Sound – How Vibrating Air Carries Meaning

Sound is one of the more ordinary phenomena to experience and one of the more surprising to examine. Nothing travels from a speaker to your ear except a pattern of pressure — air molecules being pushed slightly closer together and slightly further apart, passing the disturbance along without going anywhere themselves.

Waves of pressure, not of matter

When something vibrates, it pushes on adjacent air molecules, which crowd together, push on their neighbours, and spring back. The compression propagates outward while each individual molecule oscillates around roughly the same position.

This makes sound a longitudinal wave — the oscillation runs along the direction of travel, unlike light or a ripple on water, where the motion is perpendicular. It also means sound needs a medium. In vacuum there is nothing to compress, which is why space is silent.

In air at around twenty degrees, sound travels at roughly 343 metres per second. In water it is more than four times faster, and in steel faster still, because a stiffer medium passes the disturbance along more quickly.

Frequency and amplitude

Two properties describe a simple sound. Frequency is how many compressions pass per second, measured in hertz, and we perceive it as pitch. Amplitude is how large the pressure variation is, and we perceive it as loudness.

Human hearing spans roughly 20 Hz to 20,000 Hz, with the upper limit falling steadily throughout life. Our sensitivity is not uniform — we are most responsive between about 2,000 and 5,000 Hz, which is precisely the range that carries the consonants of human speech.

Loudness is measured logarithmically in decibels because our perception is logarithmic. A ten-decibel increase represents ten times the sound intensity but registers as roughly twice as loud.

Why a violin does not sound like a flute

Real sounds are never single frequencies. A vibrating string produces a fundamental frequency plus a series of harmonics at multiples of it, and the relative strength of those harmonics is what we hear as timbre.

Play the same note on a violin and a flute and the fundamental frequency is identical. The difference is entirely in which harmonics are present and how strong they are — and our hearing is extraordinarily good at telling those patterns apart, which is why you can recognise a voice from a single word.

How speech encodes information

Speech is a particularly clever use of the mechanism. The vocal folds produce a buzzing sound rich in harmonics. The throat, mouth, tongue and lips then form a cavity that amplifies certain frequency bands and suppresses others.

Those amplified bands are called formants, and their positions are what distinguish one vowel from another. Move your tongue and you change the cavity’s resonance, shifting the formants, and a listener hears a different vowel. Consonants add bursts, hisses and abrupt transitions on top.

All of this is visible on a spectrogram — a plot of frequency against time — where vowels appear as horizontal bands and consonants as vertical smears. Speech recognition systems work largely on this representation rather than on the raw waveform, and the difficulty of the task becomes obvious when you look at one: the boundaries between words are frequently not there at all. We hear separate words in continuous speech because we know the language, not because the acoustic signal is segmented.

This is why background noise, overlapping speakers and unfamiliar accents cause such trouble for automated systems, and why the practical limits of speech recognition on real telephone calls have more to do with acoustics than with intelligence. A telephone line transmits roughly 300 to 3,400 Hz, discarding a good deal of the signal before any processing begins.

Two ears, one location

You locate sounds using differences between your ears. A sound from the left arrives at the left ear a fraction of a millisecond earlier and slightly louder, and your brain converts those tiny differences into a direction with remarkable precision.

For higher frequencies the shape of the outer ear also filters sound differently depending on elevation, which is how you can tell whether a sound came from above or below despite having only two sensors on a horizontal line. It is a considerable amount of computation performed on nothing more than fluctuating air pressure.