The Headphone Spec Audiophiles Trust Most Hides a 100-to-1 Audibility Gap, as Exposed by Decades of Distortion Research

From inflated bass graphs to controlled listening tests, the metric failed at every level.

Headphone reviews often treat total harmonic distortion as one of the cleanest ways to judge a driver. But research by Steve Temme, president of audio measurement firm Listen Inc., and Sean Olive, formerly Harman’s director of acoustic research, show why that trust can be misplaced.

Their listening tests found that trained listeners usually needed playback around 108 to 110 dB SPL before distortion became detectable. Many people listen closer to 70 to 85 dBA. So the number can look serious long before the flaw becomes audible.

This problem goes deeper than volume. THD can exaggerate bass distortion, downplay harmonics the ear catches sooner, and miss artifacts listeners actually notice.

The Number Everyone Cites Is Broken

Total harmonic distortion is overwhelmingly shaped by the second and third harmonics, which are the lowest-order distortion products closest in frequency to the original note.

On the other hand, higher-order harmonics, such as the 7th or 10th, barely move the result because the THD formula squares each component and causes smaller values to vanish.

These nearly invisible harmonics happen to be the ones human hearing actually detects, while the dominant second and third are the ones the ear naturally suppresses.

In short, THD is dominated by low-order harmonics the ear tends to mask, while higher-order harmonics that can be more audible barely move the number. So THD can be very misleading.

Why every THD graph lies at 20 Hz

The distortion graph gets even more misleading at low frequencies, as conventional THD references every harmonic back to the fundamental at the stimulus frequency.

At 20 Hz, the main tone is already weakened because headphones roll off in deep bass. Its distortion products, however, appear at higher frequencies like 40 and 60 Hz, where the headphone may produce more output.

But conventional THD still compares those 40 and 60 Hz distortion products against the weaker 20 Hz tone. And because the reference point is already low, the percentage looks much larger than it really is.

This means that while a driver has not suddenly become far more distorted in the bass, the formula can make the bass distortion look worse than it is.

For example, a distortion product that should read as 10% distortion can show up as 71% at 20 Hz under the conventional method. At higher frequencies, the same underlying distortion can appear as only 3.1%.

Temme traces that convention to the Bruel & Kjaer chart-recorder era, when tracking filters were offset but the chart still plotted harmonics at the stimulus frequency. The workflow became standard, and the distorted-looking bass rise became part of how the industry learned to read THD.

Then in a 1993 AES paper, he proposed a fix while still at Bruel & Kjaer, then built it into SoundCheck after founding Listen Inc. two years later. His frequency-normalized method compares harmonics to the fundamental at their own measured frequencies, not the stimulus.

The correction flattens the distortion curve and exposes what the conventional method hides. Three decades later, the industry is still catching up.

Your Ears Were Already Ignoring It

Every loud tone the ear processes suppresses the frequencies around it.

Inside the cochlea, the ear groups sound into small frequency zones called critical bands. But while a loud tone can mask quieter sounds inside the same zone, those quieter sounds may still exist in the signal, and measurement gear may still capture them, even though the listener may not hear them.

Audiologists call this auditory masking.

Low-order harmonics like the second and third sit close to the fundamental in frequency and fall squarely within that curve. Meanwhile, higher-order harmonics sit further away and escape the masking, becoming audible at far lower levels.

That is a 100-to-1 gap between the harmonic that fills THD spreadsheets and the one the ear actually catches.

Where distortion vanishes entirely

At lower frequencies the masking effect becomes more extreme because the suppression curve broadens dramatically. But at 50 Hz, the curve can be so wide that, in some cases, it swallows not just the second and third harmonics but also nearly the entire distortion spectrum.

“You can actually have more than 100% distortion and not hear it in some cases,” Temme noted.

That claim sounds absurd until you consider what it means physically. When a fundamental is strong and low enough, the masking curve extends so far that distortion products matching or exceeding the fundamental’s own power remain perceptually silent.

This helps explain why many bass THD graphs look alarming even when the resulting distortion may be inaudible.

But loud playback makes the problem even stranger. Temme cautioned that subjective distortion tests should not be run too loud, not only because of hearing-risk concerns, but because louder playback widens masking curves.

That widening means the audiophile habit of cranking volume to expose flaws can bury more distortion under a broader curve. The louder you listen, the more your ear hides, which means the quieter, more careful sessions you trust least may be the ones where distortion would be easiest to detect.

