The Auditory System

The obstacle: a wave in air, a receptor in fluid

Begin with the physics, because unlike vision, where the puzzle is what the brain adds, the puzzle in hearing is how anything gets in at all.

A sound is a travelling disturbance in the pressure of a medium: molecules crowd together and thin out, and the pattern propagates. Your ear must detect it. But the sensory cells cannot simply sit in the air. They are delicate, they must be bathed in a tightly controlled ionic solution to work at all, and the precise mechanical arrangement they depend on could not be maintained in a draught. They sit, therefore, in fluid, sealed inside the densest bone in the body.

And that is where the trouble starts. Air and water differ enormously in impedance, which is a medium's resistance to being set in motion. Air is light and springy: a small force moves a lot of it a long way. Water is heavy and nearly incompressible: the same force barely budges it. When a wave meets a boundary between two media of very different impedance, most of its energy is reflected rather than transmitted. Shout at a pond and the water is essentially undisturbed; nearly all of the sound comes straight back at you.

You have felt the consequence. Under water, with air trapped in your ears, hearing is dull and muffled and direction is nearly impossible to judge: the apparatus is now being asked to pass sound from water into air-filled ears built for the reverse, and it fails. Fish, which never had this problem, never evolved a middle ear. They do not need one. Their sensory cells and the water around them have essentially the same impedance, and sound passes into a fish more or less as though the fish were not there.

So a land animal that wants to hear must find a way to force the energy of an airborne wave into a fluid-filled chamber without losing it at the boundary. That is not a biological problem. It is an engineering problem, with a known class of solutions, and the middle ear is one of them.

The middle ear is a pressure amplifier, not a bridge

The standard shorthand describes the middle ear as a set of bones that "carry vibrations from the eardrum to the inner ear". That description is not false, and it is nearly useless, because it makes the ossicles sound like a piece of plumbing whose only virtue is being connected. If mere connection were the requirement, one bone would do, or a strand of cartilage, or nothing at all: the eardrum could simply abut the fluid. The reason the middle ear has the particular form it has is that connection is not the requirement. Amplification of pressure is the requirement, and the anatomy is shaped to deliver it.

Two mechanisms are at work, and one of them does most of the job.

Combine them and the middle ear delivers a large increase in the pressure applied to the inner ear fluid relative to the pressure that arrived at the eardrum, and that gain is very close to what is needed to compensate for the impedance mismatch. The system is not merely adequate. It is matched to the physical problem, and that is what makes it satisfying: you can predict the shape of the anatomy from the physics, and then go and find that shape.

Two further details confirm the reading, because they only make sense on it.

1. The Eustachian tube exists to keep the eardrum free to move

   An amplifier that starts with a membrane requires that membrane to be able to move. If the pressure in the middle-ear cavity does not match the pressure outside, the eardrum is pushed taut and stops moving properly, and the whole gain is lost. The Eustachian tube, connecting the middle ear to the throat, opens when you swallow or yawn and equalises the two. This is exactly why your hearing goes dull as an aeroplane descends, and why swallowing restores it with a pop. Nothing has changed in the cochlea. The first stage of the amplifier has been detuned and then retuned.

2. Two tiny muscles can turn the gain down

   The tensor tympani and the stapedius, the smallest skeletal muscles in the body, attach to the ossicles and can stiffen the chain, reducing the transmission of sound. They contract reflexively to loud noise, and also just before you speak or chew, which is why your own voice does not deafen you from the inside. This is a volume control on the amplifier, and an amplifier is the only thing that needs one. The reflex is also slow relative to a sudden impulse, which is part of why gunfire and explosions damage ears that ordinary loud sustained sound would not.

3. And a third window is needed because fluid does not compress

   Push the stapes into the oval window and the fluid inside has nowhere to go, since fluid is essentially incompressible and the cochlea is walled in bone. There must be a second flexible opening for the fluid to bulge out of, and there is: the round window, which moves outward as the oval window moves in. Block it and hearing suffers, not because any sensor was touched, but because the fluid can no longer flow, and if the fluid cannot flow the membrane inside cannot be displaced.

