In a typical consumer noise-cancelling headset, the anti-noise signal has only 300 to 400 µs to get from the outer microphone, through the electronics and out of the speaker before the noise itself reaches the ear (Kwong et al., Scientific Reports 2024). That window is the hard physics behind every ANC product. A research controller sampling at 768 kHz now spends just 4.7 µs on its ADC, DSP and DAC path (Hua and Pang, Applied Sciences 2024), but ANC processing latency in general can range from tens to thousands of microseconds (Iotov et al., ICASSP 2024), and every extra sample of delay narrows the band that can be cancelled. We aggregated data from Scientific Reports, Applied Acoustics, IEEE ICASSP, Hearing Research, Trends in Hearing, Acta Acustica, Audiology Research, Frontiers in Neuroscience, the ICSV congress, and vendor documents from Analog Devices, Qualcomm, Apple and Dolphin Design (labeled as vendor claims). For the decibels that finished earbuds deliver, see our true wireless earbuds ANC statistics; for the much slower world of AI noise removal on the microphone side, see our neural noise suppression latency statistics.
TL;DR
- 300 to 400 µs: typical acoustic travel time from outer microphone to ear in consumer headsets, the full budget for feedforward ANC (Kwong et al., Scientific Reports 2024).
- 4.7 µs from ADC to DSP to DAC on a 768 kHz ANC controller (Hua and Pang, Applied Sciences 2024).
- 12.9 µs at 192 kHz vs 5 µs at 768 kHz: analog-to-analog group delay of an ANC codec (Analog Devices ADAU1787 datasheet, vendor claim).
- ANC processing latency can range from tens to thousands of microseconds (Iotov et al., ICASSP 2024).
- One sample of excess delay: about 20 dB peak reduction over about 2,500 Hz; two samples: about 15 dB over about 1,500 Hz (Zhang and Qiu, Applied Acoustics 2014).
- Optimized hybrid ANC: more than 20 dB from 100 to 400 Hz, 5 to 15 dB from 500 to 800 Hz (Wu et al., ICSV 2016).
- About 100 µs hear-through delay on an ANC earphone vs mostly 3.5 to 10 ms on hearing aids (Denk et al., Trends in Hearing 2024).
- Short delays (0 or 0.5 ms) preferred 62% of the time over 2 to 10 ms (Súsonnudóttir et al., Trends in Hearing 2024).
- Hear-through boost of up to 14.37 dB near 2.5 kHz on AirPods Pro 2 (Harrison et al., Acta Acustica 2025).
- Elevation localization error 6.79° open ear vs 19.4° in AirPods Pro transparency mode (Ausili et al., Audiology Research 2025).
1. The Latency Budget: Acoustic Head Start vs Electronic Delay
Feedforward ANC is a race. The outer (reference) microphone hears the noise first, and the anti-noise must leave the speaker in time to meet the same sound wave at the eardrum. In typical consumer headsets the noise needs only 300 to 400 µs to make that trip, according to Kwong et al. in Scientific Reports, and their own headset measured 320 µs, which the authors set as the ceiling for total processing time. Group delay, the time a signal takes to pass through the converters and filters, has to fit inside that number with room to spare.
The good news is that modern high-sample-rate converters spend very little of the budget. A 4.7 µs path would use about 1.5% of a 320 µs window (4.7 / 320 = 0.0147), although the two figures come from different devices. The bad news is that not every design runs at those rates: published estimates of ANC processing latency stretch into the thousands of microseconds (see the table), and the slow end of that range is several times longer than the acoustic head start.
