Only 39% of electronica tracks in a scan of more than 107,000 tracks peak at or below 0 dBTP, so most of them carry peaks above full scale that a sample-peak meter cannot display (Archimago’s Musings, library scan 2025). The measurement side explains why: a sample-peak meter can miss a peak by 3 dB on a tone at one quarter of the sample rate, and even a standard 4x oversampling true-peak meter can under-read by up to 0.688 dB (ITU-R BS.1770-5, 2023). Downstream, the EBU tells broadcasters to limit at -2 dBTP before lossy encoders (EBU Tech 3344, 2016), while 8 of 9 disc players in a Hi Fi News test squashed or distorted a +3 dBFS inter-sample tone (Lesurf, Over The Top measured results 2012). We aggregated data from the ITU, the European Broadcasting Union, Spotify, Apple, Benchmark Media, Sound On Sound, independent measurement archives and published hardware tests. For how engineers are adopting automated limiting and mastering tools, see our AI mastering adoption figures; this roundup stays on the peaks those tools have to control.
TL;DR
- A sample-peak meter can under-read by 3 dB, and a 4x true-peak meter by up to 0.688 dB (ITU-R BS.1770-5, 2023).
- EBU Mode meters must read true-peak test signals within +0.2/-0.4 dB (EBU Tech 3341, 2023).
- 44% of general rock tracks in a 31,400-track sample peaked above +0.1 dBTP (Archimago’s Musings, library scan 2025).
- One Steely Dan CD track holds 1,129 inter-sample overs in a little over five minutes, peaking at +0.8 dBFS (Benchmark Media, Intersample Overs in CD Recordings 2017).
- Records measured roughly 5 dB louder than 1970s releases across a 4,500-track corpus (Sound On Sound, Dynamic Range and the Loudness War 2011).
- The EBU sets the limiter in front of lossy encoders at -2 dBTP (EBU Tech 3344, 2016).
- MPEG-1 Layer II and Dolby AC-3 delivery calls for a -2 dBTP ceiling instead of -1 dBTP (EBU Tech 3343, 2023).
- Spotify asks for true peaks below -2 dBTP on masters louder than -14 LUFS (Spotify for Artists, Loudness normalization).
- Apple recommends at least 1 dB of headroom because 4x playback oversampling can clip a 0 dBFS master (Apple Digital Masters, 2021).
- Only 1 of 9 disc players came close to reproducing a +3 dBFS inter-sample tone correctly (Lesurf, Over The Top measured results 2012).
1. How Far True Peaks Hide From Sample-Peak Meters
A true peak is the highest point of the continuous waveform that a converter rebuilds from the samples. Because that point often falls between two samples, the largest stored number can understate it. The ITU’s own worked example is blunt: a sample-peak meter can under-read a tone at one quarter of the sampling frequency by 3 dB, and real transients with strong high-frequency content are “commonly” under-read by several dB (ITU-R BS.1770-5, 2023).
The fix is oversampling, meaning the meter rebuilds extra points between the original samples before it looks for the maximum. The ITU-R BS.1770 recommendation, in force in its fifth revision since November 2023, specifies 4x oversampling at 48 kHz, which lifts the measurement rate to 192 kHz. That choice is a compromise, not perfection: for content up to half the sample rate, the worst-case error only drops below a tenth of a decibel from 12x oversampling upward (0.075 dB). This is why a limiter set exactly at 0 dBTP on a 4x meter is not a guarantee of zero overs, and why the EBU keeps a 1 dB margin in production.
| Metric | Value | Source |
|---|---|---|
| Worst-case under-read of a sample-peak meter on a tone at one quarter of the sample rate | 3 dB | ITU-R BS.1770-5 (2023) |
| Maximum under-read of a 4x oversampling true-peak meter, content up to 0.45 and up to 0.5 of the sample rate | 0.554 dB and 0.688 dB | ITU-R BS.1770-5 (2023) |
| Maximum under-read at 8x, 16x and 32x oversampling, content up to 0.5 of the sample rate | 0.169 dB, 0.042 dB, 0.010 dB | ITU-R BS.1770-5 (2023) |
| Potential under-read the EBU allows for when it sets the 1 dB production margin (4x meter at 48 kHz) | about 0.5 dB | EBU Tech 3343 (2023) |
| Accepted true-peak reading error for an EBU Mode meter on test signals 15 to 23 | +0.2/-0.4 dB | EBU Tech 3341 (2023) |
| Expected reading of EBU test signal 19 (sine at 1.41 times full scale, one quarter of the sample rate, 45 degree phase) | +3.0 dBTP | EBU Tech 3341 (2023) |
| True-peak measurement tolerance in EBU R 128 (signals limited to 20 kHz bandwidth) | ±0.3 dB | EBU R 128 (2023) |
Context note: the compliance set in EBU Tech 3341 contains nine dedicated true-peak test signals, all synthesized at 48 kHz, and four of them place a single full-scale cycle at different sample offsets to check that a meter catches the peak wherever it lands between samples.