Everything in the middle ear, then, is subordinate to a single job: get the energy of an airborne wave across the air-fluid boundary without losing it. Damage this apparatus and you produce conductive hearing loss, in which the inner ear and the nerve are perfectly healthy and the sound simply cannot reach them. This is why a conductive loss can often be corrected: the sensor is intact, and it is enough to route the energy around the broken amplifier. A bone-conduction hearing aid does exactly that, vibrating the skull directly and bypassing the middle ear altogether. When the inner ear itself is damaged, no such trick is available, and the reason for that asymmetry is the subject of the sections that follow.

The cochlea performs a Fourier analysis with no neurons

The stapes now pushes on the oval window and sets the cochlear fluid moving. What happens next is the most remarkable thing in the auditory system, and it is remarkable because it is not neural at all. It is a piece of mechanics.

The cochlea is a fluid-filled tube, coiled like a snail shell, running from the oval window at its base to its far tip, the apex. Down the middle of that tube runs the basilar membrane, carrying the sensory cells on its back. And the crucial fact about the basilar membrane is that it is not the same all the way along.

Note the counterintuitive geometry, since it is a common source of confusion: the cochlear tube tapers, growing narrower toward the apex, but the membrane inside it does the opposite, growing wider and much less stiff. The membrane, not the tube, is what matters.

Now think about what a stiff, narrow strip does and what a floppy, wide one does. A short taut string on a guitar resonates at a high pitch; a long slack one resonates at a low pitch. The same holds along the basilar membrane. When the stapes drives the fluid, a wave travels along the membrane from base to apex, and as it travels it grows in amplitude, reaches a peak, and then dies away rapidly. Where it peaks depends on its frequency. A high-frequency sound peaks early, near the stiff base. A low-frequency sound travels much further before peaking, near the floppy apex.

That is the whole trick, and its consequences are large. A complex sound, a voice, a chord, a slammed door, containing many frequencies at once, sets up peaks at several places along the membrane simultaneously. The cochlea has taken a single mixed-up pressure signal, a jumble of frequencies arriving one after another in time, and laid it out as a pattern in space: this frequency here, that one there. In mathematical terms it has performed a Fourier analysis, decomposing a waveform into the frequencies that make it up. In biological terms it has done so with a strip of tissue and a stiffness gradient, and not one neuron has yet fired.

This orderly arrangement of frequency along a structure is called tonotopy, and once you know to look for it you find it everywhere upstream. The auditory nerve fibres are arranged tonotopically. So are the cochlear nuclei, the inferior colliculus, the medial geniculate nucleus of the thalamus, and the auditory cortex itself. A map established mechanically in the inner ear is copied, station by station, all the way to the surface of the temporal lobe. It is the direct counterpart of the retinotopic map in vision, with one difference worth savouring: the retina's map of space is simply a map of the receptor sheet, whereas the cochlea's map of frequency was manufactured, by a gradient of stiffness, from a signal in which frequency was not laid out in space at all.

The travelling wave was demonstrated directly by Georg von Bekesy, who observed the motion of the basilar membrane in cochleae and traced how the peak of the wave moved with frequency, work summarised in his Experiments in Hearing and recognised with the Nobel Prize in Physiology or Medicine in 1961. His preparations, being dead tissue, showed a travelling wave with a peak that was real but rather broad. Living cochleae turn out to be far more sharply tuned than his measurements suggested, and the gap between his results and reality is not an error. It is a clue, and the next two sections cash it in.

Hair cells: why hearing had to abandon chemistry

Sitting on the basilar membrane, in the structure called the organ of Corti, are the receptor cells. They are called hair cells, and the hairs are the point.

Each cell carries a bundle of stiff projections on its upper surface, the stereocilia, arranged in ranks of increasing height like a staircase. Above them lies a stiff overhanging shelf, the tectorial membrane. When the basilar membrane is displaced by the travelling wave, the geometry of the arrangement causes the bundle to be sheared sideways, and the stereocilia are deflected.

