Sound waves become electrical signals through a chain of two distinct transductions. A microphone converts pressure fluctuations into a continuously varying voltage by moving a diaphragm, coil, or capacitor plate. An analog-to-digital converter then samples that voltage thousands of times per second and rounds each sample to a discrete number. The result is a stream of binary values a processor can store, transmit, or analyze.

The first stage is physical and analog. The second is numerical and discrete. Most explanations collapse them into one step and lose the part that determines whether a recording is faithful or ruined. This article separates the two, quantifies each, and shows where the losses hide.

Pressure to voltage: the first transduction

A sound wave is a traveling pattern of pressure differences. At conversational levels the pressure swings above and below atmospheric pressure are tiny, on the order of 0.02 pascals against a baseline of about 101,325 pascals. A microphone's job is to make something measurable out of that whisper-thin oscillation.

Every microphone type does this differently, but the logic is shared. A diaphragm or membrane moves in response to pressure. That mechanical motion modulates an electrical property, and a circuit converts the modulation into a voltage that tracks the waveform. The voltage is an analog of the pressure wave, not a copy of it.

Why does the conversion need a power source or a bias? Because a passive diaphragm produces motion, not electricity. Dynamic microphones generate their own signal through electromagnetic induction, so they need no power. Condenser microphones need a charge on their capacitor plates, supplied by phantom power or a battery. That difference shapes everything downstream.

Dynamic microphones use a coil attached to the diaphragm, suspended in a magnetic field. When the diaphragm moves, the coil moves through the field, and Faraday's law of induction generates a voltage proportional to the coil's velocity. This is why dynamic designs are rugged and handle high sound pressure levels, and also why their output is faint and needs a preamplifier.

Condenser microphones place the diaphragm as one plate of a capacitor, with a fixed backplate behind it. A constant charge sits on the plates. As the diaphragm moves, the capacitance changes, and the voltage across the plates changes with it. Because the gap is small and the charge is fixed, the output tracks diaphragm displacement rather than velocity, which is one reason condensers are more sensitive and more detailed.

A third family uses a permanently charged electret material, which holds its charge without external bias. These dominate consumer devices, from headsets to hearing aids, because they are cheap and small. The physics is still capacitive, just with the charge baked in.

Piezoelectric microphones use a crystal or ceramic that generates charge when mechanically stressed. They appear in contact mics and some measurement instruments. MEMS microphones, the kind inside phones and laptops, are capacitive sensors etched onto silicon, often with the diaphragm and backplate built as microscopic structures. They are the highest-volume microphone type in the world.

The output of any of these is a small analog voltage. A dynamic mic might produce a few millivolts at normal speech. A condenser might produce tens of millivolts. Either way, the signal is weak, and it must be amplified before anything digital happens.

Sampling and quantization: the second transduction

An analog voltage is continuous in both time and amplitude. A digital system cannot store continuity. It must take snapshots at regular intervals and round each snapshot to a finite set of values. This is the analog-to-digital conversion, and it introduces two specific, quantifiable errors.

Sampling rate controls how often the voltage is measured. The Nyquist-Shannon sampling theorem states that to reconstruct a signal faithfully, you must sample at more than twice its highest frequency. Human hearing extends to roughly 20,000 hertz. That sets the practical floor for high-fidelity audio at just over 40,000 samples per second.

The standard for compact discs is 44,100 samples per second, chosen to clear the 20 kilohertz ceiling with margin for anti-aliasing filters. Professional audio often runs at 48,000 or 96,000 samples per second. Telephony traditionally sampled at 8,000 samples per second, which caps audio at about 4 kilohertz and explains why phone calls sound thin.

Why the factor of two? A sine wave sampled at exactly twice its frequency yields two points per cycle, the minimum needed to define the wave's shape. Sample any slower and the reconstruction aliases, producing a false lower frequency that cannot be distinguished from a real one. Anti-aliasing filters remove content above half the sample rate before conversion, which is why the standard sits slightly above the theoretical minimum.

Bit depth controls how finely each sample is rounded. An 8-bit converter divides the voltage range into 256 levels. A 16-bit converter uses 65,536 levels. A 24-bit converter uses about 16.7 million. More levels mean less rounding error, called quantization noise, and a wider dynamic range between the quietest and loudest representable sounds.

The rule of thumb: each additional bit adds about 6 decibels of dynamic range. Sixteen-bit audio gives roughly 96 decibels, which covers the full range of human hearing from threshold to pain. Twenty-four-bit gives about 144 decibels, well beyond what ears can use but useful for headroom during recording and mixing.

Here is the counterintuitive part. Higher sample rates and bit depths do not automatically produce better-sounding recordings for human listeners. Once the format clears the hearing range and the noise floor, further increases buy margin, not audible fidelity. The perceived quality difference between 16-bit and 24-bit playback is generally inaudible under normal conditions. The real gains from 24-bit are in the recording and processing stages, where extra headroom prevents clipping and preserves detail through repeated edits.

Quantified context: what the numbers actually mean

The specifications on a microphone or audio interface are not marketing fluff. They describe measurable limits, and knowing the ranges helps you judge equipment and diagnose problems.

