Place · Level 3
Hearing & the Cochlea
耳蜗把频率摊成一条线 · 毛细胞不长回来, 所以噪声剂量是只出不进的账
Last updated
Story path
- 1The cochlea spreads pitch along a lineThe cochlea spreads pitch along a line
- 2One row listens, three rows amplifyOne row listens, three rows amplify
- 3Hair cells don't grow back · a balance, not a flowHair cells don't grow back · a balance, not a flow
- 4High notes go first · you hear it, you can't parse itHigh notes go first · you hear it, you can't parse it
- 5Noise dose · every +3 dB halves the timeNoise dose · every +3 dB halves the time
Chapter 1
The cochlea spreads pitch along a line
The cochlea spreads pitch along a line
Once sound reaches the inner ear, it enters a bony tube coiled like a snail shell — the cochlea. Uncoiled, that tube runs roughly 33 to 35 mm long, and it varies noticeably from person to person. Running along its length is a strip called the basilar membrane. Everything that follows happens on that strip.
The important thing about the strip is that it is not uniform. At the end nearest the incoming sound (the base) it is narrow and stiff; the further in you go toward the apex, the wider and more flexible it becomes.
When a sound arrives, a traveling wave runs along the membrane from base toward apex, growing in amplitude and slowing as it goes, until it reaches a point of maximum displacement. Where it peaks depends on frequency: high frequencies peak at the base, low frequencies at the apex.
So frequency is turned into place. A membrane whose width and flexibility vary systematically vibrates maximally at different positions for different stimulus frequencies, which turns the whole strip into a map running from high pitch to low — this is tonotopy.
Note when this happens: the splitting is mechanical, and it happens before hair cells convert vibration into neural signals. The cochlea is not only an amplifier and a transducer — it is also a mechanical frequency analyzer that decomposes a complex waveform into simpler elements before anything is handed to the nerve.
Do not over-credit the passive gradient, though. The textbook's own wording is that frequency tuning is attributable in part to the geometry of the basilar membrane — it lays down the broad map; the rest is on the next page.
Remember this line. Damage and trouble following speech can both be located back onto it: which stretch of the line is in trouble. Tinnitus often is not on this line at all — that one gets its own scene later.
The important thing about the strip is that it is not uniform. At the end nearest the incoming sound (the base) it is narrow and stiff; the further in you go toward the apex, the wider and more flexible it becomes.
When a sound arrives, a traveling wave runs along the membrane from base toward apex, growing in amplitude and slowing as it goes, until it reaches a point of maximum displacement. Where it peaks depends on frequency: high frequencies peak at the base, low frequencies at the apex.
So frequency is turned into place. A membrane whose width and flexibility vary systematically vibrates maximally at different positions for different stimulus frequencies, which turns the whole strip into a map running from high pitch to low — this is tonotopy.
Note when this happens: the splitting is mechanical, and it happens before hair cells convert vibration into neural signals. The cochlea is not only an amplifier and a transducer — it is also a mechanical frequency analyzer that decomposes a complex waveform into simpler elements before anything is handed to the nerve.
Do not over-credit the passive gradient, though. The textbook's own wording is that frequency tuning is attributable in part to the geometry of the basilar membrane — it lays down the broad map; the rest is on the next page.
Remember this line. Damage and trouble following speech can both be located back onto it: which stretch of the line is in trouble. Tinnitus often is not on this line at all — that one gets its own scene later.
Mechanism · the passive gradient only sketches it
If the cochlea had only this passive membrane, its frequency resolution would not be sharp enough. The same textbook section limits itself explicitly: peripheral tuning is too sharp to be explained by passive mechanics alone, and the sensitivity comes from an active biomechanical process contributed by outer hair cells.Those cells contain a protein called prestin, which changes their length as membrane potential changes, actively amplifying vibration inside the cochlea. In mammals, this active amplification raises hearing sensitivity by more than 40 dB.
How to read that number: 40 dB corresponds to about a 100-fold difference in sound pressure — the faintest audible pressure drops to roughly one hundredth. It does not mean sound seems 100 times louder; perceived loudness and sound pressure are not the same quantity. And the evidence comes from mouse gene knockouts and isolated-cell work — animal mechanism research, not direct measurement in humans.
So the complete statement is: the passive stiffness-and-width gradient sketches the map; the active outer hair cells draw the fine lines.
Misconception · a complex sound does not smear
A piece of music, or a single spoken sentence, carries many frequency components at once. Do they smear into mush inside the cochlea?No. The vibration pattern that appears on the basilar membrane is equivalent to the superposition of the vibrations each component tone would produce on its own — the complex sound is spread out onto the positions of its components rather than blended. That is what split first, hand to the nerve second actually means.
But the uncertainty belongs in the sentence: that superposition is an idealization of a passive, linear cochlea, not a strict identity. The same section goes on to note that peripheral tuning is too sharp for passive mechanics alone, that at low sound levels the membrane moves far more than a linear extrapolation predicts, and that the sensitivity comes from an active process. When sound is faint, the cochlea is actively amplifying, and superposition is only an approximation.
Chapter 2
One row listens, three rows amplify
One row listens, three rows amplify
Sound's last stop is a small coiled tube buried deep in the skull: the cochlea. Uncoiled, it would form a tube roughly 33 to 35 mm long, varying noticeably between people. A flexible partition carrying the basilar membrane splits it down the middle, and sound makes that membrane heave. Hair cells sit on top of it, and when the membrane moves, the tiny processes on their apical ends — the stereocilia — bend, which changes the voltage across the hair cell membrane. Only at this step does sound first become electricity.
Hair cells are not one job but two. Inner hair cells are the actual microphones. Outer hair cells, though nominally sensory cells too, do something else: they are the amplifier. The textbook's reasoning for that call is restrained — tuning in the inner ear is too sharp to be explained by passive mechanics alone: at very low sound intensities the basilar membrane vibrates far more than linear extrapolation from loud sounds would predict, so the ear's sensitivity must also arise from an active biomechanical process, and outer hair cells are the most likely candidates for driving it.
Later work pointed at a specific molecule. The outer hair cell's membrane carries a protein called prestin; when membrane voltage changes, prestin changes shape and the whole cell lengthens and shortens with it. Knock the prestin gene out in mice and this electromotility disappears while cochlear sensitivity drops by more than 40 dB; heterozygotes left with roughly half the normal electromotility shift by about 6 dB. Dose tracks response. One honest addition: those knockout mice also had outer hair cells only about 60% of normal length with reduced axial stiffness, so at the time that 40 dB could not be booked entirely to the loss of electromotility.
An amplifier needs power, and the ear generates its own. A densely vascularized tissue called the stria vascularis holds the endolymph inside the duct at an unusual composition — about 150 mM K+, only 2 mM Na+, 20 μM Ca2+ — and potassium diffusing down that gradient across an electrical barrier leaves the endolymph at about +80 mV relative to blood plasma: the endocochlear potential. Those numbers were measured in guinea pigs, not taken from a human ear. The order runs like this: the pumps spend energy piling up a potassium gradient first, the voltage is a consequence of that gradient, and the voltage then feeds back as driving force for hair cell transduction — both for K+ entering the cell and for Ca2+ permeating it.
So how do we know this is a machine that is switched on and drawing power, rather than a passive drum skin? The most direct line is pulling the plug: anoxia, or ouabain (blocking the sodium-potassium pump), or bumetanide (blocking the sodium-potassium-chloride cotransporter) — any one of them blocks potassium transport in the stria, and the endocochlear potential falls. That is causation established by intervention, not two things merely seen together.
There is a second clue, but read carefully what it proves: play two tones at once and people hear a third that was not in the stimulus at all. A system that is passive and linear can only hand back the frequencies you put in, so difference tones show at minimum that the cochlea is nonlinear. On their own they do not yet prove energy is being added — what carries the word 'active' is the sharpness of tuning and the outsized movement at faint levels. The textbook's closing line on distortion is lovely: when we hear those extra tones, we are paying in distortion for an exquisitely fast and sensitive transduction mechanism.
Hair cells are not one job but two. Inner hair cells are the actual microphones. Outer hair cells, though nominally sensory cells too, do something else: they are the amplifier. The textbook's reasoning for that call is restrained — tuning in the inner ear is too sharp to be explained by passive mechanics alone: at very low sound intensities the basilar membrane vibrates far more than linear extrapolation from loud sounds would predict, so the ear's sensitivity must also arise from an active biomechanical process, and outer hair cells are the most likely candidates for driving it.
Later work pointed at a specific molecule. The outer hair cell's membrane carries a protein called prestin; when membrane voltage changes, prestin changes shape and the whole cell lengthens and shortens with it. Knock the prestin gene out in mice and this electromotility disappears while cochlear sensitivity drops by more than 40 dB; heterozygotes left with roughly half the normal electromotility shift by about 6 dB. Dose tracks response. One honest addition: those knockout mice also had outer hair cells only about 60% of normal length with reduced axial stiffness, so at the time that 40 dB could not be booked entirely to the loss of electromotility.
An amplifier needs power, and the ear generates its own. A densely vascularized tissue called the stria vascularis holds the endolymph inside the duct at an unusual composition — about 150 mM K+, only 2 mM Na+, 20 μM Ca2+ — and potassium diffusing down that gradient across an electrical barrier leaves the endolymph at about +80 mV relative to blood plasma: the endocochlear potential. Those numbers were measured in guinea pigs, not taken from a human ear. The order runs like this: the pumps spend energy piling up a potassium gradient first, the voltage is a consequence of that gradient, and the voltage then feeds back as driving force for hair cell transduction — both for K+ entering the cell and for Ca2+ permeating it.