Now comes the detail that makes hearing possible, and it repays close attention. Running from the tip of each shorter stereocilium to the side of its taller neighbour is a fine filament, the tip link. The tip link is attached to a mechanically gated ion channel. Deflect the bundle toward the tall side and the links are stretched. A stretched link pulls, physically, on the gate of the channel, and the channel opens. Ions flow in, the cell depolarises, and it releases transmitter onto the auditory nerve fibre beneath it. Deflect the bundle the other way and the links slacken, the channels shut, and the cell hyperpolarises.

Compare this with vision, and the design logic becomes obvious. In a photoreceptor, a captured photon triggers a cascade of enzymes and second messengers, and only at the end of that chain does an ion channel change state. That cascade is why the retina can detect a single photon: chemistry gives you enormous amplification, one molecule triggering many. But it is also why it is slow, and why vision has a sluggish temporal resolution, the reason a sequence of still frames can be made to look like continuous motion.

Hearing cannot afford that. Consider what it is being asked to do. Audible sound extends to frequencies of many thousands of cycles per second, so the membrane at some places is oscillating thousands of times a second, and, as the next-but-one section shows, the brain resolves differences between the two ears down to the scale of tens of microseconds. No cascade of enzymes runs on that timescale. There is not time for a second messenger to be made, to diffuse, and to act.

So hearing throws the chemistry away. The stimulus is coupled to the gate by a piece of string. The delay is essentially the time it takes to pull a taut filament, which is as close to instantaneous as biology gets. The tip link exists because hearing has to be fast, and any mechanism with a messenger in it would be too slow. The whole peculiar architecture of the hair cell, the bundle, the staircase, the tectorial shelf, the string, follows from that one requirement.

Hair cells come in two populations, and they are usually introduced as a pair. They are not a pair. They do entirely different jobs.

The ear is not only a receiver: it emits sound

Return to the loose end left with von Bekesy. His dead cochleae showed a broad, sluggish travelling wave. A living cochlea is exquisitely sharp: it can distinguish frequencies that a passive strip of tissue with that much damping could never separate, and it can detect vibrations of the membrane so small that the pressures involved are close to the noise floor set by the random motion of air molecules. A passive mechanical system cannot do this. Damping in a fluid-filled tube is heavy, and a passive resonance in such a system is inevitably blunt.

The resolution is that the living cochlea is not passive. It is an active amplifier, and the outer hair cells are the amplifying element.

Outer hair cells do something no other cell in the body does: they change length in response to changes in their own membrane voltage, and they do it fast enough to keep up with the sound. As the travelling wave depolarises and hyperpolarises them cycle by cycle, they shorten and lengthen cycle by cycle, and because they are mechanically coupled to the structures around them, they push. They push in phase with the wave, at exactly the place where the wave is already peaking, and in doing so they feed mechanical energy back into the basilar membrane.

Now derive the consequences, because a claim this strange should be made to pay for itself, and it pays three times over.

And there is a final, harder fact. In mammals, hair cells lost in adulthood are not replaced. Birds and fish regenerate theirs after damage; mammals lost the capacity, and why they lost it is not fully understood. Every hair cell you were born with is a hair cell you must make last. Damage is cumulative, and it depends on duration as well as on level: a long shift in a moderately loud factory can cost as much as a brief blast. This is the whole reason noise-induced hearing loss is permanent, and the reason the only genuinely effective intervention is not treatment but prevention.

Two cues, because one is not enough

The cochlea has told the brain what frequencies are present. It has said nothing whatever about where the sound came from, and it cannot: the basilar membrane's map is a map of frequency, and there is no space in it. Vision gets location free, because the retina is a spatial sheet and a point in the world lands on a point on the sheet. Hearing has no such sheet. Location must be computed, from scratch, by comparing what arrives at the two ears, and this is why the auditory brainstem is so much more elaborate than its visual counterpart.

The brain has two cues available, and the crucial insight, and the point of this section, is that neither one works across the whole range of hearing. Each fails precisely where the other succeeds, and the system therefore uses both. This is the duplex theory.

Set the two side by side and the design becomes inevitable. The time cue is good at low frequencies and fails at high. The level cue is good at high frequencies and fails at low. Together they cover the range, and the brain has built separate machinery for each. Neither cue is a backup for the other; each is the only option in its own band.

