Frequency response: reading the spec and the graph

How a response curve is made

Play a test signal whose level is the same at every frequency, usually a slow sine sweep or a burst of noise, and measure what comes out. For an amplifier or a DAC that means the output voltage. For a speaker it means the sound pressure at a microphone, and for a headphone the pressure inside an artificial ear. Plot the level in decibels against frequency, on a scale that gives each octave the same width, and you have the curve.

The curve is read against itself. A 3 dB peak at 3 kHz means that region plays 3 dB louder than the level around it, and a tolerance is stated from a reference level, often the level at 1 kHz.

Electronics and transducers sit at opposite ends here. Most amplifiers and DACs stay flat to within a fraction of a decibel across the audible band. Headphones and speakers routinely vary by several decibels from one region to the next, which makes response the most audible difference between them. Harman's research, summarized by Sean Olive in 2022, found that listeners' ratings of headphones can be predicted from how far each one's response strays from a target curve.

A range needs a tolerance

"20 Hz to 20 kHz" names two frequencies and stops there. It doesn't say how loud the output still is at either end, or how far it wanders between them. Output fades gradually toward the extremes, so without a stated level the endpoints can sit wherever the maker chose to read them.

A complete specification adds a tolerance. "20 Hz to 20 kHz ±3 dB" promises that the output stays within 3 dB of the reference level across the whole band. A split figure such as +1/−3 dB allows 1 dB above the reference and 3 dB below it.

Speakers are often quoted by their −3 dB points instead, which tell you how far the output has fallen at each end and nothing about the band between. The Diora Acoustics Chors 5 is quoted that way, 44 Hz to 22 kHz at −3 dB, and its review adds what an endpoint can't show: as a sealed box it rolls off slowly below 44 Hz, so there is still usable output at 35 Hz.

Every headphone in this catalogue with a published range quotes it bare. Figures far above 20 kHz describe output nobody hears, since young ears top out around 20 kHz and the limit falls with age, so a headphone quoted to 65 kHz tells you no more about the audible band than one quoted to 22 kHz. The bottom figure matters more, and the sub-bass explainer covers how to read it.

Reading a response graph

Frequency runs along the bottom, from 20 Hz at the left to 20 kHz at the right, and level runs up the side in decibels. Check the vertical scale before anything else. The same curve drawn on a grid spanning 50 dB looks far flatter than it does on one spanning 20 dB.

Smoothing matters as much. Most published curves are averaged over a fraction of an octave, which rounds off narrow peaks and dips, and two curves smoothed differently don't compare. The HIFIMAN Edition XS review points this out on the graph it reads: the headphone's trace was less smoothed than the dotted Harman target drawn beside it, which is why it shows some weird-looking dips.

Then weigh what you see by how audible it is. A broad lift of a decibel or two across an octave is easier to hear than a narrow spike several times as tall, and a peak is easier to hear than a dip of the same size. A gentle tilt deserves more attention than its smooth look suggests, and a narrow notch less. A few landmarks help you place things: bass weight sits below about 200 Hz, vocal presence between 2 and 4 kHz, and sibilance between about 6 and 10 kHz.

What flat means for speakers and headphones

For a speaker, flat means flat on its main axis, measured in an anechoic chamber or with the room's reflections cut out of the measurement. At your seat the curve changes. The room adds sound that left the speaker at an angle, which usually carries less treble, so a speaker that measures flat on axis shows a gently falling top end at the listening position. Below a few hundred hertz the room's own resonances take over, which the room mode explainer covers.

For a headphone, flat at the eardrum would sound dull. Sound from a speaker reaches your eardrum after your head, outer ear and ear canal have shaped it, adding a boost of more than 10 dB around 3 kHz. A headphone that sounds neutral has to produce a similar shape at the eardrum.

That's why headphone curves are measured on a head simulator with artificial ears and read against a target. International headphone standards recommend a diffuse-field target, modeled on sound arriving from all directions at once. Harman's research target started instead from the in-room response of an accurate loudspeaker.

So a headphone called flat is flat against the target its maker chose. The OLLO Audio X1 makes the claim with a number attached: every unit is measured and calibrated to the same flat curve, typically within ±1 dB, with the correction run as a plugin in a recording program or as EQ on an audio interface. Its review hears no boosted bass, no scooped mids and no splashy highs.

Why two graphs of one headphone disagree

A headphone graph depends on the rig as much as the headphone. Head simulators differ in their artificial ears, and pads that seal perfectly on a rig can leak on a real head, where glasses or hair let bass escape. Shifting a headphone a few millimeters on the rig moves its treble peaks, and in the top octaves real ears differ from each other, and from any rig, more than they do lower down. In-ears add insertion depth to the list, which the IEM explainer goes into.

The HIFIMAN HE400se shows how far apart the results can land. Its reviewer found published graphs that were vastly different from each other: some showed a huge drop in the bass, others a linear response down to about 20 Hz. By ear, the review found the bass extension lacking below 60 Hz.

Compare curves from one source, measured on one rig and drawn against one target. Two graphs from different places can both be correct and still disagree.

Using response when you listen and buy

Reviews on this site are written from listening, with measurements added where they reveal something audible or contradict a marketing claim, as how we test explains. They describe the shape region by region, and the overall balance they hear is summed up as a sound signature: warm, bright, V-shaped and so on.

Response is also the property of a headphone that's easiest to change after you buy it. An equalizer can move a headphone toward a target, which is how the X1's calibration is delivered, and the EQ simulator lets you hear what each band does. EQ can't make up for a seal that changes every time you put the headphones on, or for a reflection in a room, and a correction built from someone else's graph is least reliable in the top octaves, where your ears differ most from the rig.

To connect a graph with what you hear, learn the bands by ear. The EQ ear training game plays tones and sweeps from 20 Hz to 22 kHz and asks you to name the frequency. Once you can tell a lift at 3 kHz from one at 8 kHz, a response graph starts to read like a description.