| Metric | Value | Source |
|---|---|---|
| Acoustic travel time from outer microphone to ear, typical consumer headsets | 300 to 400 µs | Kwong et al., Scientific Reports 2024 |
| Same measure on the study’s own headset, and the resulting limit for total processing time | 320 µs; processing must stay below 320 µs | Kwong et al., Scientific Reports 2024 |
| Acoustic delay through a circumaural earmuff, anechoic test | About 0.29 ms | Zhang and Qiu, Applied Acoustics 2014 |
| Primary path delay (noise source to ear) with the source in front vs at the side, and secondary path delay (speaker to error microphone, including converters), at 16 kHz sampling | 7 samples vs 5 samples; about 6 samples | Zhang and Qiu, Applied Acoustics 2014 |
| Range of ANC processing latency in practice | From tens to thousands of microseconds | Iotov et al., ICASSP 2024 |
| ADC to DSP to DAC latency, research feedback ANC controller sampling at 768 kHz | 4.7 µs | Hua and Pang, Applied Sciences 2024 |
| Analog in to analog out group delay of an ANC codec at 192, 384 and 768 kHz, DSP running zero instructions | 12.9 µs, 7.5 µs and 5 µs (vendor claim) | Analog Devices ADAU1787 datasheet Rev. A, 2020 |
| Analog recording path latency, and total input to output latency, of an ANC-oriented TWS codec | Down to 3 µs; 6 to 8 µs (vendor claim) | Dolphin Design, TWS codec white paper Part 3, 2021 |
Context note: at 16 kHz one sample lasts 62.5 µs (1 / 16,000), so 7 samples is about 437.5 µs and 6 samples about 375 µs. With 7 samples of acoustic head start against about 6 samples of electronic delay, the 2014 headset was causal by a margin of only about one sample with the source in front, and with the source at the side (5 samples) the margin disappeared. Most recent available data for the path-delay breakdown: the 2014 Applied Acoustics causality study. The Analog Devices datasheet figures are the vendor’s own measurements with the fast DSP bypassed, so a real ANC filter adds to them. For comparison, a computer audio interface works in milliseconds, not microseconds, as shown in our DAW buffer size and latency statistics.
2. How Delay Caps the Cancellation Bandwidth
Late anti-noise can still cancel a sound that changes slowly, but not one that changes quickly. That is the whole mechanism behind the bandwidth limit: each sample of delay shrinks the range of frequencies the controller can predict well enough to cancel. In a pure-delay model, a single sample of excess delay at 16 kHz still allowed about 20 dB of peak reduction across roughly 2,500 Hz; a second sample cut that to about 15 dB across roughly 1,500 Hz (Zhang and Qiu, Applied Acoustics 2014).
Noise direction changes the delay budget too. When the source moved from the front to the side of the head, the acoustic head start shrank by two samples and the measured overall reduction fell by about 6.6 dB (12.2 - 5.6). Speech is the hardest case: with conventional adaptive filters, cancellation of voiced speech collapses as delay grows, which is why the ICASSP 2024 work models speech harmonics to predict ahead. The same family of adaptive filters runs in conferencing echo cancellers, covered in our acoustic echo cancellation statistics.
| Metric | Value | Source |
|---|---|---|
| Pure-delay model, anti-noise 1 sample late (16 kHz): peak reduction, bandwidth, overall reduction | About 20 dB; about 2,500 Hz; about 7.2 dB | Zhang and Qiu, Applied Acoustics 2014 |
| Same model, anti-noise 2 samples late | About 15 dB; about 1,500 Hz; about 4.6 dB | Zhang and Qiu, Applied Acoustics 2014 |
| Measured feedforward headset, anechoic room, source in front vs at the side: peak and overall reduction | About 20 dB and 12.2 dB (below 1,500 Hz) vs about 12 dB and 5.6 dB (below 1,000 Hz) | Zhang and Qiu, Applied Acoustics 2014 |
| Mean attenuation of voiced speech, conventional adaptive filter (FXNLMS), no delay, and how it falls with delay at 48 kHz | About 35 dB; halved at 5 samples, cut to a third at 15 samples | Iotov et al., ICASSP 2024 |
| Upper frequency where that conventional filter still attenuates speech | About 1 kHz; it amplifies speech from 2.5 to 3.5 kHz | Iotov et al., ICASSP 2024 |
| Harmonic-model ANC (HMD-ANC) gain over the conventional filter at delays of 9 to 60 samples, and its effective range at 20 samples | 3.0 to 7.3 dB more; up to about 2.5 kHz | Iotov et al., ICASSP 2024 |
| Simulation, babble noise, linear paths: fully causal neural model vs Wiener filter vs time-domain FxLMS (2,048 taps) | -34.48 dBA vs -30.79 dBA vs -7.43 dBA | WaveNet-Volterra Neural Networks for ANC, arXiv preprint 2025 |
Context note: at 48 kHz, 20 samples equals about 0.42 ms (20 / 48,000 = 0.000417 s). The ICASSP and arXiv results are simulations, not headphone measurements; the arXiv preprint also gives its traditional filters the exact secondary path, which flatters them. The free-field causality measurements date from 2014 (most recent available data for a measured causality study of this kind).
