Taking an insert earphone out and putting it back in changed its output at 90 Hz by up to 15 dB across 8 fittings (median 7.4 dB), while units of one in-the-ear earpiece model differed mostly by a broadband offset of about ±2 dB. That gap is the core finding of this roundup: the headphone you bought is usually more consistent than the way it sits on your head. A person-to-person standard deviation of about 3 dB (against about 1 dB within one person) and fixture-versus-ear errors of ±15 dB or more above 4 kHz complete the picture. We aggregated data from Acta Acustica, Frontiers in Psychology, Audio Engineering Society convention papers, the Hearpiece research database, RTINGS.com’s independent test lab, and vendor documents labeled as such. If you want to know which response people prefer, see our Harman target curve preference statistics; this page covers how far one model’s response actually wanders.
TL;DR
- An insert earphone varied by up to 15 dB at 90 Hz across 8 fittings, median 7.4 dB (Kohnen et al., Acta Acustica 2021).
- Units of one in-the-ear earpiece model differed by a broadband offset of about ±2 dB (Denk and Kollmeier, Hearpiece database 2020).
- Between-listener standard deviation was about 3 dB versus about 1 dB within one listener, 0.25 to 4 kHz (Denk et al., Frontiers in Psychology 2021).
- Standard ear simulators missed real ear canals by ±15 dB or more above 4 kHz (Bezzola, AES Convention 2023).
- A dummy head and a human head differed by up to 12 to 18 dB around 10 to 15 kHz (Wersényi, Forum Acusticum 2023).
- Etymotic says ER4SR channels match within 1 dB from 100 Hz to 10 kHz (Etymotic product page, vendor’s own claim).
- RTINGS.com rates a weighted left/right amplitude mismatch below 0.6 as good (RTINGS.com Test Bench 2.0, 2025).
- Across 400 different models, the 95% interval of frequency response spanned ±22 dB (Sonarworks white paper 2022, vendor’s own data).
- Four calibrated audiometers with the same headphone model varied by up to 21 dB for one tone (Barlow et al., Noise and Health 2014).
- One AirPods Pro 3 earbud varied by less than 1 dB across four placements over three days (Kim and Thibodeau, Audiology Research 2026).
- The IEC 60268-7:2010 headphone standard was superseded on June 27, 2025 (NSAI standards store listing).
1. Same-Model Unit Variation: How Far Apart Are Two Samples?
Peer-reviewed data on sample-to-sample spread is thinner than forum debate suggests, and the best numbers come from research devices rather than retail headphones. The clearest one is reassuring: units of one in-the-ear earpiece differed mostly by a broadband sensitivity offset of about ±2 dB, meaning a level shift across the whole band rather than a different tonal shape (Denk and Kollmeier, Hearpiece database 2020). A level offset is the easiest kind of difference to correct with a volume control.
The between-model picture is a different order of magnitude. A calibration vendor reports a ±22 dB spread across 400 models, and a 2017 JASA analysis found that about 90% of the variation across measured headphones fits into six basis functions, with no link to price (Breebaart, JASA Express Letters 2017). The practical reading: choosing a model matters far more than winning the sample lottery, and the vendor’s own rule of averaging at least 5 units from different production batches shows that even calibration companies treat one unit as an unreliable stand-in for its model (Sonarworks SoundID SR white paper 2022).