2. Inter-Sample Overs in Commercial Releases
No regulator or academic group publishes a catalog-wide count of inter-sample overs, so the best available evidence comes from large independent scans and from engineering measurements of individual records. The largest recent scan is the library analysis published by Archimago’s Musings in August 2025: 13,354 albums and around 150,000 tracks, measured with a BS.1770-5 true-peak scanner, with more than 107,000 tracks grouped by genre. Its result: 44% of 31,400 general rock tracks peak above +0.1 dBTP, and the top 1% of rap and electronica tracks exceed +3 dBTP.
The pattern follows production style. Classical and multichannel releases, which are rarely limited hard, stay near or below full scale, while genres mastered against a brickwall limiter push their true peaks well past it. Engineering measurements of single records point the same way: one Steely Dan track logs inter-sample overs at about 3.7 per second, even though no individual sample is clipped (Benchmark Media, Intersample Overs in CD Recordings 2017).
| Metric | Value | Source |
|---|---|---|
| General rock tracks with true peaks above +0.1 dBTP (31,400 tracks) | 44% | Archimago’s Musings, library scan (2025) |
| Electronica tracks with true peaks at or below 0 dBTP (4,400 tracks) | 39% | Archimago’s Musings, library scan (2025) |
| 99th-percentile true peak: classical, pop, rap and hip-hop, electronica | +0.79, +2.87, +3.08, +4.22 dBTP | Archimago’s Musings, library scan (2025) |
| Multichannel tracks (more than 7,000) with true peaks above 0 dBTP | 0 | Archimago’s Musings, library scan (2025) |
| Inter-sample overs in one Steely Dan CD track (Gaslighting Abbie, a little over 5 minutes), highest over | 1,129 overs, +0.8 dBFS | Benchmark Media (2017) |
| Highest inter-sample peak measured on Regina Spektor’s Fidelity | +1.49 dBFS | Benchmark Media (2017) |
Outlier note: the same library scan found a single track reaching +7.72 dBTP, the highest value in the collection, and one general rock track at +5.88 dBTP. Most recent available data for the per-track engineering measurements: Benchmark Media, 2017.
3. Loudness-War Limiting and Clip Counts
Inter-sample overs are a side effect of a long trend toward denser masters. The most cited dataset is Emmanuel Deruty’s Sound On Sound analysis of the loudness war, which measured 4,500 commercially or critically successful tracks released between 1969 and 2010 and checked the result against 20,000 songs drawn from the EMI catalogue. Records measured roughly 5 dB louder than in the 1970s, and crest factor, the gap between average (RMS) level and peak level, fell by 3 dB from the start of the 1980s. Deruty also describes an unnaturally high density of samples sitting near 0 dBFS after the early 1990s, a fingerprint of brickwall limiting. Most recent available data: Sound On Sound, 2011.
When the average level climbs toward the ceiling, every extra decibel of limiting raises the odds of reconstructed peaks above it. Apple’s mastering guide shows what that looks like in a clip report: its example readout from the afclip tool counts thousands of clipped samples per channel, split between clips on the samples themselves and clips that only appear between samples after reconstruction (Apple Digital Masters, 2021). Broadcasters moved the other way: the EBU retired the old quasi-peak ceiling in favour of a true-peak ceiling that leaves loud passages more room (EBU Tech 3343, 2023).
| Metric | Value | Source |
|---|---|---|
| Tracks in the main loudness-war corpus (released 1969 to 2010) and in the EMI cross-check corpus | 4,500 and 20,000 | Sound On Sound (2011) |
| Loudness of records at the time of the study (2011) compared with 1970s records | roughly 5 dB louder | Sound On Sound (2011) |
| Fall in crest factor (peak level minus RMS level) since the early 1980s | 3 dB | Sound On Sound (2011) |
| Clipped samples in Apple’s example afclip readout, left channel: on the sample vs between samples | 10,661 vs 2,876 | Apple Digital Masters (2021) |
| Clipped samples in Apple’s example afclip readout, right channel: on the sample vs between samples | 9,690 vs 2,487 | Apple Digital Masters (2021) |
| Legacy broadcast permitted maximum level (quasi-peak meter) made obsolete by the -1 dBTP true-peak ceiling | -9 dBFS | EBU Tech 3343 (2023) |
Context note: these clip counts come from a worked example in the guide, not from a catalog survey, so they illustrate the scale a single heavily limited file can reach rather than an industry average. The guide states that afclip runs inter-sample detection by default and prints both totals side by side.
4. Codec Overshoot and Delivery Ceilings
Lossy codecs rebuild audio from a reduced description of it, and the rebuilt waveform can peak higher than the file that went in. The EBU’s distribution guideline puts it plainly: decoder output “may peak higher than the input level due to overshoots”, and relatively low bitrates “in general cause a rise of overshoots” (EBU Tech 3344; most recent available data: EBU Tech 3344 version 2.1, 2016). The practical answer is a -2 dBTP end-stage limiter in front of the encoder, one decibel below the -1 dBTP ceiling the EBU sets for uncompressed production audio.