3. Feedforward vs Feedback vs Hybrid Control Loops
The three architectures split the spectrum. Feedforward uses an outer microphone and wins on bandwidth when the noise arrives in time; feedback uses a microphone inside the cup, needs no head start, but is limited by its loop delay and by the waterbed effect (the rule that pushing noise down in one band pushes it up in another). An optimized hybrid on one headphone reached more than 20 dB of reduction from 100 to 400 Hz and 5 to 15 dB from 500 to 800 Hz, the best of the three structures tested (Wu, Qiu, Ma and Guo, ICSV 2016).
The feedback branch is where fit and wearer variation bite. The secondary path, the acoustic route from speaker to inner microphone, changes with head shape, hair and glasses, so a controller tuned on one dummy head can lose its stability margin on a real wearer. The Hua and Pang 2024 design shows the trade-off can be managed: constraining the filter to tolerate path variation kept the waterbed lift within 5 dB. Section 2 explains why feedforward loses its edge when the noise source moves.
| Metric | Value | Source |
|---|---|---|
| Hybrid ANC on one headphone with optimized parameters: reduction from 100 to 400 Hz and from 500 to 800 Hz | More than 20 dB; 5 to 15 dB | Wu, Qiu, Ma and Guo, ICSV 2016 |
| Fully digital decoupled hybrid ANC on a dummy head: reduction from the feedforward part, noise from 0 to 1,000 Hz | Around 15 to 25 dB, vanishing near 50 Hz | Rivera Benois, Nowak and Zölzer, 2017 |
| Extra reduction from the feedback (internal model control) branch in one sub-band | Around 1 to 3 dB | Rivera Benois, Nowak and Zölzer, 2017 |
| Waterbed lift (noise boost outside the cancelled band) of a feedback controller designed with a secondary-path uncertainty constraint | Within 5 dB across all measured secondary paths (simulation) | Hua and Pang, Applied Sciences 2024 |
| Tunable ANC circuit: band handled by ANC vs band left to passive cup attenuation | Below 2 kHz vs above 2 kHz | Kwong et al., Scientific Reports 2024 |
| Effective band of feedforward ANC vs feedback ANC vs passive isolation in TWS earbuds | About 80 Hz to 2 kHz vs about 50 Hz to 800 Hz vs above 1 kHz (vendor claim) | Dolphin Design, TWS codec white paper Part 3, 2021 |
| Earbud chip with dedicated ANC and transparency cores: ANC generation and DSP clocks | 4th Generation Adaptive ANC; 1x 350 MHz plus 1x 150 MHz DSP (vendor claim) | Qualcomm Snapdragon S5 Gen 1 Sound Platform brief |
Context note: the ICSV 2016 paper is closed access and the figures come from its published abstract; the Rivera Benois et al. 2017 paper reads its dB values from a figure. Both are most recent available data for side-by-side structure comparisons. The Dolphin Design article and the Qualcomm platform brief are vendor documents; Qualcomm publishes no decibel figure for its ANC.