So how do we know this is a machine that is switched on and drawing power, rather than a passive drum skin? The most direct line is pulling the plug: anoxia, or ouabain (blocking the sodium-potassium pump), or bumetanide (blocking the sodium-potassium-chloride cotransporter) — any one of them blocks potassium transport in the stria, and the endocochlear potential falls. That is causation established by intervention, not two things merely seen together.
There is a second clue, but read carefully what it proves: play two tones at once and people hear a third that was not in the stimulus at all. A system that is passive and linear can only hand back the frequencies you put in, so difference tones show at minimum that the cochlea is nonlinear. On their own they do not yet prove energy is being added — what carries the word 'active' is the sharpness of tuning and the outsized movement at faint levels. The textbook's closing line on distortion is lovely: when we hear those extra tones, we are paying in distortion for an exquisitely fast and sensitive transduction mechanism.
机制 · 这一幕的完整推导: 从解剖到电压
声音走到最后一站, 是耳朵最深处、埋在颅骨里的一小段卷起来的管子 —— 耳蜗。把它拉直, 是一根约 33 到 35 毫米长的管子, 人和人之间差得不少。管子中间被一张基底膜分开, 声音让这张膜起伏。后来的研究把方向指到了一个具体的分子上。外毛细胞的膜上嵌着一种叫 prestin 的蛋白, 膜电位一变, 它就跟着改变形状, 整个细胞随电压伸长、缩短。把小鼠的 prestin 基因敲掉, 外毛细胞的这种电能动性 (electromotility) 消失, 同时耳蜗灵敏度掉了 40 分贝以上; 只让 prestin 剩一半的杂合子小鼠, 阈值也相应地移了约 6 分贝。剂量和反应对得上。要补一句诚实的话: 那批敲除小鼠的外毛细胞本身也只有正常长度的约 60%、轴向刚度也降了, 所以当年还不能把这 40 分贝全部记在电能动性丢失这一件事上。
放大器要用电, 而这份电是耳朵自己发的。耳蜗里有一层布满血管的组织叫血管纹 (stria vascularis), 它把管腔里的内淋巴维持在一个很不寻常的成分上: 钾约 150 mM、钠只有 2 mM、钙 20 μM; 钾顺着浓度差跨过一道电学屏障扩散, 内淋巴因此相对于血浆带上约 +80 mV 的电位, 这就是内淋巴电位 (endocochlear potential)。这几个数是在豚鼠身上测到的, 不是从人耳里取到的。顺序是这样的: 离子泵先烧掉能量把钾堆出浓度差, 电压是这个浓度差的结果; 电压再回过头去给毛细胞的换能加驱动力 —— 钾流进细胞、钙透过来, 都靠它。
怎么知道这是一台开着机、在耗电的机器, 而不是一面被动的鼓皮? 最直接的一条是拔电源: 让耳蜗缺氧, 或者用哇巴因 (ouabain, 挡住钠钾泵)、布美他尼 (bumetanide, 挡住钠钾氯共转运体) 阻断血管纹的钾转运, 内淋巴电位就掉下来。这是干预出来的因果, 不是看见两件事同时发生。
还有一条线索, 但要读准它到底证明了什么: 同时放两个音, 人会听见刺激里原本不存在的第三个音 (差音)。一个被动而且线性的系统只能把送进来的频率还回去, 所以差音至少说明耳蜗不是线性的。它本身还不足以证明有能量被注入 —— 真正撑起主动这两个字的, 是上面那条调谐太尖锐、以及轻声时振动远超线性外推。教科书对差音的收尾很漂亮: 我们听见它, 是在为一套极快、极灵敏的换能机制付失真的代价。
Mechanism · a place first, a gain second
Before it amplifies anything, the cochlea does something more basic: it spreads frequency out into place.The basilar membrane is not uniform. It is wider and more flexible at the apical end, and narrower and stiffer at the basal end. When sound arrives, movement begins at the stiff end (the base) and propagates toward the flexible end (the apex).
Working with tubular models and human cochleas taken from cadavers, von Békésy found that an acoustic stimulus initiates a traveling wave that runs from base toward apex, growing in amplitude and slowing in velocity until it reaches a point of maximum displacement — and where that point falls is set by frequency. High frequencies peak at the base, low frequencies at the apex, turning frequency into a map (tonotopy).
Three things must not be overstated.
First, the textbook's own wording is that frequency tuning in the inner ear is attributable in part to the geometry of the basilar membrane. The passive width and stiffness gradient only sets up the coarse place-frequency correspondence; the genuinely sharp part comes from active amplification.
Second, the textbook's elegant line — that a complex sound produces a vibration pattern equivalent to the superposition of the vibrations of its component tones — describes a passive, linear membrane. The real cochlea is nonlinear and manufactures tones that were never in the stimulus, so superposition is an approximation, not an identity.
Third, the 33 to 35 mm measures the whole cochlear tube uncoiled, not the length of the basilar membrane itself.
Units · what 40 dB actually buys, and in whom
40 dB does not mean 100 times louder. A decibel is a ratio, and here the ratio is of sound pressure: 40 dB = 10^(40/20) = 100-fold in pressure. The same 40 dB expressed as intensity (energy) is 10,000-fold — and subjective loudness is not linear in pressure either, so neither 'a hundred times louder' nor 'ten thousand times louder' is correct. What the amplifier actually buys, stated precisely: the faintest audible sound can have about one hundredth the sound pressure.Two different 40 dBs live in that paper; do not splice them. The abstract's 'more than 40 dB (that is, 100-fold)' describes how much gain mammalian hearing gets from mechanical amplification in total; the 40 dB used above describes how much sensitivity the knockout mice lost (the paper's range is 40 to 60 dB). The two happen to be compatible, but quoting 'more than 40 dB' off the knockout data is taking the bottom of a range, not being more conservative than the source.
This was measured in mice. The evidence is gene-knockout mice plus isolated electromotility measurements — mechanism and animal level. Humans have not been measured the same way. So the sentence belongs to the mammalian cochlea, not to 'a hundredfold measured in your ear'.
What was measured was threshold, not basilar membrane displacement. The study measured cochlear thresholds, and its own abstract uses hedged phrasing along the lines of 'thought to be generated by'. So when 'amplifier' is cashed out as a number, that number is a shift in hearing threshold — not someone directly measuring the membrane being amplified a hundredfold.
Metabolism · the amplifier has a power bill
Endolymph has a composition rarely seen in the body: about 150 mM K+, only 2 mM Na+, 20 μM Ca2+, and it sits about +80 mV relative to blood plasma or perilymph. Holding a pool that anomalous requires the transporters of the stria vascularis to work continuously.The evidence for this is interventional, not merely observational: anoxia, ouabain (inhibiting the sodium-potassium pump), and bumetanide (inhibiting the sodium-potassium-chloride cotransporter) each block potassium transport in the stria; K+ rises in the intrastrial space, falls inside the marginal cells, and both the intrastrial potential and the endocochlear potential drop with it. Cut the energy supply and the voltage goes.
This is where picturing the cochlea as a passive drum skin misleads most. A drum skin does not fail for lack of oxygen; this apparatus does. It behaves more like an instrument that must be kept powered to stay at its operating point — sensitivity is not an intrinsic property it has, but a state it maintains, moment to moment.
Two boundaries. First, this +80 mV comes from in vivo electrophysiology in guinea pigs, so the accurate phrasing is 'about +80 mV measured in the mammalian cochlea', not 'measured in the human ear'. Second, this page explains why the apparatus is structurally dependent on metabolism — it is not guidance on managing any hearing problem. For anything specific to you, that judgment belongs to a clinician who knows your situation.
Chapter 3
Hair cells don't grow back · a balance, not a flow
Hair cells don't grow back · a balance, not a flow
The cochlea is a bony tube in your inner ear, coiled like a snail shell; uncoiled, it would be about 33 to 35 mm long, and it varies noticeably between people. Running along it is a thin sheet called the basilar membrane, and lined up on that sheet are rows of sensory cells — hair cells. Sound sets the membrane rippling, the hair bundles on top bend, and that bending is converted into electrical signals for the brain.
This row of cells has one decisive property: lose one, and you are one short. In the adult mammalian cochlea, spontaneous regeneration of hair cells has not been observed to date. So the hair-cell account is not like muscle, which comes back when you train it, nor like gut lining, which swaps itself for a fresh batch every so often. It behaves more like an account that only ever pays out: the hair cells you get for a lifetime are the deposit you were born with minus every withdrawal since, not income that arrives fresh each month.
To be precise: this is not the only account in the cochlea. Hearing can also decline through the auditory nerve, or through the tissue that powers the whole thing — one of those shows up later in this scene. But the hair-cell ledger really does only run one way.
And what goes missing is not a lump sum but a specific band of frequencies. The basilar membrane is narrow and stiff at its base and wider and more flexible at its apex; that gradient makes high frequencies peak at the base and low frequencies at the apex, so frequency is laid out as a map of position. Which stretch of that map goes first has a typical shape: early or moderately advanced noise-induced hearing loss usually shows a dip on the audiogram between 3 and 6 kHz, most often near 4 kHz, with some recovery again by 8 kHz. So the world does not get uniformly quieter — particular drawers get pulled out. Worth flagging: the exact position of that dip shifts with the frequency of the damaging noise and with individual anatomy, and no dip does not mean nothing is wrong — it is a chart a clinician reads alongside your history, not a self-test marker.