The time cue deserves a closer look, because on the face of it the brain should not be able to use it at all. An action potential lasts on the order of a millisecond. The interaural time differences the brain must resolve are on the order of tens of microseconds, which is to say, a small fraction of the duration of a single spike. The system is extracting a timing difference far finer than the width of the events it is made of. How?

1. Fire in step with the wave

   At low frequencies, auditory nerve fibres do not fire at random within the sound. They fire at a consistent point in the cycle of the wave, a behaviour called phase locking. Each spike therefore carries a timestamp, and the timestamp is far more precise than the spike is long, because what matters is not the width of the spike but the reliability of the moment it starts.

2. Send both ears to one place

   Fibres from the two ears converge on the medial superior olive, low in the brainstem, and they converge on the same individual neurons. This is why the pathway must cross early, and it answers a question the next section makes explicit: comparison requires convergence, and you cannot compare two ears in structures that never meet.

3. Make each neuron a coincidence detector

   A medial superior olive neuron fires strongly only when its two inputs arrive together. It is not measuring a delay. It is answering a much simpler question, which biology is very good at: did these two spikes land at the same instant?

4. Vary the wiring, and the delay becomes a place

   Now make the axons feeding these neurons different lengths, so that each neuron has a different built-in delay on one side. A neuron whose left input is delayed by 30 microseconds of extra axon will see coincidence only when the sound genuinely arrived at the left ear 30 microseconds early. So each neuron is tuned to a particular interaural delay, and therefore to a particular direction. Which neuron is firing tells you where the sound is. The brain has converted an impossible timing measurement into a question about which cell in a row is active, and that is the sort of question a nervous system can answer easily.

It is worth pausing on how good this is. The precision achieved by this circuit is among the finest in all of neurobiology, and it is achieved not by making neurons faster, but by arranging slow neurons cleverly, so that a fine temporal question is re-posed as a coarse spatial one.

Both cues share one blind spot, and it is a geometric consequence rather than a flaw in the machinery. Consider a sound directly in front of you and a sound directly behind you. Both are equidistant from the two ears. Both therefore produce zero time difference and zero level difference, and no amount of neural cleverness can distinguish them, because the information is not there. The same holds for a whole surface of points sharing the same pair of values: the cone of confusion. Two solutions exist, and you use both, constantly, without noticing. The first is the outer ear itself: the folds of the pinna reflect and filter incoming sound differently depending on whether it arrives from above, in front, or behind, stamping a subtle frequency signature onto it that the brain learns to read, which is why localisation degrades when the pinna is bypassed by headphones. The second is simply to move your head. Rotate slightly and a sound in front behaves quite differently from a sound behind. That small involuntary turn toward an unexpected noise is not idle curiosity. It is a measurement, taken to break a symmetry that the ears alone cannot break.

Why the auditory pathway crosses so early and so often

Set the auditory pathway next to the visual one and the difference is immediately striking. Vision is almost brutally direct: retina, one crossing at the chiasm, thalamus, cortex. Hearing is nothing of the kind. It has multiple stations in the brainstem, crossings at several levels, and a great deal of information passing back and forth between the two sides before anything reaches the forebrain at all. The temptation is to record this as a list of names. The better question is why the architecture differs, and the previous section has already supplied the answer.

Vision arrives pre-localised and hearing does not. A visual location is given by which part of the retina was struck, and the brain can therefore afford to send the signal upward with minimal processing. An auditory location exists nowhere in the signal from either ear; it exists only in the relationship between the two. So the two ears must be brought together, and they must be brought together early, while the timing information is still fresh and phase locking is still intact. Every microsecond of extra processing before the comparison risks blurring the very quantity being measured. The elaborate, heavily crossed brainstem of the auditory system is not a historical accident. It is the direct anatomical consequence of a computation that requires both ears at once.

One clinical consequence follows immediately from all this crossing, and it is worth stating because it inverts the familiar rule. In vision, damage to one side of the brain produces a clean loss of one half of the world. In hearing, because each ear's information reaches both hemispheres after the superior olive, damage to the auditory cortex on one side does not make you deaf in the opposite ear. Total deafness of one ear points to the ear or the nerve, before the crossing. A one-sided cortical lesion instead produces subtler deficits, most notably in localising sounds and in picking a voice out of a crowded room, and these deficits make sense on exactly the account above: what a single hemisphere uniquely holds is not a half of the auditory world, but a set of comparisons.

