Reading Audio Specifications

A specification is a measurement, and every measurement was taken under conditions somebody chose. The number itself is usually honest, because a manufacturer is on the hook for what it prints. What that number was measured with, at what level, across what band and through which filter is the part left to judgement — and that is where the truth about a piece of gear actually lives.

Musicians read spec sheets the way they read prices, as though two figures in the same units could simply be set against each other. Little of that survives contact with how the figures are made. Two preamps quoting the same noise number can be ten decibels apart once you measure them the same way, because one ran its figure through a filter that throws away the frequencies you were worried about — and both spec sheets are correct.

A Range With No Tolerance Is Not a Response

Two numbers and a dash are not a frequency response. They are a frequency range — the region where something is still coming out of the far end. A box quoted as 45 Hz to 21 kHz has told you nothing about whether it is flat, because nobody said what "to" means.

Add the tolerance and it becomes a claim you can act on. 45 Hz to 21 kHz, ±2 dB describes a genuinely even device — the same band at ±10 dB describes something with strong opinions about the top end. Half a decibel is what a careful manufacturer publishes, three decibels is the common looser one, and much wider is a tone control sold as a specification.

A tolerance also needs an anchor. It is quoted against a reference tone, conventionally 1 kHz, so ±2 dB means two decibels from wherever the device sits at 1 kHz, not two decibels from flat. Where the band stops is a separate decision about how far down the maker was willing to let the response fall, and −3 dB is the usual answer because that is exactly half the power. Even a stated tolerance flattens the shape. Two decibels of wander can be a gentle tilt across the top two octaves, which colours everything you put through it — or one narrow dip you would never find by ear. Only the curve separates them.

A Noise Figure Is Three Numbers Wearing One Coat

The headroom lesson established that a dB figure with no reference and no weighting compares to nothing. The next question is sharper — which weighting, because that choice moves the printed number more than most of the engineering behind it. A-weighting was built from how people hear at quiet levels, and it is aggressive. It discounts 50 Hz by 30.27 dB and 20 Hz by 50.39 dB, as the decibel lesson sets out, while handing back 1.27 dB around 2.5 kHz where hearing is keenest. What that does to a noise figure is not what most people assume — run it over genuinely white noise across the audio band and it improves the number by barely two decibels.

It flatters a real product by far more than that, because a noise floor in the wild is not white. They carry mains hum. A-weighting removes 16.7 dB from the second harmonic of a 60 Hz supply and 12 dB from the third, so the filter is most generous to the product with the worst hum problem — backwards from what a buyer needs to know. Measure the same noise to ITU-R 468 instead and the figure climbs roughly 11 dB again. One device, three defensible numbers, and only the smallest of them reaches the brochure.

Bandwidth is the third variable, conventionally 22 kHz so the limiting filter has somewhere to work without touching 20 kHz. All three conditions land on the one noise spec that genuinely separates preamps — equivalent input noise. Its input is terminated in 150 ohms, the impedance a microphone would present, and its gain is run wide open, because nothing downstream can undo the noise that resistor itself makes. The headroom lesson carries that floor and what good electronics achieve against it. What belongs here is that two preamps two decibels apart on paper may be two preamps and one filter.

THD Describes One Tone at One Level

Total harmonic distortion means one pure tone in, everything harmonically related to it out, summed as a fraction of the signal. THD+N folds in every other thing the device contributed — hiss, hum, buzz, radio interference. For one device the plus-N figure is always the larger of the two. That makes a bare THD number the flattering choice, and one company's THD against another's THD+N no comparison at all.

The plus-N figure also depends on the window it was measured through. Bandwidths of 22, 30 and 80 kHz are all in use, and a wider window admits more noise from the same box. Level does similar work. THD+N is a ratio, and the noise half of it does not shrink when you turn the signal down. One channel measured near the top of its range and measured quietly yields two very different percentages — so a spec naming no level has left out the thing that set it. Frequency is the third lever, and 0.004 % at 1 kHz and 0.02 % at 12 kHz can be one channel described twice.