4. Measured Attenuation Outside the Simulator
Simulations show what a control loop could do; measurements show what survives real noise, real ears and real levels. Two over-ear ANC headphones reduced background white and pure-tone noise by up to 20 dB, with the largest cuts at low frequencies, across ten audiometric frequencies from 125 to 8,000 Hz (Byun et al., Hearing Research 2026). That ceiling is consistent with the roughly 20 dB peaks in the delay-limited lab headset in section 2.
Level matters as much as frequency. In a hearing-screening study of three ANC setups, how faithfully a test tone came through fell steeply as background noise rose, and at 90 dB SPL the pass rate at 0.5 to 1 kHz was 0% for all three setups (Cheng et al., Sensors 2022). The turboprop study below shows the practical upside: in cabin noise dominated by a tone around 100 Hz, exactly the band ANC handles best, ANC headphones were the only option whose comfort gain over no protection survived statistical correction.
| Metric | Value | Source |
|---|---|---|
| Reduction of background white and pure-tone noise by ANC, 2 over-ear models, 125 to 8,000 Hz | Up to 20 dB, larger at low frequencies | Byun et al., Hearing Research 2026 |
| ANC output stability under steady white noise vs dynamic white noise | Under 1 dB SPL deviation vs under 5 dB fluctuation | Byun et al., Hearing Research 2026 |
| Correlation of a test tone heard through ANC with the same tone in quiet, averaged over 3 ANC setups: background at 50 dB SPL vs 90 dB SPL | 83% vs 35% | Cheng et al., Sensors 2022 |
| Hearing screening pass rate at 0.5 to 1 kHz with 90 dB SPL background, all three ANC setups | 0% | Cheng et al., Sensors 2022 |
| Turboprop cabin noise reproduced in the lab, and its dominant tone | 84.2 to 86.3 dB(A); tonal component around 100 Hz | Vledder et al., Work 2026 |
| Comfort with ANC headphones vs no protection in turboprop noise (24 participants) | Significantly better, p < .001, still significant after Bonferroni correction | Vledder et al., Work 2026 |
| Comfort with foam earplugs vs no protection in the same noise | Not significant, p = .102 | Vledder et al., Work 2026 |
Context note: the Vledder et al. study used Bose NC 700 headphones and reports comfort, not measured attenuation; ANC headphones also raised local discomfort around the ear over time. The Cheng et al. setups were Apple AirPods Pro, Bose QuietComfort 35 and EarPods under a QC35. Lab results will not match every ear and fit.
5. Hear-Through Latency: Microseconds vs Milliseconds
Transparency (hear-through) mode runs the control problem in reverse: the outer microphones feed ambient sound to the speaker so the wearer can hear the room. Because some outside sound still leaks past the ear tip, any processing delay makes the leaked and the reproduced sound add up out of step, which carves regular notches into the spectrum, a comb-filter effect. An ANC insert earphone in hear-through mode delayed sound by about 100 µs, while high-end hearing aids set to transparency mostly took 3.5 to 10 ms (Denk et al., Trends in Hearing 2024), and only the earphone showed no significant speech-in-noise penalty.
Listener studies explain why the microsecond class matters. People prefer the shortest delays, and longer delays make one’s own voice sound metallic and sharp. In a simulated one-sided open fit, raising delay from 1.75 to 7 ms worsened the speech reception threshold (the signal-to-noise ratio at which half the speech is understood) by 3.4 dB (-14.7 - (-18.1)), although with both ears fitted the effect of delay was not significant (p = 0.303) (Roth et al., Frontiers in Neuroscience 2023).