| Metric | Value | Source |
|---|---|---|
| Difference between individual units of one in-the-ear earpiece model, measured on a KEMAR test head | Mostly a broadband offset of about ±2 dB | Denk and Kollmeier, Hearpiece database 2020 |
| Error when one model-average calibration is applied to an individual unit | Up to 3 dB | Sonarworks blog 2016 (vendor’s own claim) |
| Minimum units per model, taken from different production batches, for a model-average profile | 5 | Sonarworks SoundID SR white paper 2022 (vendor’s own claim) |
| 95% interval of frequency response across 400 different models (between models, not between units) | ±22 dB from 20 Hz to 20 kHz; ±13 dB from 100 Hz to 10 kHz | Sonarworks SoundID SR white paper 2022 (vendor’s own data) |
| Share of variation across measured headphone responses explained by six basis functions | About 90% | Breebaart, JASA Express Letters 2017 |
| Relationship between retail price and measured response or estimated quality | None found | Breebaart, JASA Express Letters 2017 |
| Spread of hearing-test results when calibration coefficients are reused on other sets of the same model (standard deviation versus clinical audiometry) | Below 8.5 dB | Masalski, Medical Devices: Evidence and Research 2024 |
| Same spread when each device is calibrated individually by a normal-hearing listener | 7.3 to 10.7 dB | Masalski, Medical Devices: Evidence and Research 2024 |
Context: the two Masalski rows measure hearing thresholds, not frequency response, but they answer a practical question. Reusing one model’s calibration on other units of that model (Masalski 2024) lands inside the error range of calibrating each device individually, which suggests unit spread is small compared with the noise of a hearing test. Most recent available data for the Sonarworks unit claim: Sonarworks blog, 2016.
2. Left/Right Channel Matching: The Mismatch Inside One Pair
Channel mismatch is the one form of unit variation a listener hears without comparing two products, because it shifts the stereo image. Etymotic states the ER4SR’s left and right channels match within 1 dB from 100 Hz to 10 kHz, and ships a channel-matching compliance certificate with each pair (Etymotic ER4SR product page). That is a vendor’s own claim, and it is unusual precisely because most makers publish no matching tolerance at all.
Independent scoring has moved from a pass/fail idea to graded thresholds. RTINGS.com’s Stereo Mismatch test, introduced with Test Bench 2.0 in April 2025, splits the problem into amplitude (level), frequency (shape) and phase (timing) differences between the two drivers (RTINGS.com Stereo Mismatch test 2025). Production test vendors frame it the same way: Klippel’s quality-control note compares each channel against reference units, and its worked example slightly widens the left/right limit above 2 kHz.
| Metric | Value | Source |
|---|---|---|
| Left/right channel matching specification for the ER4SR in-ear monitor, 100 Hz to 10 kHz | Within 1 dB | Etymotic ER4SR product page (vendor’s own claim) |
| Weighted amplitude mismatch (A-weighted level difference between left and right drivers) rated good | Below 0.6 | RTINGS.com Test Bench 2.0, Stereo Mismatch 2025 |
| Weighted frequency mismatch (RMS deviation between left and right frequency responses) rated good | Below 1.2 | RTINGS.com Test Bench 2.0, Stereo Mismatch 2025 |
| Weighted phase mismatch between left and right drivers rated good | Below 1 degree | RTINGS.com Test Bench 2.0, Stereo Mismatch 2025 |
| Smallest change in the amplitude or frequency mismatch score treated as noticeable | 0.1 | RTINGS.com Test Bench 2.0, Stereo Mismatch 2025 |
| Upper frequency limit of the left/right difference check in a production quality-control example (limit widened above 2 kHz) | Below 10 kHz | Klippel Application Note 73, revision 1.8, March 13, 2026 (vendor document) |
Note: the independent lab publishes thresholds, not a typical value per product category, so this table shows what “good” means rather than how many headphones reach it.
3. Fit and Repositioning on Real Heads: The Biggest Variable
This is where the decibels really move. In a two-lab study, 14 listeners each fitted three headphones 8 times, and the insert earphone varied by up to 15 dB at 90 Hz, with a median of 7.4 dB (Kohnen et al., Acta Acustica 2021). That bass swing comes from the seal: a slightly looser insertion leaks low frequencies, which is why tips and insertion depth matter so much (see our ear tip passive isolation statistics).
