Headphone Measurements | Audeze

How to measure headphones

Headphones can’t be accurately measured by just placing a microphone between the headphone drivers. That’s why we use artificial heads with different mic placements to measure our high-quality, high-end headphones.

All the technology in the world means nothing if it doesn’t sound right. Over the years Audeze has developed a wide variety of technologies to improve the sound quality of our headphones. We record live music and our engineers listen closely to ensure the sound is accurate. We engage with music professionals, and sound engineers, and often incorporate their feedback into our designs.

Headphone measurement is an art as much as a science. We begin by measuring the transducers in a standard IEC- baffle to verify their response and performance. Audeze transducers are matched in pairs using these measurements with typical variations within a very tight range of +/- 0.5dB. Then we use artificial measuring heads.

Artificial Measurement Heads?

There are several different types of artificial measurement heads on the market, each with multiple sets of ears (small/medium/large/male/female). Microphone placement within the heads is quite different. Some heads have the microphones near the entrance of the ear canal, some have them near the eardrum.

At Audeze we use multiple testing rigs to measure our headphones. In production, we use a Neumann KU100 with microphones positioned at the entrance of the ear canal. Headphone positioning on the artificial head creates large variations in response above 2kHz, so in the design phase we take measurements at five different positions with each headphone to create an average frequency response benchmanrk graph, then we match that response as closely as possible in production, checking the production graphs against the benchmark graph.

Frequency Response

The question of ideal headphone frequency response is the subject of heated debate among audio professionals and enthusiasts. It’s no secret that speaker manufacturers design for flat frequency response to 1kHz with a slightly down-sloping response of about 6dB between 1kHz and 22kHz measured at the listening position. This response curve results in natural sounding and pleasurable listening. It’s comparatively easy to measure speakers. It can be as simple as placing a calibrated microphone at the listening position or in an anechoic chamber and measuring the frequency response using one of many measurement software packages.

Headphones, in sharp contrast, cannot be accurately measured by just placing a microphone between the headphone drivers. That’s why we use artificial heads to measure our headphones, including, but not limited to, Neumann KU100 mics and dedicated measurement hardware and software. Measuring speakers through microphones placed inside artificial heads is very different from measuring speakers with microphones in free space. With full-range speakers it’s reasonable to expect fairly flat frequency response in a well-treated room or anechoic chamber. It’s also normal to see some roll-off at low and high frequencies. It’s not at all the same with artificial heads and in-ear microphones. That’s because the geometry of the outer and inner ear significantly affects frequency response when measured at the eardrum or ear canal entrance. Even if the driver has a perfectly flat response, the geometry of the outer and inner ear act as signal processors by boosting certain frequencies and attenuating others. By the time sound waves reach your eardrum their frequency and phase content have been altered and are no longer flat. Frequency response measured at the eardrum is more likely to stay flat up to about 200Hz and gradually increase by about 15-to-20dB, reaching a peak at around 3kHz and then rolling off again as illustrated in the figure below.


Just to complicate matters, the geometry of the head and torso contribute to measured frequency response. That’s called the HRTF (Head-Related Transfer Function). Since the geometry of head, torso and ears are unique to each person, their HRTFs are also unique. Of course nobody’s perfectly symmetrical, so HRTFs for the left ear are different from the right ear. Artificial heads try to model an average response for both. Measuring them at the same distances from the left and right ears at the same angle you can still see a difference in frequency response measured by the left and right ears. Though transducer frequency response measured by in-ear microphones isn’t flat, artificial head manufacturers provide frequency response curves that tell us what an ideal response looks like with the artificial head. These frequency response calibration curves are applied to the measured response data to see the response as if without the artificial head in the mix. You’d think by now we could determine if our headphone’s response is perfectly flat, but no. Unlike a free-standing speaker, the headphone earpad acts as an acoustic coupler that creates a small chamber in which your ear and headphone driver sit. While the rest of your head and torso have a lesser effect on frequency response, your ear chambers have a significant effect on frequency response and audio perception.