| Metric | Value | Source |
|---|---|---|
| Processing delay of an ANC insert earphone in hear-through mode | About 100 µs, with a peak near 1 ms around 1.5 kHz | Denk et al., Trends in Hearing 2024 |
| Processing delay of six high-end hearing aids set to transparency, and of a research platform | Mostly 3.5 to 10 ms (one model about 1 ms at 500 Hz and below); 10 ms | Denk et al., Trends in Hearing 2024 |
| Speech-in-noise disadvantage vs the open ear (18 normal-hearing listeners) | About 0.5 to 1 dB for hearing aids; not significant for the ANC earphone | Denk et al., Trends in Hearing 2024 |
| Processing delays reported for hearing devices, and left-right delay mismatch | 0 to 9.7 ms; mismatches up to 9.5 ms | Ausili et al., Audiology Research 2025 |
| Share of choices for short delays (0 or 0.5 ms) vs longer delays (2, 5 or 10 ms), 29 listeners | 62% vs 38% | Súsonnudóttir et al., Trends in Hearing 2024 |
| Own-voice rating on the tonal (metallic, sharp) scale, 0.5 ms vs 10 ms delay, 24 hearing-aid wearers | Median 1.00 vs 7.50, p < .001 | Súsonnudóttir et al., Trends in Hearing 2026 |
| Speech reception threshold in a simulated one-sided open fit, 1.75 ms vs 7 ms delay | -18.1 vs -14.7 dB SNR (higher is worse) | Roth et al., Frontiers in Neuroscience 2023 |
| Processing rate of an earbud hearing protection feature, and the sustained level it is not intended for | 48,000 times per second; above 110 dBA (vendor claim) | Apple AirPods Pro 3 product page, 2026 |
Context note: the Súsonnudóttir et al. 2024 study found no difference between 0 and 0.5 ms or among 2, 5 and 10 ms; the split is short vs longer. In the 2026 follow-up, the overall preference for 0.5 ms was not significant after Bonferroni-Holm correction, while the own-voice tonal difference was. Hearing aids are not ANC earbuds, but they are the best-documented case of how delay sounds when mixed with leaked direct sound. This section is factual only and is not medical advice.
6. What Hear-Through Mode Does to the Sound at the Ear
Low latency does not make a transparency mode transparent. In the hear-through test of three current earbuds, the strongest setting boosted sound by 14.37 dB near 2.5 kHz on AirPods Pro (2nd generation), 13.77 dB near 2 kHz on Bose QuietComfort Earbuds II and 12.28 dB at 800 Hz on Sony LinkBuds S, while cutting the top octave by 20 to 30 dB (Harrison et al., Acta Acustica 2025). The authors concluded none of the modes provided full acoustic transparency, and suggest the speech-band boost may be intended for media listening.
Spatial hearing takes the hit. Timing cues between the ears survived almost intact, which matches the microsecond-class delays in section 5, but level cues and the high-frequency spectral cues the outer ear uses for elevation did not (Ausili et al., Audiology Research 2025). The open-ear route avoids this by never sealing the canal, as covered in our open-ear dipole cancellation statistics.
| Metric | Value | Source |
|---|---|---|
| Peak hear-through boost, strongest setting: AirPods Pro 2nd gen / Bose QC Earbuds II / Sony LinkBuds S | 14.37 dB near 2.5 kHz / 13.77 dB near 2 kHz / 12.28 dB at 800 Hz | Harrison et al., Acta Acustica 2025 |
| Peak hear-through boost, weakest boosting setting, same three earbuds | 8.54 dB / 3.12 dB / 0.24 dB at 800 Hz | Harrison et al., Acta Acustica 2025 |
| High-frequency deviation from transparency, strongest hear-through setting | About -20 dB at 16 to 20 kHz (Apple); -28.24 dB at 20 kHz (Bose); about -30 dB at 16 to 20 kHz (Sony) | Harrison et al., Acta Acustica 2025 |
| Time before the AirPods Pro 2 applied its hear-through spectral shaping vs the other two earbuds | About 2 s vs within milliseconds | Harrison et al., Acta Acustica 2025 |
| Interaural time difference error with AirPods Pro in transparency mode vs open ear, quiet | Less than 45 µs (open-ear range ±700 µs) | Ausili et al., Audiology Research 2025 |
| Largest interaural level difference increase in transparency mode | 8.0 dB | Ausili et al., Audiology Research 2025 |
| Mean elevation localization error, open ear vs transparency mode (10 adults, quiet) | 6.79° vs 19.4° (p < 0.001) | Ausili et al., Audiology Research 2025 |
| Characteristic spectral notches and peaks above 5 kHz with the earbuds in | Eliminated | Ausili et al., Audiology Research 2025 |
Context note: Harrison et al. measured one pair per model on an acoustic test fixture in 80 dB SPL diffuse pink noise, and say fixture results likely exaggerate attenuation compared with real listeners. The localization study tested 10 normal-hearing adults; its abstract gives the azimuth error with the device as 19.6°, a value the results section reports only for the noisiest condition, so we quote the elevation figures.
