent more sound pressure and 26 percent more acoustic power than the other. Those are not large numbers in any manufacturing sense. They are the sort of numbers that fall out of two components built on the same line on the same day.

The drivers alone predict it

A dynamic driver's sensitivity depends on its force factor, the product of the magnet gap's flux density and the length of wire in it, divided by the moving mass. Every one of those varies a little between units: magnets vary in remanence, gaps vary in width, coils vary in turns and in glue mass, diaphragms vary in thickness.

Suppose two drivers end up 5 percent apart on force factor, which would be a tight pair. That is 20 times the base ten logarithm of 1.05, which is 0.42 decibels, which is 4.2 percent off centre. At 10 percent apart it is 0.83 decibels and 8.3 percent off centre. So a perfectly healthy pair of headphones with no fault of any kind sits a few percent off centre, and a merely acceptable pair sits most of a tenth of the way across.

Makers who care about this sort their drivers into matched pairs and publish a channel balance figure, usually as plus or minus one or plus or minus three decibels. Nobody we carry publishes one. Three of our own records say as much in their own drawbacks, which is the honest way to handle it: a Poly Blackwire 3325 stereo USB headset lists among its weak points "No frequency response, driver size or mic pattern published", and its record goes further and refuses to dress the gap up: What Poly does not publish is the problem. Enhanced Audio is described only as Poly signature audio quality and reliability. That is exactly right, and it means the figures on this page are derived rather than read. Our Blackwire 3325 headset review has the rest.

The fit supplies more than the drivers do, and it changes every time

Now the variable that is yours rather than the factory's, and it is the larger one.

A sealed in-ear tip and your ear canal form a closed cavity. At low frequencies the driver is not radiating into open air at all, it is squeezing a trapped pocket of it, so the pressure it generates goes as one over the volume of that pocket. Make the pocket smaller and the bass gets louder, by exactly the ratio of the volumes.

Put numbers on it. An ear canal of about 7.5 millimetres bore has a cross section of 44.2 square millimetres. Call the residual volume between the tip and the eardrum 800 cubic millimetres. Push one tip in a single millimetre further than the other and you have removed 44.2 cubic millimetres, which is 5.5 percent of the pocket, which is 20 times the log of 1 over 0.945, or 0.49 decibels. Two millimetres is 11.1 percent and 1.02 decibels.

So a couple of millimetres of asymmetry, which is less than you can feel, is a tenth of the way across your head. And it only does this to the bass, because the sealed cavity argument only applies where the wavelength is long compared with the canal. The treble is radiated and reflected rather than compressed, so it barely moves. That is why the experience is so specific and so confusing: the voice and the bass drift toward one side while the cymbals stay put, which no amount of staring at the mix will explain.

And it is the reason the problem comes and goes. If the driver spread and the fit asymmetry happen to point the same way you get 1.84 decibels and an image 18 percent off centre. If they oppose you get 0.19 decibels and 2 percent, which nobody notices. Same headphones, same track, same ears, two sessions, two answers.

Which brings up a product designed to make that asymmetric on purpose. A two in one set of flat sleep earbuds for side sleepers exists specifically so that one earpiece can be pressed into a pillow all night, and its record is clear about the geometry: the buds are "ultra-thin and lightweight", the fit is "secure but pressure-free, resting gently in the ear without the deep insertion that feels intrusive at bedtime", and "Soft ear tips form a gentle seal". Read that with the arithmetic above. A gentle seal with shallow insertion is a large residual volume, and a pillow pressing on one side of your head for six hours reduces it on that side only. The record never mentions that this will shift the balance, and for white noise and sleep stories it genuinely does not matter, which is a fair reason for the omission. Worth knowing anyway if you also listen to music in bed. Our sleep earbuds review has the fit detail.

Your two ears are not identical either, and that is not a fault

The ear canal behaves as a tube closed at one end, so it has a quarter wave resonance at the speed of sound divided by four times its length. At 343 metres per second a 25 millimetre canal rings at 3,430 hertz and a 27 millimetre canal at 3,176 hertz, which is 7.4 percent apart. Nobody's two canals are the same length, and that resonance is worth roughly 10 decibels of gain at its peak, so two canals of slightly different length apply slightly different gain curves to the same signal.