Intermodulation distortion is the more revealing test and the less advertised one. Two tones go in, and the device manufactures sum and difference products related to neither. The SMPTE method uses 60 Hz and 7 kHz at a four-to-one amplitude ratio; the CCIF method pairs 19 kHz and 20 kHz at equal level and measures the 1 kHz difference product they should never have produced. Both reach the page as "IMD" with nothing further attached — and they will not agree. IMD tracks what you hear better than THD for a plain reason: a second harmonic sits an octave above the note that made it and hides inside it, while an intermodulation product is musically unrelated to anything on the stage.

An Interface Spec Is Half of a Sentence

An impedance figure and a level figure describe a relationship, not a property. Print either one on its own, with nothing said about what it connects to, and there is nothing there to read. Nominal impedance is a bracket, not a reading. Under the EIA convention any microphone measuring anywhere between 80 and 300 ohms is sold as 150 ohms nominal — a label two mics can share while measuring almost four times apart. Loudspeakers stretch the same label over a curve that moves with frequency, which the power and impedance lesson takes apart properly.

At line level the two ends are deliberately mismatched. Outputs sit between 100 and 600 ohms, inputs at 10 kΩ or higher. The job is to hand over voltage rather than power — a load an order of magnitude above the source barely draws on it, so the voltage arrives intact. The cables lesson covers the case that actually bites players: a passive pickup meeting an input built for a microphone.

Level standards fail in the same shape. Professional gear centres on +4 dBu and consumer gear on −10 dBV, and the gap is 11.8 dB rather than the 14 still printed in reference works — the decibel lesson explains why subtracting the two numbers does not work. The other half of a level spec is the ceiling. "+22 dBu balanced, into 10 kΩ, at 0.5 % THD" is a claim you can plan a system around; "+22 dBu" alone is not, because the load and the distortion limit are what set it. The distance from nominal up to that maximum is your headroom.

Most Specs Stopped Discriminating Years Ago

A specification is one point on a curve, and music happens across the whole curve at once. The conditions a manufacturer picks are nearly always the ones where a device is at its best: mid-band, unity gain, nominal level, an easy resistive load. None of that describes a channel wound up with a snare on it — and the gap between the sheet and the room opens there, not in any dishonesty.

That gap has closed in one direction and widened in another. Noise and distortion in ordinary modern gear passed the point of audibility years ago, so choosing between two interfaces whose THD+N figures differ in the fourth decimal place is choosing on a difference no listener recovers in a full room. What still separates one box from another is mostly unpublished — how a preamp behaves at the top of its gain range, what a microphone does with sound off its axis, whether a converter stays graceful in its last three decibels.

Other specs are true, current and beside the point. A response flat to 40 kHz is a real achievement — and it will not survive the first reflection off a back wall. A digital peak meter reading −0.2 dBFS is accurate about the samples it saw and still lets the signal clip on the way out, because reconstruction interpolates between them. True-peak metering exists to oversample and catch what sample peaks structurally cannot. Even the way the numbers are made has moved: a published response curve comes from a swept sine now, because a real impulse pushes most gear out of the linear region it was meant to describe.

What to Check Before You Believe a Number

• Find the tolerance before you read the band — a range with no ± says where sound still emerges, not how flat the box is.
• Check what the tolerance is measured against — ±2 dB means two decibels from a 1 kHz reference tone, not two decibels from flat.
• Ask which weighting — A-weighted and unweighted measure one device differently, and ITU-R 468 lands about 11 dB above the A-weighted figure.
• Distrust a noise number generous by about what hum would cost — A-weighting removes 16.7 dB at 120 Hz, rewarding the product you would least want to own.
• Compare THD with THD and THD+N with THD+N — the plus-N figure is always the larger, and it moves with measurement bandwidth and signal level too.
• Check which IMD standard was used — SMPTE's 60 Hz and 7 kHz pair and CCIF's 19 kHz and 20 kHz pair provoke different failures under the same three letters.
• Read impedance as a pair — nominal is a bracket, and a line output near 100 ohms feeding an input of 10 kΩ or more is what everything else assumes.
• Subtract nominal level from maximum output — that gap is the headroom you own, and neither figure means anything without its load and distortion limit.
• Prefer the graph — a curve shows the shape, the frequency and the exception; a single figure is that curve with the inconvenient parts averaged out.