One group of these cells is itself an amplifier. Outer hair cells carry a membrane protein called prestin; when membrane voltage changes, prestin makes the cell shorten and lengthen, actively driving the vibration of the organ of Corti larger. In mammals this active amplification raises hearing sensitivity by more than 40 dB — meaning the faintest detectable sound pressure falls to roughly one hundredth of what it would otherwise be. The membrane's passive gradient only sets up the coarse position mapping: peripheral tuning is far too sharp to be explained by passive mechanics alone, and it is the active process that sharpens it. So when a patch of outer hair cells goes, what is lost is not a few signals — it is the entire amplifier for that band.
Why can it not be replaced? Hair cells do have neighbours — supporting cells — which in principle retain the potential to become hair cells. In mice, for a short period after birth, that potential is genuinely usable: supporting cells can either change identity directly into hair cells, or re-enter the cell cycle and divide to produce new ones. But the ability is progressively lost after birth.
Two things have to be stated precisely. First, that window was measured in mice; there is no equivalent human data, so it must not be read as human infants having a regenerative period that then closes. Second, spontaneous hair-cell regeneration has not been observed in the adult mammalian cochlea — which is a failure to find, not a proof of impossibility. That distinction comes back when regenerative-medicine marketing gets discussed.
So everything the rest of this island covers — volume, duration, earplugs, when to step outside for a while — comes down to the same act: spending a little less from an account that does not top itself up.
This row of cells has one decisive property: lose one, and you are one short. In the adult mammalian cochlea, spontaneous regeneration of hair cells has not been observed to date. So the hair-cell account is not like muscle, which comes back when you train it, nor like gut lining, which swaps itself for a fresh batch every so often. It behaves more like an account that only ever pays out: the hair cells you get for a lifetime are the deposit you were born with minus every withdrawal since, not income that arrives fresh each month.
To be precise: this is not the only account in the cochlea. Hearing can also decline through the auditory nerve, or through the tissue that powers the whole thing — one of those shows up later in this scene. But the hair-cell ledger really does only run one way.
And what goes missing is not a lump sum but a specific band of frequencies. The basilar membrane is narrow and stiff at its base and wider and more flexible at its apex; that gradient makes high frequencies peak at the base and low frequencies at the apex, so frequency is laid out as a map of position. Which stretch of that map goes first has a typical shape: early or moderately advanced noise-induced hearing loss usually shows a dip on the audiogram between 3 and 6 kHz, most often near 4 kHz, with some recovery again by 8 kHz. So the world does not get uniformly quieter — particular drawers get pulled out. Worth flagging: the exact position of that dip shifts with the frequency of the damaging noise and with individual anatomy, and no dip does not mean nothing is wrong — it is a chart a clinician reads alongside your history, not a self-test marker.
One group of these cells is itself an amplifier. Outer hair cells carry a membrane protein called prestin; when membrane voltage changes, prestin makes the cell shorten and lengthen, actively driving the vibration of the organ of Corti larger. In mammals this active amplification raises hearing sensitivity by more than 40 dB — meaning the faintest detectable sound pressure falls to roughly one hundredth of what it would otherwise be. The membrane's passive gradient only sets up the coarse position mapping: peripheral tuning is far too sharp to be explained by passive mechanics alone, and it is the active process that sharpens it. So when a patch of outer hair cells goes, what is lost is not a few signals — it is the entire amplifier for that band.
Why can it not be replaced? Hair cells do have neighbours — supporting cells — which in principle retain the potential to become hair cells. In mice, for a short period after birth, that potential is genuinely usable: supporting cells can either change identity directly into hair cells, or re-enter the cell cycle and divide to produce new ones. But the ability is progressively lost after birth.
Two things have to be stated precisely. First, that window was measured in mice; there is no equivalent human data, so it must not be read as human infants having a regenerative period that then closes. Second, spontaneous hair-cell regeneration has not been observed in the adult mammalian cochlea — which is a failure to find, not a proof of impossibility. That distinction comes back when regenerative-medicine marketing gets discussed.
So everything the rest of this island covers — volume, duration, earplugs, when to step outside for a while — comes down to the same act: spending a little less from an account that does not top itself up.
机制 · 这本账开在哪里: 解剖、频率地图, 与那台放大器
耳蜗 (cochlea) 是内耳里一条盘成蜗牛壳的骨管, 把它拉直大约 33 到 35 毫米长, 而且人和人差得不小。管子里顺着一张叫基底膜 (basilar membrane) 的薄膜, 铺着一排排感觉细胞 —— 毛细胞 (hair cell)。声音把膜推得起伏, 毛细胞顶上的纤毛跟着弯一下, 这个弯曲就被换成电信号发给大脑。账上少掉的也不是一个笼统的总数, 而是具体某一段频率。基底膜的底端窄而硬, 顶端宽而软; 这个渐变让高频的振动峰落在底端、低频落在顶端, 于是频率被摊成了一张位置地图。这张地图上哪一段先出问题, 是有典型形状的: 早期或中度进展的噪声性听力损失, 听力图上常见的是 3 到 6 kHz 的一个凹陷, 最常见落在 4 kHz 附近, 到 8 kHz 又回升一些。所以不是整个世界统一调小了音量, 而是某几个抽屉被抽走了。要提醒的是, 这个凹陷的位置会随致害噪声的频率与个人条件浮动, 没有凹陷也不等于没事 —— 它是医生连同病史一起读的图, 不是自测标记。
而且这排细胞里有一组本身就是放大器。外毛细胞 (outer hair cell) 的膜上有一种叫 prestin 的蛋白, 膜电位一变它就让细胞伸缩, 主动把耳蜗内柯蒂器的振动推大。在哺乳动物身上, 这套主动放大把听觉灵敏度提高了 40 分贝以上, 也就是能被听见的最小声压降到大约百分之一。基底膜的被动渐变只把粗略的位置对应搭好: 外周的频率调谐尖锐到光靠被动力学解释不了, 真正把它调尖的是这套主动过程。所以外毛细胞掉一片, 丢的不只是几个信号, 是那一段频率上的整个放大器。
那为什么补不上? 毛细胞旁边其实有一群叫支持细胞 (supporting cell) 的邻居, 它们在原理上保有变成毛细胞的潜力。在小鼠出生后最初的一小段时间里, 这个潜力是真能用的: 支持细胞既可以直接改变身份变成毛细胞, 也可以重新进入细胞周期、分裂出新的毛细胞。但这项能力在出生后逐步丧失。
有两处必须把话说准。第一, 上面这个窗口是在小鼠身上测出来的, 人类没有对应数据, 不能读成人类婴儿有一段可再生期然后关闭。第二, 成年哺乳动物耳蜗迄今没有观察到毛细胞自发再生 —— 这是没找到, 不等于已经证明不可能。这个区别在后面讲再生医学宣传时还会用到。
Mechanism · the door isn't broken, it's locked
Supporting cells have not lost the genes needed to become hair cells. The genes are still there — they just cannot be read anymore.A mouse study gave that lock a concrete shape: as development proceeds, supporting cells progressively accumulate DNA methylation at the promoters of developmentally regulated hair-cell genes. DNA methylation is a chemical tag stuck onto a gene's switch; once it is there, that stretch of DNA becomes harder to read. More pointedly, the methylation overlaps with binding sites for Atoh1 — Atoh1 being the key transcription factor that decides whether a cell becomes a hair cell. The commander is still present; the buttons it needs to press have been sealed over.
The timeline lines up too: methylation at these sites climbs steadily through development, across exactly the window in which supporting cells lose their capacity to change identity (transdifferentiate).
Three limits belong with this mechanism, not after it:
It is all mouse, and the crucial identity-switching experiments were done on explanted cochlear tissue in culture — not in a living animal, and certainly not in a human.The authors' own verb is contributes to, not causes. Methylation is one layer of a multi-layer epigenetic barrier — earlier work found enhancer decommissioning as another layer, and each only partly explains the failure to regenerate.So this page cannot be read as: erase the methylation and the ear grows hair cells again. In the paper, the step showing that demethylation is required for neonatal identity-switching rests on a single chemical inhibitor, with no genetic corroboration.
The point of this page is neither hope nor despair. It is an explanation of why this is hard: what has to be prised open is not a broken part, but a program that development locked shut one layer at a time.
nguyen-2023-cochlea-methylation
Clinical · the audiogram resets; the account may not
The night the concert ends, your ears feel stuffed, they ring, voices sound like they come through cotton wool. By morning it is gone. So you conclude the night was free.In mice, that assumption has been tested directly, and the answer is no.
Researchers exposed mice to 100 dB SPL noise in an 8-16 kHz band for two hours. Thresholds (the faintest audible sound) did rise — and recovered completely to pre-exposure levels within two weeks. Counting hair cells afterwards: not one was missing, inner or outer, and that was still true a year later. Up to here, everything matches the impression that a night's sleep fixes it.
Yet in those same ears, in the region tuned to 32 kHz, the connections between inner hair cells and the auditory nerve — the ribbon synapses — fell from about 16 per inner hair cell to under 7 within 24 hours, and had not recovered by the 8-week follow-up. A slower step followed: the spiral ganglion neurons that lost those connections — the cell bodies of the auditory nerve, the ones that actually carry signal to the brain — were still normal in number at two weeks, began falling significantly around a year, and were down by roughly half in the high-frequency region at two years.