The earpad and driver chamber has its own acoustics. Similar to any other room, there are reflections, cancellation and absorption that result in peaks and valleys in frequency response. Even changing the position of the headphone effects frequency response. When listening to speakers in a room, we’re actually listening to the speakers and the room together. The room’s geometry causes peaks and nulls, absorption, reflections and cancellations, just like your ear chamber. Fortunately the small size of the ear chamber reveals peaks and nulls only above 2 or 3kHz where we’re less sensitive to variations in frequency response, unless they’re large peaks or wide valleys. Our brain also ignores some variations specific to our personal ear geometry and, after all, we’re very accustomed to listening through them.

Peaks and valleys that are pushed to higher frequencies cause us to perceive sound as far less colored despite what it looks like on a frequency response graph. That makes it even harder to reliably measure or judge headphone frequency response above 2kHz. If the response is taken on an artificial head at a specific position, it’s going to look different wearing the same headphone if measuring at your eardrum! That’s not to say that frequency response doesn’t provide useful information above 2kHz; there are a few clues to the sound. We do multiple measurements by adjusting the headphone position relative to the artificial head and use multiple heads looking for common features like large peaks or wide valleys. Just as room design and treatment can vastly improve free-standing speaker sound, we analyze the measurements and make adjustments to the geometry of our earpads, their material and angle, as well as driver design. We go through a series of critical listening sessions with physical modeling and analysis.

At Audeze our goal is producing natural sounding, accurate headphones. Our target frequency response is one that’s flat until about 2kHz and gently slopes down from there. We believe the gentle slope mimics the high-frequency energy from a good floor-standing speaker and compensates for the close proximity of the drivers resulting in a more natural-sounding treble.

Below are measurements of LCD-4 made at DRP (ear-drum reference point) using different artificial heads by both Audeze and third party. As expected, there is quite a bit of variability, which is not surprising due to the difference in ear geometry and how the artificial ears model the inner ear.

Below are the same measurements but we have superimposed what an ideal measurement may look like (in red)

The ideal measurement would look different from person to person, hence these measurements can only be used to get a rough idea of the frequency response and are not a substitute for listening. More over, the human brain compensates for the ear geometry and what one hears when the frequency response follows the ideal response would be a slight slanting but flat response going from the low frequencies to the high frequencies.

The frequency response, though very important, does not completely define a headphone’s sonic characteristics. Frequency response only tells us how a specific frequency is reproduced; boosted, attenuated or remains the same. It doesn’t tell us how these frequencies arrive in time, called time alignment. Audeze measures impulse response to judge time alignment as well.


The Importance of Impulse Response

Impulse response helps us judge the transient reproduction response capabilities of the headphone, one of the more important characteristics that create the sense of transparency and focused imaging. Two sources with the same frequency response might have dramatically different time alignment and so they’ll sound very different. For example, recording a shotgun and calculating its frequency response results in a fairly uniform curve over a wide range of frequencies, but so does recording a white noise! With a gunshot all frequencies are generated within a very short timeframe while white noise arrives over a longer span.

A headphone with perfect impulse response reproduces sound exactly as recorded. If you place a perfect microphone next to a headphone transducer with perfect impulse response you couldn’t can’t distinguish the difference between them.

Unfortunately this perfect model is, in reality, impossible to achieve -- blame the laws of physics and especially inertia. Audeze’s diaphragms are incredibly thin, light, extremely fast with uniform drive across the entire surface, all in an effort to do the impossible and reach the that ideal impulse response. The nano-grade diaphragms of the LCD-4, each of which takes weeks to manufacture, are beginning to blur the line between accepted technology and Audeze’s pushing the boundaries of what’s possible.

Unlike frequency response, impulse response measurements are even harder to interpret. A perfect impulse response has a single point that rises from zero to peak then falls back to zero with minimal oscillation. The faster the rise-time and the faster it settles, the more transparent the sound.

Return to Technology Page