ANC Makes It Better, Not Worse

ANC exposes the central problem with THD from the opposite direction. In Temme’s measurements, noise cancellation sometimes reduced harmonic distortion even while perceived sound quality got worse.

The spec improved, but the listening experience did not.

He had the same instinct most audiophiles do. After years of listening to ANC headphones and noticing the sound quality degrade when noise cancellation kicked in, he assumed harmonic distortion was the culprit. Besides, ANC circuits process the audio signal in real time, so introducing measurable artifacts seemed like the obvious explanation.

“I assumed it was because of distortion and traditional distortion,” he recalled. “Boy, was I wrong.”

When he engaged ANC on one pair of high-end over-ear headphones, bass THD dropped from 8% to 3% at 20 Hz, cutting the distortion by more than half. A second pair already sat at just 0.25% distortion at 90 dB SPL and barely shifted with ANC on.

The feedback circuits were not adding harmonic distortion. In fact, they were reducing it in the first headphone.

Apparently, because ANC works by generating an anti-phase signal to cancel unwanted sound, the same feedback loop can also suppress some harmonic products inside its operating range.

This left a more important question: If ANC made THD look better, why did the headphones still sound worse with noise cancellation engaged?

To find the missing damage, Temme ran the recordings through Perceptual Objective Listening Quality Analysis (POLQA), a perceptual quality metric built to catch signal changes THD can miss. The results moved in the opposite direction from the distortion graph, as the quality fell when ANC was engaged, even as harmonic distortion dropped.

So the likely problem was timing, not harmonic distortion. ANC systems constantly adapt as they try to cancel outside noise. And fast transients may expose moments when the circuit cannot keep up, creating artifacts that listeners hear but THD does not register.

That means THD doesn’t just get some numbers wrong. It can miss the kind of damage people actually notice.

The Proof and Its Limits

If THD misses masked harmonics and fails to explain time-domain artifacts, the next question is whether listeners can hear the distortion it measures at all.

While Temme’s measurements built the theoretical case, Sean Olive set out to test whether it held up under controlled conditions with real listeners and real headphones.

This distortion study used binaural recordings of five headphones made through a B&K 5128 ear simulator, all equalized to an identical frequency response so that distortion was the only variable.

Olive designed the test to isolate distortion from everything else listeners might notice.

The method had three key controls:

• He compared five headphones across a wide quality range, from the low-distortion Audeze LCD-X to the budget Loxjie X6.
• He equalized all recordings to the same frequency response, so listeners were judging distortion rather than tonal balance.
• He played the recordings at 83 dBA and normalized them to -16 LUFS, so volume differences between tracks could not bias the results.

To find the threshold where listeners could detect distortion, Olive then used an ABX adaptive test staircase method.

Listeners started with an obviously distorted version, then the test gradually made the difference harder to hear until they could no longer identify it reliably.

The surprising result

Across the three best-behaved headphones, listeners did not reliably detect distortion until playback reached 108 to 110 dB SPL. Most individual thresholds fell between 105 and 115 dB SPL.

This surprised the researchers as this was not an untrained or unusually forgiving panel. Participants ranged from 29 to 51 years old, with a mean age of 34, and 71% had completed 30 hours of distortion-detection training over three months.

Plus, the 83 dBA playback level also made the test more sensitive than a louder real-world test would have been.

The unresolved Critical Red anomaly

Olive did not present these results as the final word, and the places where he hedged tell you as much as the data itself.

“The sample size is very small,” he acknowledged. “So when you’re doing regression analysis on six data points, it’s heavily influenced by outliers. That’s a big caveat.”

The Critical Red was the clearest example of this.

According to the research notes, it showed audible artifacts below 90 dB despite having fairly low THD, which made it difficult to explain through the metric the study was testing.

So, the anomaly remains unresolved.

This limitation matters because it keeps the article from overstating the case. The study strongly suggests that ordinary headphone THD is often far below audibility, but it does not prove that every audible defect will show up in, or be ruled out by, the THD number.

But even with those caveats, the study leaves the audiophile community’s reliance on THD looking much weaker than the spec sheets suggest.

The researchers who spent a decade measuring, modeling, and testing headphone distortion found that their own listeners usually couldn’t hear it until playback reached levels no one should be using.

Olive is the first to say more work is needed, but the gap between 70 dBA and roughly 110 dB is not the kind of gap that a few more headphones are likely to erase.

 

 

Article written by Colin Toh
Reprint from website: https://www.headphonesty.com