Which is to say: on the morning the audiogram read zero, a withdrawal had already been made — one the measurement of the day simply could not see. It cuts the wiring first, and kills the cells slowly afterward.
Now the restraint, which matters more than the conclusion.
All of the above is mouse data at a laboratory dose (100 dB, two hours). It is not the same thing as listening to music on headphones.What about people? Two questions have to be kept apart. That this synaptic loss exists in human ears is not much disputed — post-mortem temporal-bone histology shows it directly, particularly the age-related kind. What remains unsettled is the other two: whether noise is its cause in humans, and whether it can be measured in a living person at all.In the living, synapses can only be inferred indirectly from electrophysiological and audiological measures, and those methods return conflicting results; a 2019 expert review summed up the state of play as: the available metrics are imperfect, and a satisfactory diagnostic approach is still far off.
A related clue that cannot stand in for the one above: one study recruited 116 adults with entirely normal conventional audiograms (every threshold from 0.25 to 8 kHz at 20 dB HL or better, where HL is a scale zeroed on normal young hearing), mean age 29.5, range 18 to 65. Of them, 74 (64%) already had elevated thresholds in the extended high frequencies above 8 kHz, and 39 (34%) reported difficulty following speech in noisy places. The extended-high-frequency average separated the people who reported difficulty better than the conventional average did: pick one complainer and one non-complainer at random, and the extended high frequencies guess which is which about 81% of the time, versus about 71% for the conventional range.
Read it carefully: 64% is the proportion in that recruited sample, not a population prevalence; and discriminating better means concurrent statistical separation, not prospective prediction, and certainly not that fixing high-frequency hearing would make speech in noise easier. It also measured no synapses at all, so it cannot be used to prove that the hidden damage above happens in humans.
What it does establish is one thing, and one thing is enough: a normal conventional audiogram does not mean nothing has happened in the cochlea.
Practical · the dose has already been written down
If the account cannot be topped up, only one question is left: how much goes out at a time.First, connect that back to the body. What the cochlea receives is not loudness but energy: every 3 dB rise in sound pressure level doubles the energy the sound carries, so the same energy dose only lasts half as long. This modelling assumption — the equal-energy principle — is the algorithm under every number below. It welds intensity and duration into a single quantity, which is exactly the account metaphor: what you spend was never volume, it was volume multiplied by time. Honesty requires saying that this is a calculation convention standards adopt, not a physiological law measured directly in people.
The occupational figure: the US NIOSH recommended exposure limit is 85 dBA as an 8-hour time-weighted average (dBA meaning decibels weighted for the sensitivity of human hearing). It comes with a 3-dB exchange rate: every 3 dB louder halves the recommended duration. So on the same table, 85 dBA maps to 8 hours, 88 dBA to 4 hours, and 91 dBA to 2 hours. That conversion is the thing to take away: a little louder costs half the time, not a little less time.
Two limits travel with it:
It is a recommended exposure limit, not a legal one. The legally enforced US figure (OSHA) uses a different set of numbers — 90 dBA with a 5 dB exchange rate — so seeing two numbers is not a contradiction, it is two conventions.It is an occupational figure, built on an 8-hour workday across a 40-year working life. It answers how long you can be somewhere at work; it does not answer what level is absolutely safe, and carrying it straight over to concerts, headphones, and gyms goes beyond what the document covers.
The recreational figure: the WHO-ITU safe listening standard sets two reference exposures for personal audio devices — the adult mode is equivalent to 80 dB for 40 hours per week, and the mode for sensitive users such as children is equivalent to 75 dB for 40 hours per week. The standard also recommends that devices track the user's sound exposure, and requires that users be able to see how much of their allowance they have used. It manages exposure as a weekly allowance that runs out — the engineering version of the deposit metaphor.
Just as plainly: this standard is voluntary, not law; a manufacturer that ignores it breaks nothing. And what it gives is an allowance, which is not the same as a guarantee that staying under it is safe.
Why it is worth the trouble: WHO notes that the most common causes of hearing loss in adults — including exposure to loud sound and ototoxic medicines — are preventable. The same source gives a scale: over 1 billion young adults are at risk of permanent, avoidable hearing loss from unsafe listening (at risk — not a count of people already affected).
And earplugs? They help, though perhaps not in the way you assume. In a Cochrane review, two randomized trials (140 people between them) found that people trained in proper insertion got 8.59 dB better attenuation from their earplugs than untrained people (95% CI 6.92 to 10.25), on moderate-quality evidence.
Read that precisely: it measured how many decibels the plug blocked, a surrogate — not hearing loss itself — and only at short-term follow-up. Two other conclusions in the same review — that stricter legislation lowers workplace noise, and that better hearing-protector use within prevention programmes lowers risk — rest on very low-quality evidence.
So the claim that survives is narrow, and useful enough: inserting an earplug properly has measured value on its own.
tikka-2017-noise-hearing-loss
Myths · so can any of it be put back?
Once you know the account cannot be topped up, the next thought is almost inevitable: is there a drug, are there stem cells, is there anything that grows it back?Three common claims, taken separately.
Claim one: an antioxidant can prevent noise-induced hearing loss. Half true. Take N-acetylcysteine (NAC): it has consistently reduced permanent noise-induced hearing loss in laboratory studies, while its clinical efficacy remains controversial. The human trial that tested it is worth reading closely — 566 military personnel, double-blind and placebo-controlled, and the primary endpoint was not met; the investigators' own words were that the present study design failed to confirm it. Signals did appear in secondary and post-hoc analyses — and treating a secondary endpoint as proof of efficacy is exactly the move this kind of marketing makes.
Claim two: stem cells or regenerative medicine will fix this any day now. The direction is real; the timescale has been compressed. The real part: supporting cells do retain the potential to become hair cells, a potential that is progressively switched off after birth. The compressed part: spontaneous hair-cell regeneration has not been observed in the adult mammalian cochlea, and prising that lock open means facing an epigenetic barrier laid down layer by layer through development — on evidence from mice, with the key experiments done on explanted tissue.
Claim three: block DNA methylation and the ear regrows hair cells. This one does not hold up yet. In the relevant work, the step showing that demethylation is required for neonatal identity-switching rests on a single chemical inhibitor with no genetic corroboration; and the authors' own framing is that methylation contributes to the failure to regenerate, as one of several layers, not as the single switch.
How to use all three? There is really only one test: is the claim about spending less, or about restoring the balance?
On the spending-less side there is something solid: a dose-and-duration table already written down that anyone can look up, and measured evidence that inserting an earplug properly blocks several more decibels. On the restoring-the-balance side, no method has yet been shown to bring back hair cells that have already died in an adult cochlea — note that this too is not yet, not impossible, and the review carrying that sentence was published in 2017.
If a pitch promises the second one, three questions are worth asking first: in whom was this done? In a living animal, or in a dish? And was the primary endpoint met?
kopke-2015-nac-nihl-rctnguyen-2023-cochlea-methylationtikka-2017-noise-hearing-loss
Chapter 4
High notes go first · you hear it, you can't parse it
High notes go first · you hear it, you can't parse it
The cochlea sits deep in your inner ear — a tube coiled into a snail shell; uncoiled, it would be about 33 to 35 mm long. Stretched along its whole length is a membrane called the basilar membrane. When sound arrives, the membrane doesn't vibrate all at once — it vibrates at particular places. That division of labour by place is the precondition for talking about losing the high notes first.
The membrane is not uniform: the end nearest the entrance (the base) is narrow and stiff, and it grows wider and more flexible the deeper you go toward the apex. Sound sets off a traveling wave that runs from base toward apex like a flick down a rope, growing until it peaks at one location and then dying away fast. The narrow, stiff end responds most to high frequencies, the wide, floppy end to low frequencies — so frequency gets translated into position.
To be precise: this passive stiffness gradient only lays down the rough correspondence — the textbook's own wording is that tuning is attributable to the membrane's geometry only in part. The genuinely sharp tuning takes the outer hair cells: they actively change length as their membrane voltage swings, amplifying the vibration inside the cochlea — and in the mammalian cochlea that active amplification raises sensitivity by more than 40 dB. One conversion that often gets mangled: 40 dB means 100 times in sound-pressure amplitude, not 100 times as loud to you.
So before any signal becomes a nerve impulse, the ear has already sorted it by frequency. The textbook calls the cochlea a mechanical frequency analyzer: a complex sound produces a vibration pattern equivalent to superimposing the vibrations its individual component tones would each produce. Note the word equivalent rather than equal — at low sound levels the cochlea actively amplifies, and superposition is then only an approximation.
Once the labour is divided by place, damage acquires a favourite address. Early or moderately advanced noise-induced hearing loss typically shows up as a dip in the audiogram at 3 to 6 kHz — clinically, a notch — most often centred near 4 kHz, with some recovery again by 8 kHz. Why that band? The review attributes it not to anything inside the cochlea but to the two stages in front of it: ear-canal resonance plus middle-ear mechanics already boost those frequencies before they ever reach the cochlea. In other words, the high end goes first mainly because more energy is delivered there to begin with, not because of the place mapping itself. One further limit: this shape describes the early stage only.
That division of labour may be connected to what many people actually report: the volume is still there, a quiet room is fine, but a noisy restaurant turns speech into something you can't grab hold of. And it doesn't necessarily wait for the audiogram to go bad — among young adults with normal conventional audiograms, elevated thresholds in the extended high frequencies above 8 kHz are fairly common, and that range is simply not covered by the conventional chart. The actual numbers, and what they can and cannot show, are laid out on the pages that follow.