Auditory cortex: from frequency to meaning

The medial geniculate nucleus projects to the primary auditory cortex, on the upper surface of the temporal lobe, largely hidden within the Sylvian fissure on a ridge called Heschl's gyrus. See the brain lobes page for where this sits in the wider geography.

The map the cochlea built survives to the end. Primary auditory cortex is tonotopic: move across its surface and the preferred frequency of the neurons shifts smoothly, low at one end, high at the other. The gradient of stiffness in a strip of tissue in the inner ear is, in a real sense, still legible on the surface of your temporal lobe. But a frequency map is not hearing, any more than a map of edges in V1 is seeing, and beyond the primary area the same logic that governed the visual system reasserts itself: successive stages combine simple features into more abstract ones, and neurons come to respond not to tones but to combinations, to changes over time, to the sweeps and bursts and harmonic stacks that make up real sounds. A sound in the world is almost never a pure tone. It is a structured event in time, and cortex is where time-structure becomes an object.

The two-stream organisation appears here too, and it is not merely an analogy imported from vision. A ventral, anterior pathway running forward in the temporal lobe is concerned with what a sound is: identifying a voice, a word, an instrument, a dog. A dorsal, posterior pathway running toward the parietal lobe is concerned with where it is and with using it to guide action. The reappearance of the same division in a completely different sense, arriving in cortex by a completely different route, is one of the better arguments that the split reflects something deep about what brains are for, rather than a quirk of any one system.

Speech is where all of this culminates and where hearing stops being a physics problem. The auditory input for the word "cat" varies enormously between a child and an adult, a whisper and a shout, a Glasgow accent and a Texan one: the physical waveforms have little in common. Yet all are heard as the same word. Extracting a stable linguistic category from a wildly variable acoustic signal is a problem of exactly the same shape as recognising a face across changes of lighting and angle, and it is solved the same way, by a hierarchy that discards the irrelevant variation and keeps the invariant. The regions that do this cluster around the superior temporal cortex and connect to the classical language areas, and they are treated in full on the language and the brain page.

Finally, the auditory cortex demonstrates its own necessity, in the same way the visual cortex does, by failing. In cortical deafness, both ears work, the cochleae respond, the nerve fires, the brainstem passes signals up, and the patient cannot hear. In the more selective condition of auditory agnosia, the patient hears sounds perfectly well and cannot recognise them: they hear that someone is speaking without being able to extract words, or hear a phone ringing without knowing what it is. The ear is not the hearing. The ear is a magnificent piece of engineering that delivers a well-formed frequency code to a brain, and the hearing happens afterwards.

What hearing is not

What remains contested

The engineering account above is solid at its base and genuinely unsettled at its edges.

How pitch is actually coded. Place along the basilar membrane is one code, and the timing of phase-locked spikes is another, and both plainly carry information. Which one the brain relies on, and where in the range each dominates, has been argued for over a century. Place theory struggles to explain the fine pitch discrimination people achieve at low frequencies, where the membrane's peak is broad; timing theory struggles at high frequencies, where phase locking fails. Most modern accounts use both and disagree about the boundary.

The molecular identity of the transduction channel. The tip link and the mechanically gated channel have been understood in outline for decades, but the exact identity of the channel protein, and how the tip link couples to it, has been contested and revised repeatedly, and the field has not fully closed the question.

Why mammals cannot regenerate hair cells. Birds and fish do it routinely. Mammals lost the ability, and it is not clear why, nor whether the block can be lifted. Considerable effort is going into trying, and it would transform the treatment of deafness if it succeeded. It has not yet.

How a voice is picked out of a crowd. The ability to attend to one speaker in a noisy room is easy for a healthy listener and remains difficult to explain, and harder still to replicate in a machine. It clearly involves the tonotopic separation performed by the cochlea, but also attention, prediction, memory, vision, and the direction cues computed in the brainstem, and how they combine is not settled.

Related topics