Open over-ears barely moved in the same test, and the closed over-ear sat in between, with its own bass sensitivity near 90 Hz. Cushion pressure and seal on the head are linked, which is one reason headphone clamping force statistics matter for consistency as well as comfort. Higher up the picture flips: earlier repositioning studies summarized by the same authors found about 1.5 dB of variation up to 5 kHz but up to 15 dB above 7 kHz, so treble differences between two reviews of the same model are often fit, not factory.
| Metric | Value | Source |
|---|---|---|
| Insert earphone (Etymotic ER4) at 90 Hz, spread between highest and lowest of 8 fittings | Up to 15 dB (median 7.4 dB) | Kohnen et al., Acta Acustica 2021 |
| Closed over-ear (Beyerdynamic DT770) near 90 Hz, same spread | Up to 4.5 dB (right-ear median 2.8 dB) | Kohnen et al., Acta Acustica 2021 |
| Open over-ear (Sennheiser HD650) from 90 to 2500 Hz, same spread | Less than 1.4 dB | Kohnen et al., Acta Acustica 2021 |
| Standard deviation across 8 repositionings within one listener, Sennheiser HD 650, 0.25 to 4 kHz | About 1 dB | Denk et al., Frontiers in Psychology 2021 |
| In-the-ear earpiece re-inserted 5 times, driver response at the eardrum below 8 kHz | Less than ±2 dB | Denk and Kollmeier, Hearpiece database 2020 |
| One AirPods Pro 3 earbud measured in four positions rotated 45 degrees, over three measurement days | Less than 1 dB variation | Kim and Thibodeau, Audiology Research 2026 |
Outlier note: the AirPods figure comes from a single right earbud in a hearing-aid test setup (Kim and Thibodeau 2026), so it shows placement repeatability for one device, not a population result.
4. Real Heads vs Test Fixtures, and Person-to-Person Spread
A test fixture (an artificial head with an ear simulator, the microphone-and-cavity assembly that stands in for an ear canal) gives one repeatable answer. Real listeners give a distribution. On 40 listeners wearing the same Sennheiser HD 650, the standard deviation was about 3 dB between people versus about 1 dB within one person from 0.25 to 4 kHz (Denk et al., Frontiers in Psychology 2021). In other words, the person changes the result about three times as much as re-seating the headphone does in that band.
In-ear products are worse because the ear canal itself becomes part of the acoustic system. Simulations on MRI scans of 10 people (20 ears, 5 insertion depths each, 100 personal responses) compared real canals with a standard head and torso simulator: the error was small in the bass but grew to ±15 dB or more above 4 kHz (Bezzola, AES Convention 2023). That is the physical argument behind personal fit and individual equalization explored in our custom 3D printed ear canal fit statistics.
| Metric | Value | Source |
|---|---|---|
| Standard deviation between 40 listeners, Sennheiser HD 650, 0.25 to 4 kHz | About 3 dB | Denk et al., Frontiers in Psychology 2021 |
| Difference between 25 subjects for an in-the-ear earpiece, 2 to 6 kHz | Less than ±5 dB, mostly a broadband offset | Denk and Kollmeier, Hearpiece database 2020 |
| Difference between subjects for the same earpiece above 6 kHz | 30 dB and more | Denk and Kollmeier, Hearpiece database 2020 |
| Error of a standard simulator (HATS with 711-type couplers) versus simulated real ear canals, below 1 kHz | Less than ±1 dB | Bezzola, AES Convention 2023 |
| Same error, 1 to 4 kHz | Within ±5 dB | Bezzola, AES Convention 2023 |
| Same error, above 4 kHz | ±15 dB or more | Bezzola, AES Convention 2023 |
| Supra-aural (on-ear) headphones on a dummy head versus a human head, around 10 to 15 kHz | Differences up to 12 to 18 dB | Wersényi, Forum Acusticum 2023 |
| Frequency above which dummy-head and human-head results diverge in general | Above 8 to 10 kHz | Wersényi, Forum Acusticum 2023 |
Context: the Bezzola figures assume no leakage, so real-world in-ear errors with an imperfect seal can only be larger in the bass. The Wersényi comparison covers open ear canal headsets and on-ear models (Wersényi, Forum Acusticum 2023).