Specification Typical consumer value Typical professional value What it governs
Sample rate 44.1–48 kHz 48–192 kHz Maximum representable frequency
Bit depth 16-bit 24-bit Dynamic range and noise floor
Dynamic range 90–96 dB 110–130 dB Distance from noise floor to clipping
Frequency response 50 Hz–16 kHz 20 Hz–20 kHz Range of pitches captured faithfully
Self-noise 25–35 dBA 5–15 dBA Baseline hiss the mic adds
Maximum SPL 110–130 dB 130–160 dB Loudest sound before distortion

These values interact. A microphone with excellent dynamic range but a noisy preamp will still produce a hiss floor set by the preamp. The weakest link in the chain sets the practical limit, which is why professionals match components rather than buying one premium piece.

Signal-to-noise ratio deserves attention. At a 96-decibel dynamic range, the quietest representable sound sits 96 decibels below the loudest. If your recording peaks at a healthy level, ambient room noise and microphone self-noise fill the bottom of that range. Recording too quietly wastes the top of the range on silence and pushes quiet details into the noise floor. Recording too loudly risks clipping, which is unrecoverable.

Edge cases: where the simple model breaks

The clean model of pressure to voltage to numbers assumes a well-behaved source, a clean room, and adequate equipment. Reality violates all three, and the failures are informative.

Ultrasound sits above human hearing but can still matter. Bat calls and medical imaging use frequencies from 20 kilohertz to several megahertz. These require sample rates well above the audio standard, often 192 kilohertz or higher, and microphones with response extending into that range. A standard 44.1 kilohertz recording of an ultrasound source aliases it into the audible band as a false tone.

Infrasound sits below hearing, roughly under 20 hertz. Wind, machinery, and some natural events produce it. It can still overload a microphone's diaphragm or preamp even though it is inaudible, which is why recording setups use low-cut filters. A rumble you cannot hear can still ruin a take.

Impulse sounds, like a gunshot or a handclap, have extremely fast attack times. They stress the diaphragm's mechanical response and the preamp's slew rate. A microphone rated for steady 130 decibels may distort on a transient that briefly exceeds it.

Digital transmission has its own failure modes. Bluetooth audio compresses the stream, and codecs like SBC or AAC discard information to fit bandwidth. The losses are often audible in music. Latency from buffering can exceed 100 milliseconds, which is why wireless monitors are unsuitable for live performance without careful design.

Hearing aids and cochlear implants present a striking edge case. Hearing aids amplify and shape acoustic input, still an acoustic-to-electric conversion inside the device. Cochlear implants go further, converting sound to electrical pulses that directly stimulate the auditory nerve. That is a prosthesis, not a recording, and it bypasses the mechanical transduction of the inner ear entirely.

Misconceptions and the professional caveat

Several beliefs about sound conversion are common and wrong. Correcting them sharpens practical judgment.

Higher sample rates always sound better. False. Beyond the hearing range and noise floor, increases are inaudible in playback. Bit depth and sample rate are the same thing. False. One controls frequency range, the other controls amplitude resolution. A digital signal is an exact copy of the analog wave. False. It is a rounded, sampled approximation, though one good enough that the error is usually below audibility. Microphone specifications predict sound quality. Partly false. Response curves and noise floors matter, but placement and room acoustics often dominate. Clipping can be fixed in software. Mostly false. Clipping destroys information permanently; plugins can disguise it but cannot restore it.

The professional caveat: specs describe limits, not character. Two microphones with identical dynamic range and frequency response can sound different because of transient response, off-axis behavior, and harmonic distortion at high levels. Engineers choose by ear and by use case, not by datasheet alone.

Practical actions: getting the conversion right

You do not need a physics background to apply this. A few concrete steps prevent most of the common failures in recording and playback.

Set sample rate above twice your highest frequency of interest. For speech, 16 kilohertz is ample. For music and general audio, use 44.1 or 48 kilohertz. Reserve 96 kilohertz and above for ultrasonic work, heavy processing, or specific measurement needs.

Choose bit depth by headroom needs. Sixteen-bit is fine for final delivery. Record and edit at 24-bit to leave room for level changes without accumulating quantization error.

Aim peak levels around minus 12 to minus 6 decibels full scale during tracking. This keeps transients from clipping while staying well above the noise floor. Do not normalize to zero; leave margin.

Match your preamp to your microphone. A low-output dynamic mic needs a preamp with at least 60 decibels of clean gain. A high-output condenser needs far less. A mismatch shows up as hiss or distortion, not as a spec-sheet failure.

Treat the room before the gear. A quiet room with modest equipment outperforms a noisy room with premium equipment. Background noise sets a floor no converter can beat.

Limitations: what conversion cannot do

Every stage in the chain is lossy in some dimension, and honest design means knowing which losses matter and which do not.

Microphones impose their own coloration. No capsule is perfectly linear across all frequencies and all angles. Off-axis sound is shaped differently from on-axis sound, which affects how a recording captures a room or a group of instruments. Specifications usually describe on-axis behavior and say little about the rest.

Analog electronics add noise and distortion. Preamplifiers, cables, and converters all contribute. The cumulative effect is usually small when components are well matched and can be large when they are not.

Digital conversion introduces quantization and aliasing, both bounded by design choices. Proper filtering and adequate bit depth keep these below audibility for most purposes.

The final limit is perceptual. Human hearing itself is nonlinear, frequency-dependent, and adaptive. A recording that measures well can still sound wrong, and one that measures imperfectly can sound right. The conversion chain is a tool, and its success is judged by the ears it serves, not by the numbers alone.