The membrane is not uniform: the end nearest the entrance (the base) is narrow and stiff, and it grows wider and more flexible the deeper you go toward the apex. Sound sets off a traveling wave that runs from base toward apex like a flick down a rope, growing until it peaks at one location and then dying away fast. The narrow, stiff end responds most to high frequencies, the wide, floppy end to low frequencies — so frequency gets translated into position.
To be precise: this passive stiffness gradient only lays down the rough correspondence — the textbook's own wording is that tuning is attributable to the membrane's geometry only in part. The genuinely sharp tuning takes the outer hair cells: they actively change length as their membrane voltage swings, amplifying the vibration inside the cochlea — and in the mammalian cochlea that active amplification raises sensitivity by more than 40 dB. One conversion that often gets mangled: 40 dB means 100 times in sound-pressure amplitude, not 100 times as loud to you.
So before any signal becomes a nerve impulse, the ear has already sorted it by frequency. The textbook calls the cochlea a mechanical frequency analyzer: a complex sound produces a vibration pattern equivalent to superimposing the vibrations its individual component tones would each produce. Note the word equivalent rather than equal — at low sound levels the cochlea actively amplifies, and superposition is then only an approximation.
Once the labour is divided by place, damage acquires a favourite address. Early or moderately advanced noise-induced hearing loss typically shows up as a dip in the audiogram at 3 to 6 kHz — clinically, a notch — most often centred near 4 kHz, with some recovery again by 8 kHz. Why that band? The review attributes it not to anything inside the cochlea but to the two stages in front of it: ear-canal resonance plus middle-ear mechanics already boost those frequencies before they ever reach the cochlea. In other words, the high end goes first mainly because more energy is delivered there to begin with, not because of the place mapping itself. One further limit: this shape describes the early stage only.
That division of labour may be connected to what many people actually report: the volume is still there, a quiet room is fine, but a noisy restaurant turns speech into something you can't grab hold of. And it doesn't necessarily wait for the audiogram to go bad — among young adults with normal conventional audiograms, elevated thresholds in the extended high frequencies above 8 kHz are fairly common, and that range is simply not covered by the conventional chart. The actual numbers, and what they can and cannot show, are laid out on the pages that follow.
机制 · 位置分工是怎么建立的
耳蜗在你的内耳深处, 是一根卷成蜗牛壳形状的管子; 把它展开, 大约 33 到 35 毫米长。管子里从头到尾绷着一条膜, 叫基底膜 (basilar membrane)。声音进来以后, 这条膜不是整条一起响, 而是在不同的位置响 —— 位置分工, 是能谈高频先掉的前提。这条膜不是均匀的: 靠近入口的那一端 (底部) 又窄又硬, 越往深处 (顶端) 走越宽、越软。声音一进来, 会像抖一条绳子那样从底部往顶端推出一列行波 (traveling wave), 走到某个位置振幅最大, 然后迅速衰减。窄而硬的那头对高频响应最强, 宽而软的那头对低频响应最强, 于是频率被翻译成了位置。
要说准一点: 这条被动的软硬渐变只奠定了粗略的对应关系, 教科书自己的措辞是它只解释了其中一部分。真正锐利的分辨还要靠外毛细胞 (outer hair cell): 它们随膜电位变化主动伸缩, 把耳蜗里的振动放大 —— 在哺乳动物的耳蜗里, 这套主动放大让灵敏度提高 40 dB 以上。顺带说清一个常被写错的换算: 40 dB 指的是声压幅值的 100 倍, 不是听起来响 100 倍。
所以在信号变成神经冲动之前, 耳朵已经先做了一次分频。教科书把耳蜗直接称为一台机械频率分析器: 复杂声音在膜上产生的振动, 相当于把它各个成分单音各自引起的振动叠加起来。这里说相当于而不是等于是有原因的 —— 声音很轻时耳蜗会主动介入放大, 那时叠加只是一个近似。
Numbers · what that study actually measured
The line on the previous page — that a normal conventional audiogram can still sit on top of something — comes from a study of 116 adults. The numbers deserve to be opened up, because they are easy to overuse.Every participant's conventional audiogram (0.25 to 8 kHz) was within 20 dB HL at every frequency (HL being a scale zeroed on normal young hearing). Of them, 74 (64%) had elevated thresholds in the extended high frequencies above 8 kHz, and 39 (34%) reported difficulty hearing speech in noise.
The extended-high-frequency average separated those reporting difficulty better than the conventional range did: pick one person who reports difficulty and one who doesn't, and the extended high frequencies identify which is which about 81% of the time, versus about 71% for the conventional range.
The limits have to be said plainly:
That is statistical discrimination at one point in time, not a forecast, and certainly not grounds for claiming that fixing extended high-frequency hearing would restore speech in noise.64% is the proportion inside this recruited sample, not a population prevalence.Participants averaged 29.5 years with a range of 18 to 65 — they were not all young.It is a single-centre, single cross-section sample reporting an association, not a cause.
What it establishes firmly is one thing: a normal conventional audiogram does not mean nothing has happened at the top end.
Myth · a notch is not noise's signature
The notch gets treated as noise's fingerprint, and that claim outruns the evidence.The review's own description of the notch carries two qualifiers: it is the typical shape of early or moderately advanced loss, and once aging is layered on top the notch may become less prominent. Which means: a notch does not prove noise caused it, and no notch does not prove nothing is wrong. This is a chart a clinician reads alongside your history, not one to draw your own conclusions from.
Separately, one thing is completely unlike the chronic change this scene describes and is worth remembering on its own. Clinical guidelines define sudden sensorineural hearing loss as a rapid drop occurring within a 72-hour window, of 30 dB or more across at least three consecutive frequencies. Note that the 72 hours is the window in which the loss develops — not 72 hours you have in which to act. The literature identifies this as a situation calling for prompt medical attention, with the guideline's actionable anchor being an audiogram within 14 days of symptom onset. Two caveats matter just as much: the 30 dB across three frequencies is a research and enrolment convention, and the guideline itself notes that clinicians may reasonably act on smaller drops; and that number is measured by a clinician on an audiogram, usually against the opposite ear as reference — it is not something you can measure yourself.
chandrasekhar-2019-sudden-hearing-loss
Terms · normal audiogram means normal where, exactly
A normal audiogram sounds like a verdict on your hearing, but it has a defined boundary — and the boundary falls exactly where this scene is looking.A conventional audiogram tests 0.25 to 8 kHz, and the threshold for calling it normal is that every frequency is within 20 dB HL. Everything above 8 kHz is the extended high frequency (EHF) range; the 116-adult study measured 10 to 16 kHz. Put the two together and it gets interesting: the typical noise notch sits at 3 to 6 kHz, comfortably inside the conventional range, while the range above it is simply not covered by that chart.
That is how the study can hold two statements at once: those 74 people did have normal conventional audiograms by the 20 dB HL line, and they did have elevated thresholds above 8 kHz. No contradiction — the two statements measure different stretches.
The uncertainty stays on the page too: that was a single-centre, single cross-section sample of 116, reporting an association rather than a cause, and not a way to forecast any one person's future from one chart.
Practice · the dose is countable
Damage favours the high-frequency end, and what the cochlea gets billed for is sound energy: every 3 dB rise in level doubles the energy carried. So only two things are countable: how loud, and how long. Noise standards are written on that equal-energy convention — and honesty requires saying it is a calculation convention standards adopt, not a physiological law verified directly in people.Occupational framing: NIOSH sets its recommended exposure limit at 85 dBA as an 8-hour time-weighted average, with a 3 dB exchange rate — every 3 dB rise halves the allowed time: 88 dBA leaves 4 hours, 91 dBA leaves 2. Two qualifiers travel with it: this is a recommended, not a legal, limit (the legally enforced US figure, OSHA, uses 90 dBA with a 5 dB exchange rate), and it is derived for an occupational 8-hour working day, so applying it directly to concerts or headphones is extrapolation.
Personal-device framing is a separate document: the WHO-ITU safe listening standard sets the reference exposure for adults at 80 dB for 40 hours per week, and for sensitive users such as children at 75 dB for 40 hours. The standard requires devices to let the user see how much of their sound allowance they have used, and recommends that devices track the level and duration of exposure. The key point: it is a voluntary standard, not binding regulation — whether your headphones do this is not decided by law.
How you insert an earplug is worth more discussion than whether you own one: in a Cochrane review, 2 RCTs totalling 140 participants found that attenuation was 8.59 dB better with insertion instruction than without (95% CI 6.92 to 10.25, moderate-quality evidence). Two honest limits: what was measured is attenuation, a surrogate, not hearing loss itself, and follow-up was short-term only. Two other conclusions in the same review — that stricter legislation lowers workplace noise, and that better hearing-protector use within prevention programmes reduces risk — rest on very low-quality evidence.
Can you take something to cancel it out? The antioxidant N-acetylcysteine (NAC) has consistently reduced permanent noise-induced hearing loss in the laboratory, but its clinical efficacy remains controversial. A double-blind placebo-controlled trial in 566 military personnel failed to confirm efficacy on its primary endpoint. In that trial at least, a pill did not substitute for turning the volume down.
All of this is worth the effort because the direction is right: WHO notes that the most common causes of hearing loss in adults, such as exposure to loud sounds and ototoxic medicines, are preventable.
tikka-2017-noise-hearing-losskopke-2015-nac-nihl-rct
Chapter 5
Noise dose · every +3 dB halves the time
Noise dose · every +3 dB halves the time
Noise damage happens in the cochlea of the inner ear: a tube coiled like a snail shell, with a row of hair cells running along it. As earlier scenes covered, no spontaneous hair cell regeneration has been observed in the adult mammalian cochlea. Broken parts do not get restocked, which leaves exactly one lever: do not break them.