5. Measurement Repeatability Across Rigs and Labs
Even when the headphone and the head stay the same, the measuring chain adds its own spread. When the same 14 people were measured at two university labs, high-frequency headphone responses deviated by up to 10 dB, which the authors say probably stem from differences in the measured microphone sensitivities (Kohnen et al., Acta Acustica 2021). A frequency response graph is therefore a statement about a headphone, a head and a microphone calibration at once.
Clinical audiology shows how large the combined error can get. Four recently calibrated audiometers, all with TDH39 headphones and with the headphones swapped for each test, produced up to 21 dB of variation for the same tone, worst at 6 kHz (Barlow et al., Noise and Health 2014). Review labs respond with repetition: RTINGS.com averages repositionings and human passes before scoring consistency (RTINGS.com Frequency Response Consistency test 2025), and production-test guidance asks for repeatability checks before any limit is set (Klippel Application Note 73, 2026).
| Metric | Value | Source |
|---|---|---|
| High-frequency deviation in headphone responses of the same 14 people measured at two labs | Up to 10 dB | Kohnen et al., Acta Acustica 2021 |
| Maximum sound pressure variation at the ear for the same tone across 4 calibrated audiometers with TDH39 headphones | 21 dB (highest at 6 kHz) | Barlow et al., Noise and Health 2014 |
| Test-retest deviation of a clinical audiometer used as a reference | About 6 dB | Masalski, Medical Devices: Evidence and Research 2024 |
| Independent lab consistency protocol for over-ear and on-ear headphones | 5 repositionings on a B&K 5128 test head, plus 5 people measured 3 times each (human passes scored from 20 Hz to 2 kHz); a score below 1 is rated good | RTINGS.com Test Bench 2.0, Frequency Response Consistency 2025 |
| Repeated measurements of one headphone in a production-test repeatability example | 6 | Klippel Application Note 73, revision 1.8, 2026 (vendor document) |
| Minimum signal-to-noise ratio recommended at every tested frequency in production headphone testing | At least 30 dB | Klippel Application Note 73, revision 1.8, 2026 (vendor document) |
| Status of the IEC 60268-7:2010 headphone measurement standard | Superseded on June 27, 2025 by IEC 60268-7:2025 | NSAI standards store listing, IEC 60268-7 |
| Edition the 2010 standard replaced | Second edition of 1996 | NSAI standards store listing, IEC 60268-7 |
Context: IEC 60268-7 specifies which characteristics manufacturers should state and how to measure them; it does not set a unit-to-unit tolerance, which is why vendor matching claims are not comparable with each other.
Summary: Headphone Unit-to-Unit Frequency Response Variation by the Numbers
| Metric | Value | Source |
|---|---|---|
| Insert earphone at 90 Hz across 8 fittings | Up to 15 dB (median 7.4 dB) | Kohnen et al., Acta Acustica 2021 |
| Difference between units of one in-the-ear earpiece model | About ±2 dB broadband offset | Denk and Kollmeier, Hearpiece database 2020 |
| Between-subject difference for that earpiece above 6 kHz | 30 dB and more | Denk and Kollmeier, Hearpiece database 2020 |
| Standard deviation between 40 listeners, HD 650, 0.25 to 4 kHz | About 3 dB | Denk et al., Frontiers in Psychology 2021 |
| Standard deviation within one listener across repositionings, HD 650, 0.25 to 4 kHz | About 1 dB | Denk et al., Frontiers in Psychology 2021 |
| Standard simulator versus real ear canals, 1 to 4 kHz | Within ±5 dB | Bezzola, AES Convention 2023 |
| Standard simulator versus real ear canals, above 4 kHz | ±15 dB or more | Bezzola, AES Convention 2023 |
| Dummy head versus human head, on-ear models, 10 to 15 kHz | Up to 12 to 18 dB | Wersényi, Forum Acusticum 2023 |
| One AirPods Pro 3 earbud across four placements and three days | Less than 1 dB | Kim and Thibodeau, Audiology Research 2026 |
| ER4SR left/right matching, 100 Hz to 10 kHz | Within 1 dB | Etymotic ER4SR product page (vendor’s own claim) |
| Weighted left/right amplitude mismatch rated good | Below 0.6 | RTINGS.com Test Bench 2.0, Stereo Mismatch 2025 |
| Spread across 400 different models, 20 Hz to 20 kHz | ±22 dB (95% interval) | Sonarworks SoundID SR white paper 2022 (vendor’s own data) |
| Variation across headphone responses captured by six basis functions | About 90% | Breebaart, JASA Express Letters 2017 |
| Hearing-test spread when one model’s calibration is reused on other units | Below 8.5 dB | Masalski, Medical Devices: Evidence and Research 2024 |
| Same-tone variation across 4 calibrated audiometers with TDH39 headphones | Up to 21 dB | Barlow et al., Noise and Health 2014 |
| IEC 60268-7:2010 superseded | June 27, 2025 | NSAI standards store listing, IEC 60268-7 |
Methodology and Sources
- Kohnen, Denk, Llorca-Bofí, Vorländer and Kollmeier, Cross-site investigation on head-related and headphone transfer functions, Acta Acustica 5, 58 (2021): 14 listeners, 2 labs, 3 headphones, 8 fittings each.