Summary: Noise-Cancelling Headphone DSP Latency by the Numbers
| Metric | Value | Source |
|---|---|---|
| Acoustic head start for feedforward ANC, typical consumer headset | 300 to 400 µs | Scientific Reports 2024 |
| Processing time limit on the study’s own headset | Below 320 µs | Scientific Reports 2024 |
| ADC to DSP to DAC latency at 768 kHz | 4.7 µs | Applied Sciences 2024 |
| ANC codec group delay, 192 kHz vs 768 kHz | 12.9 µs vs 5 µs (vendor claim) | Analog Devices ADAU1787 datasheet |
| Range of ANC processing latency in practice | Tens to thousands of µs | ICASSP 2024 |
| Pure-delay model, 1 vs 2 samples late at 16 kHz | About 20 dB over 2,500 Hz vs about 15 dB over 1,500 Hz | Applied Acoustics 2014 |
| Conventional adaptive filter on voiced speech, no delay | About 35 dB, halved at 5 samples of delay | ICASSP 2024 |
| Optimized hybrid ANC, 100 to 400 Hz | More than 20 dB | ICSV 2016 |
| Feedforward part of a digital hybrid, 0 to 1,000 Hz | Around 15 to 25 dB | Rivera Benois et al. 2017 |
| Waterbed lift with uncertainty-constrained feedback design | Within 5 dB | Applied Sciences 2024 |
| ANC reduction of background noise, 2 over-ear models | Up to 20 dB | Hearing Research 2026 |
| Screening pass rate at 0.5 to 1 kHz in 90 dB SPL noise | 0% | Sensors 2022 |
| Hear-through delay: ANC earphone vs hearing aids | About 100 µs vs mostly 3.5 to 10 ms | Trends in Hearing 2024 |
| Preference for 0 or 0.5 ms delay vs 2 to 10 ms | 62% vs 38% | Trends in Hearing 2024 |
| Speech reception threshold, 1.75 ms vs 7 ms delay (one-sided open fit) | -18.1 vs -14.7 dB SNR | Frontiers in Neuroscience 2023 |
| Peak hear-through boost, AirPods Pro 2nd gen | 14.37 dB near 2.5 kHz | Acta Acustica 2025 |
| Elevation error, open ear vs transparency mode | 6.79° vs 19.4° | Audiology Research 2025 |
Methodology and Sources
- Kwong, Choy, Chan and Mung, Tunable active noise control circuit topology for multiple-feature applications, Scientific Reports 2024.
- Zhang and Qiu, Causality study on a feedforward active noise control headset with different noise coming directions in free field, Applied Acoustics 2014.
- Iotov, Nørholm, McCutcheon and Christensen, Improving Speech Attenuation in Headphones using Harmonic Model Decomposition and Multiple-Frequency ANC, IEEE ICASSP 2024.
- Hua and Pang, Uncertainty Constraint on Headphone Secondary Path Function for Designing Cascade Biquad Feedback Controller, Applied Sciences 2024.