Worse, your ears hand you a false all-clear. In mice, a single laboratory noise exposure (100 dB SPL for two hours) raised hearing thresholds — the faintest audible sound got louder — and those thresholds then recovered completely within two weeks, with no loss of inner or outer hair cells at all. Yet the ribbon synapses between inner hair cells and the auditory nerve were extensively lost within 24 hours and had still not recovered eight weeks on, and spiral ganglion neurons — the cell bodies of the auditory nerve, the ones that carry signal to the brain — began dying off noticeably about a year later. Thresholds returning to normal, in other words, is not proof that nothing was lost. The boundaries matter: that is a mouse, at a laboratory noise dose, and it does not convert directly into everyday listening. In people, this kind of synaptic loss can currently only be confirmed by post-mortem temporal bone analysis, which makes it very hard to study in living humans; the various non-invasive measures tried so far disagree with one another, and whether noise is the cause in humans remains unsettled. One clarification is owed: age-related synaptic loss in the human inner ear is not much disputed. What is unresolved is whether it can be measured in the living, and whether noise is to blame.
You cannot see it afterwards and it does not grow back, so what is left is doing the arithmetic beforehand.
Essentially every noise exposure standard assumes that what the cochlea gets billed for is not how loud, but how loud × how long — the equal-energy principle. The catch is that decibels are a logarithmic scale: add 3 to the number and the sound energy doubles; double the energy and the permitted time has to halve. Honesty requires two additions: this is a modelling convention standards adopt, not a physiological law measured directly in people; and not every standard uses 3 — the legally enforced US figure (OSHA) uses a 5 dB exchange rate, while NIOSH and the international line (ISO 1999, the EU, WHO) use 3.
The recommended exposure limit from the US National Institute for Occupational Safety and Health (NIOSH) is written on exactly that 3 dB rule: 85 dBA for 8 hours, then halve the time for every 3 dB up — 88 dBA for 4 hours, 91 dBA for 2 hours, 94 dBA for 1 hour. (The A in dBA means decibels weighted for the sensitivity of human hearing — a different ruler from the unweighted dB SPL used for the mouse experiment above, so the two '100's cannot be lined up.)
Carry the same rule further yourself and the numbers fall much faster than intuition expects. The rows below are computed from the 3 dB rule, not lifted from the original table:
97 dBA — 30 minutes100 dBA — 15 minutes103 dBA — about 7.5 minutes110 dBA — about a minute and a half
From 85 to 110, on the usual rule of thumb that roughly every 10 dB doubles perceived loudness, sounds maybe five or six times louder. On the ledger it is eight hours shrinking to ninety seconds — a factor of over three hundred. If this scale teaches one thing, it is that your intuition about decibels cannot be trusted.
Three boundaries have to be stated plainly.
First, this is a recommendation, not a legal limit. The document is formally titled *criteria for a recommended standard*. It is not an enforcement threshold, and it does not mean anything below the line is safe.
Second, this is an occupational frame. Its model is one person working 8 hours a day for decades. Concerts, headphones and gyms are not inside those assumptions, and lifting those rows straight across is using them out of scope — their job here is to show you how steep a logarithmic scale is, not to issue permits for recreational listening.
Third, recreational listening has a framework of its own. The 2019 WHO-ITU standard for safe listening devices sets two reference exposures for personal audio devices: 80 dB over 40 hours per week for adults, and 75 dB over 40 hours per week for sensitive users such as children. It turns that exposure into a spendable sound allowance: the standard recommends that the device track your level and duration, and requires that you be able to see how much of the day's and the week's allowance you have used.
Two points of proportion are worth keeping. Compliance with the standard is voluntary, not a regulation — a manufacturer that ignores it breaks no law. And what a standard specifies is a different sentence from what is safe to listen to: the first has a primary document behind it, the second would need its own body of evidence, and this page makes no such second claim.
The allowance being settled weekly is the practically useful part: one very loud exposure does not just spend today, it takes a bite out of the whole week, and the days that follow draw on the same account.
The last word goes to scale. The most common causes of hearing loss in adults — exposure to loud sound and ototoxic medicines — are preventable; and more than 1 billion young adults worldwide are at risk of permanent, avoidable hearing loss from unsafe listening practices.
Worse, your ears hand you a false all-clear. In mice, a single laboratory noise exposure (100 dB SPL for two hours) raised hearing thresholds — the faintest audible sound got louder — and those thresholds then recovered completely within two weeks, with no loss of inner or outer hair cells at all. Yet the ribbon synapses between inner hair cells and the auditory nerve were extensively lost within 24 hours and had still not recovered eight weeks on, and spiral ganglion neurons — the cell bodies of the auditory nerve, the ones that carry signal to the brain — began dying off noticeably about a year later. Thresholds returning to normal, in other words, is not proof that nothing was lost. The boundaries matter: that is a mouse, at a laboratory noise dose, and it does not convert directly into everyday listening. In people, this kind of synaptic loss can currently only be confirmed by post-mortem temporal bone analysis, which makes it very hard to study in living humans; the various non-invasive measures tried so far disagree with one another, and whether noise is the cause in humans remains unsettled. One clarification is owed: age-related synaptic loss in the human inner ear is not much disputed. What is unresolved is whether it can be measured in the living, and whether noise is to blame.
You cannot see it afterwards and it does not grow back, so what is left is doing the arithmetic beforehand.
Essentially every noise exposure standard assumes that what the cochlea gets billed for is not how loud, but how loud × how long — the equal-energy principle. The catch is that decibels are a logarithmic scale: add 3 to the number and the sound energy doubles; double the energy and the permitted time has to halve. Honesty requires two additions: this is a modelling convention standards adopt, not a physiological law measured directly in people; and not every standard uses 3 — the legally enforced US figure (OSHA) uses a 5 dB exchange rate, while NIOSH and the international line (ISO 1999, the EU, WHO) use 3.
The recommended exposure limit from the US National Institute for Occupational Safety and Health (NIOSH) is written on exactly that 3 dB rule: 85 dBA for 8 hours, then halve the time for every 3 dB up — 88 dBA for 4 hours, 91 dBA for 2 hours, 94 dBA for 1 hour. (The A in dBA means decibels weighted for the sensitivity of human hearing — a different ruler from the unweighted dB SPL used for the mouse experiment above, so the two '100's cannot be lined up.)
Carry the same rule further yourself and the numbers fall much faster than intuition expects. The rows below are computed from the 3 dB rule, not lifted from the original table:
97 dBA — 30 minutes100 dBA — 15 minutes103 dBA — about 7.5 minutes110 dBA — about a minute and a half
From 85 to 110, on the usual rule of thumb that roughly every 10 dB doubles perceived loudness, sounds maybe five or six times louder. On the ledger it is eight hours shrinking to ninety seconds — a factor of over three hundred. If this scale teaches one thing, it is that your intuition about decibels cannot be trusted.
Three boundaries have to be stated plainly.
First, this is a recommendation, not a legal limit. The document is formally titled *criteria for a recommended standard*. It is not an enforcement threshold, and it does not mean anything below the line is safe.
Second, this is an occupational frame. Its model is one person working 8 hours a day for decades. Concerts, headphones and gyms are not inside those assumptions, and lifting those rows straight across is using them out of scope — their job here is to show you how steep a logarithmic scale is, not to issue permits for recreational listening.
Third, recreational listening has a framework of its own. The 2019 WHO-ITU standard for safe listening devices sets two reference exposures for personal audio devices: 80 dB over 40 hours per week for adults, and 75 dB over 40 hours per week for sensitive users such as children. It turns that exposure into a spendable sound allowance: the standard recommends that the device track your level and duration, and requires that you be able to see how much of the day's and the week's allowance you have used.
Two points of proportion are worth keeping. Compliance with the standard is voluntary, not a regulation — a manufacturer that ignores it breaks no law. And what a standard specifies is a different sentence from what is safe to listen to: the first has a primary document behind it, the second would need its own body of evidence, and this page makes no such second claim.
The allowance being settled weekly is the practically useful part: one very loud exposure does not just spend today, it takes a bite out of the whole week, and the days that follow draw on the same account.
The last word goes to scale. The most common causes of hearing loss in adults — exposure to loud sound and ototoxic medicines — are preventable; and more than 1 billion young adults worldwide are at risk of permanent, avoidable hearing loss from unsafe listening practices.