- Denk et al., The Missing 6 dB Revisited, Frontiers in Psychology (2021): 40 listeners, Sennheiser HD 650 from one purchase batch.
- Denk and Kollmeier, The Hearpiece database of individual transfer functions, arXiv 2004.06579 (2020): 25 subjects plus KEMAR, 5 re-insertions per ear.
- Bezzola, In-Ear Headphones on Ear Canal Simulator vs Real Human Ear Geometries, AES Convention, October 25, 2023 (Samsung Research America).
- Wersényi, Comparison of Transfer Functions of Open Ear Canal Headsets Measured on a Dummy-Head and a Human Head, Forum Acusticum 2023.
- Breebaart, No correlation between headphone frequency response and retail price, JASA Express Letters 141(6) EL526 (2017).
- Barlow, Davison, Ashmore and Weinman, Amplitude variation in calibrated audiometer systems, Noise and Health 16(72) (2014).
- Masalski, The Hearing Test App for Android Devices, Medical Devices: Evidence and Research (2024).
- Kim and Thibodeau, Electroacoustic Verification Comparison of AirPods Pro 2nd and 3rd Generations and Traditional Hearing Aids, Audiology Research 16(2) 55 (2026).
- RTINGS.com (independent lab): Frequency Response Consistency test, Test Bench 2.0, August 11, 2025 and Stereo Mismatch test, Test Bench 2.0, April 28, 2025.
- Vendor documents, labeled as vendor claims throughout: Sonarworks SoundID SR white paper (2022), Sonarworks blog on same-model variation (2016), Etymotic ER4SR product page and Klippel Application Note 73, QC Headphone Testing, revision 1.8 (2026).
- NSAI standards store listing for IEC 60268-7:2010, Sound system equipment, Part 7: Headphones and earphones.
- Data watch: independent, multi-unit studies of retail headphones are scarce, so the best unit-to-unit figure comes from a research earpiece measured on KEMAR (one poorly fitting vented unit fell outside the ±2 dB pattern), and the retail-unit figures are vendor claims. The Sonarworks ±22 dB interval describes spread between 400 different models, not between units; the same white paper also cites 875+ models and 7000+ unit measurements without reconciling the two counts. The Masalski rows measure hearing-test agreement, not frequency response, and the 6 dB audiometer figure is a value that paper cites from earlier work. The Barlow audiometer result mixes device, calibration and headphone placement error. Kohnen et al. and Denk et al. (2021), the Hearpiece database (2020), Breebaart (2017), the Sonarworks blog (2016) and Barlow et al. (2014) are older than three years: most recent available data. The Kim and Thibodeau result covers one earbud. Pay-walled AES papers on headphone leakage on real heads were not used because their figures could not be read in full.
Last updated: November 6, 2026. We update this roundup quarterly, and the next refresh is expected when the IEC 60268-7:2025 edition text and new Test Bench results from RTINGS.com become available for review.