- WaveNet-Volterra Neural Networks for Active Noise Control: A Fully Causal Approach, arXiv preprint 2504.04450, 2025.
- Wu, Qiu, Ma and Guo, Comparison of feedforward, feedback and hybrid structures on an ANC headphone, ICSV 2016.
- Rivera Benois, Nowak and Zölzer, Evaluation of a Decoupled Feedforward-Feedback Hybrid Structure for Active Noise Control in a Multi-Source Environment, 2017.
- Byun, Kwak, Lee, Cho and Kong, Feasibility study on noise attenuation and stability of active noise cancelling headphones for mobile hearing screening, Hearing Research 2026.
- Cheng et al., Objective Signal Analysis for Investigating Feasibility of Active Noise Cancellation in Hearing Screening, Sensors 2022.
- Vledder, Yao, Song and Vink, Effectiveness of earplugs and noise cancelling headphones to improve turboprop acoustic comfort, Work 2026.
- Denk, Wiederschein, Kemper and Husstedt, (Why) Do Transparent Hearing Devices Impair Speech Perception in Collocated Noise?, Trends in Hearing 2024.
- Súsonnudóttir, Kowalewski, Stiefenhofer and Neher, Individual Differences Underlying Preference for Processing Delay in Open-Fit Hearing Aids, Trends in Hearing 2024 and Súsonnudóttir et al., Own-Voice Perception with Different Processing Delays in Open-Fit Hearing Aids, Trends in Hearing 2026.
- Roth, Müller, Angermeier, Hemmert and Zirn, Effect of a processing delay between direct and delayed sound in simulated open fit hearing aids on speech intelligibility in noise, Frontiers in Neuroscience 2023.
- Harrison, Archer-Boyd and Murphy, Measurement of insertion loss and gain in variable hear-through and noise cancellation modes of true wireless earbuds, Acta Acustica 2025.
- Ausili, Erthal, Bennett and Snapp, Sound Localization with Hearables in Transparency Mode, Audiology Research 2025.
- Vendor documents (labeled as vendor claims throughout): Analog Devices ADAU1787 datasheet Rev. A, Dolphin Design, Paving the way for the next generation audio codec for TWS applications, Part 3, March 29, 2021, Qualcomm Snapdragon S5 Gen 1 Sound Platform brief, Apple AirPods Pro product page.
- Data watch: (1) Peer-reviewed measurements of the internal latency of shipping ANC earbuds are almost absent from open literature; manufacturers do not publish them, and the only commercial hear-through delay we found measured is the single ANC earphone in Denk et al. 2024 (model not named in the data we read). The 2025 localization study describes the AirPods Pro transparency mode as having minimal processing latency but gives no figure, and Apple publishes none. (2) Several core physics results are older than three years (Zhang and Qiu 2014, Wu et al. 2016, Rivera Benois et al. 2017, Dolphin Design 2021); they are flagged as most recent available data. (3) The 1.5% share of the delay budget combines a controller latency from one study with an acoustic delay from a different headset in another; it illustrates scale, not one device. (4) Hua and Pang report a 4.7 µs latency on an ADAU1777 at 768 kHz, close to but not identical with the 5 µs that Analog Devices claims for the ADAU1787 at the same rate. (5) The localization study’s abstract and results differ on the quiet-condition azimuth error (19.6°), so we report elevation only. (6) The 2024 delay-preference study reports the spectral ripple discrimination effect as η² = 0.30 in its results and 0.29 in its abstract; the 2026 abstract’s preference claim is stronger than its corrected statistics. (7) We found no open-access AES paper with usable latency numbers for this topic; AES library papers were not used. (8) Simulation results (ICASSP 2024, arXiv 2025, Hua and Pang waterbed figure) are not headphone measurements.
Last updated: October 27, 2026. We update this roundup quarterly, and the next refresh is expected when the IEEE ICASSP 2027 proceedings are published.