机制 · 听力恢复了, 不等于没有损失
声音的伤害发生在内耳的耳蜗里: 那是一条卷成蜗牛壳的管子, 里面排着一列毛细胞。前几幕说过, 在成年哺乳动物的耳蜗里, 迄今没有观察到毛细胞自发再生。零件坏了不补货, 所以这件事只剩一根杠杆: 别把它用坏。更麻烦的是, 耳朵还会给你一个假的安全信号。在小鼠身上, 一次实验室噪声暴露 (100 dB SPL, 两小时) 造成的听阈升高 —— 也就是能听见的最小声音变大了 —— 在两周内完全恢复, 内外毛细胞一个都没少; 但内毛细胞与听神经之间的带状突触在 24 小时内就大量丢失, 到暴露后 8 周仍未恢复, 而螺旋神经节细胞 (听神经的胞体, 把信号送去大脑的那批细胞本体) 在约一年后开始明显死亡。听力恢复到原样, 未必等于没有损失。边界要说清: 这是小鼠, 是实验室级的噪声剂量, 不能直接换算成日常听音乐。换到人身上, 这类突触丢失目前只能靠尸检颞骨确认, 因此在活人身上很难研究; 已经试过的多种无创指标结论互相矛盾, 噪声是不是人身上的成因也仍未定论。要补一句: 人的内耳存在年龄相关的突触丢失, 这一点争议不大; 悬而未决的是活体能不能测出来, 以及噪声该不该为它负责。
分寸 · 三套口径各自管到哪里, 以及问题有多大
先补一条标准之间的差别: 并不是所有标准都用 3 —— 美国法定的 OSHA 口径用的是 5 分贝交换率, 采用 3 分贝的是 NIOSH 与国际上的 ISO 1999 / 欧盟 / WHO 这一系。再说回休闲聆听那份 WHO-ITU 标准。它把这份暴露量做成一笔可以花的声音额度 (sound allowance): 标准建议设备跟踪你的音量与时长, 并规定你必须能看到这一天、这一周已经用掉了多少。
两处分寸值得留意。这份标准是自愿采纳的, 不是强制法规, 厂商不做也不违法。另外, 标准怎么定, 与这样听是否安全, 是两句不同的话: 前者有一手文件为凭, 后者需要另一套证据, 本页不作后一种承诺。
最后一句留给规模。资料指出, 成人最常见的听力损失原因 —— 接触强声与耳毒性药物 —— 是可以预防的; 而全球超过 10 亿年轻人因不安全的聆听习惯, 正处在永久性、可避免的听力损失风险之中。
Debunked · auditory training trains the brain, not the cochlea
Train a bit and get your hearing back — that pitch contains a swap. To take it apart, separate two things: one sits in the cochlea and is the hardware that picks sound up; the other sits in the brain and is what gets made of the sound that arrives. Training can reach the second one.Why can it not reach the first? No spontaneous hair cell regeneration has been observed in the adult mammalian cochlea. In mice, during the first postnatal week some supporting cells still retain the ability to give rise to new hair cells, and that ability is progressively lost after birth. At least one layer of the reason is epigenetic: mouse supporting cells progressively accumulate DNA methylation at the promoters of hair-cell genes, and that methylation overlaps the binding sites of Atoh1, a key transcription factor for hair cell fate.
Both lines need their limits attached: the data are from mice, the work was done on cultured cochlear explants rather than in a living animal, let alone a human; and methylation is only one of several epigenetic layers, not the single switch.
So what does training actually improve? A 2013 systematic review pooled 13 articles on individual computer-based auditory training for people with hearing loss: learning did generalize to untrained measures of speech intelligibility (in 11 of the 13), but the improvements were small and not robust; study quality ranged from very low to moderate, and the authors concluded the evidence cannot reliably be used to guide intervention.
Two boundaries: the review covers self-administered computer-based training and does not cover group aural rehabilitation or clinician-led training; and it describes the state of evidence in 2013, not a settled answer today.
The judgment tool: next time you meet a hearing-training pitch, ask two questions. Does it improve thresholds on an audiogram, or a score on the trained task? And does it say how large the generalization to everyday speech actually was? Training helping someone is one claim; training repairing a cochlea is a different one.
nguyen-2023-cochlea-methylationhenshaw-2013-auditory-training
Debunked · can a supplement protect your ears?
Part of the damage loud sound does inside the cochlea involves oxidative stress, so protecting ears with antioxidants is a natural idea in the laboratory. The problem is the next step: it has not cashed out in humans.Take N-acetylcysteine (NAC). Researchers summarize it this way: it has consistently reduced permanent noise-induced hearing loss in laboratory studies, but its clinical efficacy is still controversial. The paper that wrote that sentence is itself part of the controversy: in a double-blind, placebo-controlled trial of 566 military personnel, the primary endpoint (rate of significant threshold shift) did not come out positive — the authors' own wording is that the study design failed to confirm it. Only a secondary endpoint and post-hoc analyses showed a signal. When the primary endpoint is negative, using a secondary result to declare efficacy is cherry-picking.
On tinnitus, the language is blunter. The 2014 clinical practice guideline on tinnitus from the American Academy of Otolaryngology-Head and Neck Surgery includes a recommendation *against*: clinicians should not recommend Ginkgo biloba, melatonin, zinc, or other dietary supplements for treating persistent, bothersome tinnitus.
Keep the proportion. That is the middle tier, a recommendation (against), not a strong recommendation; the guideline graded the aggregate evidence quality for this statement as C, with the underlying trials conflicting and methodologically flawed; and the guideline panel itself was divided on the harm-benefit balance. So the accurate phrasing is not shown to work, not shown not to work. The scope is narrow too: adults, primary persistent bothersome tinnitus, and supplements taken as a tinnitus treatment — it says nothing about whether zinc matters to normal ear physiology.
Ginkgo later got a harder pooled look, and the answer was still we do not know. A 2022 Cochrane review found 12 studies with 1915 participants in total, but only 2 studies, 85 people, could actually be pooled for tinnitus symptom severity: Ginkgo may have little to no effect compared with placebo (Tinnitus Handicap Inventory, 0-100 scale, lower is better, mean difference -1.35, 95% CI -8.26 to 5.55), on very low-certainty evidence; the authors concluded there is uncertainty about both benefits and harms. That interval is wide enough to hold either answer — the lower bound of -8.26 cannot rule out a meaningful benefit, and the upper bound of 5.55 cannot rule out a small harm. It neither proves the supplement works nor has the power to prove it does not. Reading '1915 people studied' as a sign the conclusion is solid is exactly the trap here.
Ginkgo is not free of bodily cost either. The guideline notes that its flavonoids and terpenoids have antiplatelet activity and inhibit clotting when combined with anticoagulants, with reported bleeding, hematoma, neurological deficit and even death; it may also interact with thiazide diuretics to raise blood pressure.
There is a deeper mismatch of mechanism, too. Many pitches built on improving inner-ear blood flow assume tinnitus is the ear ringing. But surgically sectioning the auditory nerve does not eliminate tinnitus in every case, which favors a more central than peripheral origin. The same review then takes half of that back: many forms of tinnitus are now considered a complex interaction between peripheral and central mechanisms along the auditory pathway — cochlear damage is often the trigger, but the percept is sustained centrally. An approach that only pushes blood into the ear may be aimed at the wrong place from the start.
For contrast: the only tinnitus intervention that has ever reached moderate certainty is cognitive behavioural therapy, aimed at the brain — and what it changes is the distress tinnitus causes, not the sound going away.
kopke-2015-nac-nihl-rcttunkel-2014-tinnitus-cpgsereda-2022-cochrane-tinnitushaider-2018-tinnitus-central-originfuller-2020-cbt-tinnitus
Practice · earplugs only count if they are seated right
If not breaking it is the only lever, then hearing protection is the most direct gate you have. But it comes with a prerequisite people skip: how you wear it decides how much it actually blocks.In a Cochrane systematic review, two randomized controlled trials (140 people in total) compared having been taught how to insert earplugs properly against not: the instructed group got 8.59 decibels more attenuation (95% CI 6.92 to 10.25), on moderate-quality evidence.
Three limits cannot be dropped:
What was measured is earplug attenuation, a surrogate outcome, not hearing loss itself; no study in the review showed that insertion training reduced actual hearing lossOnly short-term follow-upTwo other conclusions in the review (stricter legislation reducing workplace noise, better hearing-protection-device use within hearing loss prevention programmes reducing risk) rest on very low-quality evidence
Even so, this remains the single most directly evidenced actionable step in this scene: the same pair of earplugs, seated differently, measured 8.59 dB apart.
tikka-2017-noise-hearing-loss
Red flag · sudden hearing loss is an emergency
Everything above is the slowly accumulating ledger. One situation is completely different and deserves its own slot in memory: sudden sensorineural hearing loss (SSNHL).The clinical practice guideline defines it as hearing that drops rapidly within a 72-hour period, is sensorineural, measures 30 decibels or more, and affects at least three consecutive frequencies.
There is one thing here that is very easy to read backwards, so it has to be said outright: the 72 hours is the window in which the loss develops, not 72 hours you have in which to get treated. Its job is to separate this kind of rapid onset from hearing that erodes over years.
Two more points need stating:
The 30 decibels and three consecutive frequencies are numbers on an audiogram, a convention for research and trial enrolment, not a clinical gate. The guideline itself notes that in clinical practice, expanding the definition to cases with less than 30 decibels of loss may be considered. Falling short of 30 decibels does not rule the condition out.You cannot measure this yourself; a clinician confirms it with an audiogram. And because a pre-illness audiogram usually does not exist, the practical reference is the threshold in the opposite ear.
Why the urgency? The literature places it in the see-someone-quickly category, notes that recognizing and managing it promptly may improve the chance of hearing recovery, and treats it as a signal that must be worked up for vestibular schwannoma (acoustic neuroma), stroke, and malignancy. The timing anchor the guideline gives is as soon as possible: an audiogram within two weeks of symptom onset.
This page prescribes nothing and does not replace a doctor. It only wants one thing to occupy a slot in your head: one ear going muffled and unclear over hours to days, with tinnitus or vertigo alongside, is not poor sleep and is not earwax — it is something the guidance explicitly places in the see-someone-quickly category.