The consequence is that the offset you hear may be frequency dependent in a way no headphone is responsible for. A vocal with most of its presence energy around 3 kilohertz is sitting right where the two canals disagree most. This is worth knowing mainly because it tells you when to stop hunting for a fault, and because there is a ten second test that separates it from everything else.

The free test, which beats every purchase in this article

Swap the earpieces. Put the left one on your right ear and the right one on your left, or simply rotate a headband 180 degrees and wear it backwards for ten seconds. Then listen to where the vocal is.

If the image stays on the same side of your head, the cause is downstream of the hardware: your own ears, or the recording. If the image jumps to the other side, the cause is the hardware, and you have identified it exactly, with no meter and no purchase.

A Poly Blackwire 3310 monaural USB headset is useful here as the end of the scale rather than as a cure. One earpiece is an infinite level difference, so the image is fully at one ear by construction, and its record puts the trade plainly: You give up stereo imaging and any pretence of media listening, and you gain situational awareness. It is also the one headset in this article that ships with the swap test built in, because its "180 degree pivoting speaker" exists so that the single earpiece can be worn on either ear. It is the same headset as the 3325 above, from the same range, with the same boom mic, differing in channel count and in having no 3.5 mm socket. If you want to hear what 10 decibels and beyond sounds like, borrowing one for a minute will tell you more than any description. Our Blackwire 3310 review covers the contact centre case it is actually for.

Two more tests that are free. Play something you know to be mono, or switch on the mono audio toggle that both major desktop and mobile platforms carry in their accessibility settings. A mono signal is identical in both channels by construction, so any offset you still hear is everything downstream of the file, which removes the recording from the suspect list in one click. And then, if you have established that the fault is in the hardware, the same accessibility panels carry a left and right balance slider. Nulling a 1 decibel imbalance means trimming the loud side by about 11 percent of its amplitude, which is a small nudge on that slider, and it is a complete fix that involves buying nothing.

The case where the seal is not a variable at all

There is one category where the whole fit argument above simply does not arise, and it is worth understanding as a trade rather than as a solution. Baseus Bowie MC2 Air clip-on open-ear earbuds work by not sealing anything. Their record is explicit: Instead of a silicone tip pushed into your ear canal, soft air cushions and an extended ultra-thin wave-shaped C-ring flex to different ear shapes, and "nothing seals the ear".

No seal means no trapped cavity, which means the one over volume relation has nothing to act on, which means insertion depth cannot produce a bass imbalance. The driver spread still can, and so can your canals, but the largest and most variable of the three terms is gone. The honest other half is in the same record and the record states it without being asked: bass will never hit as hard as a sealed tip delivers. There is less bass to be imbalanced. That is the whole trade, and it is the same one our page on open-ear versus sealed headphones frames correctly as isolation against awareness, with the imaging consequence simply not mentioned on either side. Our Bowie MC2 Air review has the specification.

The honest limit, and when it really is the record

The limit is blunt. The single number this entire page is about, channel balance tolerance in decibels, is published by none of the five audio products touched here, and in the wider catalogue it is published by none at all. Driver sensitivity is not published either. Neither is frequency response. So the 0.42 and 0.83 decibel driver figures are computed from assumed tolerances on the force factor, the insertion figures are computed from an assumed canal bore and residual volume, and all of them are labelled as such in the drawing. What the records do publish, to their credit, is that they publish nothing: three of the four say so in their own drawbacks rather than inventing a figure, and that is the behaviour to reward.

And yes, sometimes it really is the song. A lead vocal can be deliberately placed a few degrees off centre, a double tracked lead can be panned narrow rather than stacked, and a mono vocal with a reverb return that is wider on one side will read as offset even though the dry voice is centred. Mixes from analogue sources can carry a genuine interchannel time offset from head alignment, which the ear reads as position just as readily as it reads level. None of that is detectable by listening harder. It is detectable by swapping the earpieces, which takes ten seconds and settles the question completely, and it is the only diagnostic in this article that works on the first attempt.

One thing not to do: raise the volume. Interaural level difference is a ratio, so it survives every volume change you can make. The image stays exactly where it was and everything gets louder, which is the one outcome nobody wanted.