One note on scope. The World Health Organization states that the most common causes of hearing loss in adults are preventable. Preventable and emergency are not in conflict: the first governs the day-in day-out allowance, the second governs the morning that plays by no rules.
chandrasekhar-2019-sudden-hearing-loss
Practical · make fewer gaps · dose is intensity times time
Since central gain follows the input gap, creating fewer gaps at the input end is the one link in this chain you can act on directly.Start with the premise: the World Health Organization states that the most common causes of hearing loss in adults — such as exposure to loud sounds and ototoxic medicines — are preventable. The same source notes that over 1 billion young adults are at risk of permanent, avoidable hearing loss due to unsafe listening practices. Note the wording: at risk, not already affected — those are entirely different quantities.
Dose = intensity × time, and that's the key to every noise standard. Each 3-decibel rise in level doubles the energy the sound carries, so the time has to halve for the total to stay level. That convention — the equal-energy principle — is the modelling basis of the standards, not a physiological law verified directly in people. The US National Institute for Occupational Safety and Health writes its recommended exposure limit on exactly that rule: 85 dBA as an 8-hour time-weighted average, with a 3-dB exchange rate — 85 dBA for 8 hours, 88 dBA for 4 hours, 91 dBA for 2 hours.
Two limits travel with it: this is a recommended exposure limit (the legally enforced US figure, OSHA, uses a different set of numbers), and its frame is occupational exposure over an 8-hour workday; lifting it directly onto concerts, headphones, or gym volume goes beyond what that document covers.
Consumer electronics have their own standard. The 2019 WHO-ITU safe listening standard defines two modes: 80 dB for 40 hours per week for adults, and 75 dB for 40 hours per week for sensitive users such as children. It recommends that personal audio devices include software tracking the level and duration of a user's sound exposure as a percentage of a reference exposure — the so-called sound allowance — and it requires that users be able to see how much of that allowance they have used, presented in an easy-to-understand way. The allowance being settled weekly is the practically useful part: one very loud exposure does not just spend today, it takes a bite out of the whole week.
Two honest additions. First, the standard is voluntary, not law; a manufacturer that ignores it is not breaking any rule. Second, and more important: what the standard says and listening this way is definitely safe are two different claims. The first is what the document contains; the second would need its own body of evidence. Don't stitch them together.
On earplugs, there's a counterintuitive finding. Two randomized trials (140 participants) in a Cochrane review found that people given instruction in proper insertion achieved earplug attenuation 8.59 dB better (95% CI 6.92 to 10.25) than those without instruction, on moderate-quality evidence. Same earplug, different insertion — that is the size of the difference.
Three caveats, stated plainly: what was measured is earplug attenuation, a surrogate — not hearing loss itself; follow-up was short-term only, and the review's authors say long-term follow-up is still needed; and two other conclusions in the review (that stricter legislation reduces workplace noise, and that better hearing-protector use within hearing loss prevention programmes reduces risk) rest on very low-quality evidence.
Is there a pill that shields the ear? N-acetylcysteine (NAC) is among the most studied antioxidants here: researchers summarise it as having consistently reduced permanent noise-induced hearing loss in the laboratory, while its clinical efficacy remains controversial. The randomized, double-blind, placebo-controlled trial in military personnel (566 participants; 277 NAC, 289 placebo) was negative on its primary endpoint — the authors' own words are that the study design failed to confirm it, with signals appearing only in secondary and post-hoc analyses.
So the correct reading is: consistent in the laboratory, unsettled in humans. Using a secondary endpoint or a post-hoc analysis to declare it effective is exactly the manoeuvre this site exists to take apart.
To close the page: what you can change directly isn't the knob in the brain — it's how much energy reaches your cochlea each day. Level, duration, and getting your hearing protection seated properly: those three levers are more substantial than any bottle on a shelf.
tikka-2017-noise-hearing-losskopke-2015-nac-nihl-rct
References · 13
- Purves, D., Augustine, G. J., Fitzpatrick, D., Katz, L. C., LaMantia, A.-S., McNamara, J. O., & Williams, S. M. (Eds.). (2001). The inner ear. In Neuroscience (2nd ed.). Sinauer Associates. https://www.ncbi.nlm.nih.gov/books/NBK10946/ www.ncbi.nlm.nih.gov/books/NBK10946
- Purves, D., Augustine, G. J., Fitzpatrick, D., Katz, L. C., LaMantia, A.-S., McNamara, J. O., & Williams, S. M. (Eds.). (2001). The inner ear. In Neuroscience (2nd ed.). Sinauer Associates. An acoustic stimulus initiates a traveling wave that propagates from the base toward the apex of the basilar membrane; because the membrane is narrower and stiffer at the base and wider and more flexible at the apex, high frequencies peak at the base and low frequencies at the apex, giving rise to a topographical (tonotopic) mapping of frequency. www.ncbi.nlm.nih.gov/books/NBK10946
- Purves, D., Augustine, G. J., Fitzpatrick, D., Katz, L. C., LaMantia, A.-S., McNamara, J. O., & Williams, S. M. (Eds.). (2001). The inner ear. In Neuroscience (2nd ed.). Sinauer Associates. The cochlea not only amplifies sound and transduces it into neural signals but also acts as a mechanical frequency analyzer: because basilar membrane stiffness is graded from base to apex, high frequencies peak at the base and low frequencies at the apex (tonotopy), and complex sounds produce a vibration pattern equivalent to the superposition of the vibrations generated by their individual component tones. The same section notes that peripheral tuning is too sharp to be explained by passive mechanics alone and that an active biomechanical process (outer hair cells) contributes, so the superposition description is a passive-linear idealization. www.ncbi.nlm.nih.gov/books/NBK10946
- Liberman, M. C., Gao, J., He, D. Z. Z., Wu, X., Jia, S., & Zuo, J. (2002). Prestin is required for electromotility of the outer hair cell and for the cochlear amplifier. Nature, 419(6904), 300-304. https://doi.org/10.1038/nature01059 10.1038/nature01059
- Nin, F., Hibino, H., Doi, K., Suzuki, T., Hisa, Y., & Kurachi, Y. (2008). The endocochlear potential depends on two K+ diffusion potentials and an electrical barrier in the stria vascularis of the inner ear. Proceedings of the National Academy of Sciences, 105(5), 1751-1756. 10.1073/pnas.0711463105
- Franco, B., & Malgrange, B. (2017). Concise review: Regeneration in mammalian cochlea hair cells: Help from supporting cells transdifferentiation. Stem Cells, 35(3), 551-556. https://doi.org/10.1002/stem.2554 10.1002/stem.2554
- Le, T. N., Straatman, L. V., Lea, J., & Westerberg, B. (2017). Current insights in noise-induced hearing loss: A literature review of the underlying mechanism, pathophysiology, asymmetry, and management options. Journal of Otolaryngology - Head & Neck Surgery, 46, 41. https://doi.org/10.1186/s40463-017-0219-x. Early or moderately advanced NIHL produces the typical notch at 4 kHz, spreading to the neighbouring frequencies of 3 kHz and 6 kHz, with some hearing recovery at 8 kHz; the ~4 kHz vulnerability is attributed to ear-canal resonance plus middle-ear mechanics, and the notch fades with aging. 10.1186/s40463-017-0219-x
- Motlagh Zadeh, L., Silbert, N. H., Sternasty, K., Swanepoel, D. W., Hunter, L. L., & Moore, D. R. (2019). Extended high-frequency hearing enhances speech perception in noise. Proceedings of the National Academy of Sciences, 116(47), 23753-23759. Among 116 adults with normal conventional audiograms (0.25-8 kHz, all thresholds 20 dB HL or better; mean age 29.5 y, range 18-65), 74 (64%) had elevated thresholds above 8 kHz and 39 (34%) reported difficulty hearing speech in noise; the extended high-frequency threshold average discriminated those reporting difficulty better (ROC area 0.81) than the conventional-frequency average (ROC area 0.71). 10.1073/pnas.1903315116
- Kujawa, S. G., & Liberman, M. C. (2009). Adding insult to injury: Cochlear nerve degeneration after "temporary" noise-induced hearing loss. The Journal of Neuroscience, 29(45), 14077-14085. https://doi.org/10.1523/JNEUROSCI.2845-09.2009 10.1523/JNEUROSCI.2845-09.2009
- Bramhall, N., Beach, E. F., Epp, B., Le Prell, C. G., Lopez-Poveda, E. A., Plack, C. J., Schaette, R., Verhulst, S., & Canlon, B. (2019). The search for noise-induced cochlear synaptopathy in humans: Mission impossible? Hearing Research, 377, 88-103. https://doi.org/10.1016/j.heares.2019.02.016 10.1016/j.heares.2019.02.016
- National Institute for Occupational Safety and Health. (1998). Criteria for a recommended standard: Occupational noise exposure, revised criteria 1998 (DHHS (NIOSH) Publication No. 98-126). U.S. Department of Health and Human Services, Centers for Disease Control and Prevention. Sets the recommended exposure limit at 85 dBA as an 8-hr time-weighted average with a 3-dB exchange rate; Table 1-1 lists 85 dBA for 8 hr, 88 dBA for 4 hr, and 91 dBA for 2 hr. 10.26616/NIOSHPUB98126
- World Health Organization, & International Telecommunication Union. (2019). Safe listening devices and systems: A WHO-ITU standard. World Health Organization. https://www.who.int/publications/i/item/9789241515276 www.who.int/publications/i/item/9789241515276
- World Health Organization. (2026). Deafness and hearing loss [Fact sheet]. Geneva: WHO. www.who.int/news-room/fact-sheets/detail/deafness-and-hearing-loss