Streaming services publish the same logic in their own terms. Spotify’s loudness guidance recommends true peaks below -1 dBTP for its lossy Ogg Vorbis and AAC streams and tightens that to -2 dBTP for loud masters, while Apple tells engineers to check the actual 256 kbps AAC encode for clipping, because levels that show no overs in PCM can still clip once encoded (Apple Digital Masters, 2021). Lower bitrates are where margins shrink fastest, which matters for speech-heavy formats served at modest bitrates; our podcast hosting bandwidth data shows the delivery side of that trade-off, and our lossless audio statistics cover the formats where codec overshoot does not apply.
| Metric | Value | Source |
|---|---|---|
| End-stage limiter setting before a lossy encoder | -2 dBTP | EBU Tech 3344 (2016) |
| Maximum true peak for MPEG-1 Layer II and Dolby AC-3 data reduction | -2 dBTP | EBU Tech 3343 (2023) |
| Clipping point of an AC-3/E-AC-3 decoder in RF mode (1 kHz sine), recommended maximum programme output | -3 dBTP, -4 dBTP | EBU Tech 3344 (2016) |
| Spotify true-peak recommendation for masters (lossy Ogg Vorbis and AAC delivery) | below -1 dBTP | Spotify for Artists |
| Spotify true-peak recommendation for masters louder than -14 LUFS integrated | below -2 dBTP | Spotify for Artists |
| Spotify playback limiter on the Loud setting: threshold, attack, decay | -1 dB sample value, 5 ms, 100 ms | Spotify for Artists |
| Apple headroom recommendation for masters bound for its 256 kbps AAC encoding | at least 1 dB | Apple Digital Masters (2021) |
| Limiter ceiling suggested by Sound On Sound when testing MP3 encodes | -3 dBFS | Sound On Sound (2016) |
Context note: the AC-3 figure is a reminder that the decoder, not only the encoder, sets the ceiling. In RF mode the decoder output is lowered by 3 dB to align loudness, so a 1 kHz tone clips at -3 dBTP; the EBU accepts the -2 dBTP studio limiter anyway because AC-3 carries its own overload protection. The Sound On Sound ceiling comes from a reader Q&A on inter-sample peaks and mix distortion and is a working suggestion, not a measured threshold.
5. The Playback Chain: DACs, Players and Headroom
The last place an inter-sample over can turn into audible distortion is the digital-to-analogue converter (DAC). Most converters oversample internally; if their digital filter has no room above 0 dBFS, a reconstructed peak above full scale is clipped inside the chip. The clearest published test is old but stark: in measurements by Keith Howard for Hi Fi News, 8 of 9 disc players squashed or distorted a +3 dBFS inter-sample test tone, and only one came close to the ideal (Lesurf, Over The Top measured results, 2012, republishing 2007 measurements; most recent available data: Lesurf, 2012).
The test follows the theoretical worst case. When full-scale samples straddle the peaks of a tone at one quarter of the sample rate, the rebuilt wave rises about 3 dB above full scale, the same condition the EBU’s test signal 19 checks in meters (see section 1). Converter makers that design for it add digital headroom above 0 dBFS; buyers comparing products can find the market context in our DAC and amplifier market statistics.
| Metric | Value | Source |
|---|---|---|
| Disc players tested with a +3 dBFS inter-sample test tone (Hi Fi News, 2007) | 9 | Lesurf, Over The Top measured results (2012) |
| Players that squashed or distorted the +3 dBFS peaks | 8 of 9 | Lesurf, Over The Top measured results (2012) |
| Test tone frequency (one quarter of the 44.1 kHz CD sample rate) and peak level between samples | 11,025 Hz, +3 dBFS | Lesurf, Over The Top measured results (2012) |
| Digital headroom above 0 dBFS built into Benchmark DAC2 and DAC3 (the vendor’s own design figure) | 3.5 dB | Benchmark Media (2017) |
| Maximum gap between a 4x true-peak reading and the analogue level after D/A conversion | less than 1 dB (typically a few tenths of a dB) | EBU Tech 3344 (2016) |
Context note: Benchmark states that every D/A chip and sample-rate-converter chip it has tested shows an inter-sample clipping problem. That is the manufacturer’s own claim, made while promoting a product designed around it, so treat it as an engineering observation rather than an industry statistic.
Summary: Mastering Limiter True-Peak and Inter-Sample Clipping by the Numbers
| Metric | Value | Source |
|---|---|---|
| Maximum under-read of a 4x true-peak meter (content up to half the sample rate) | 0.688 dB | ITU-R BS.1770-5 (2023) |
| Under-read the EBU allows for at 4x oversampling and 48 kHz | about 0.5 dB | EBU Tech 3343 (2023) |
| EBU Mode true-peak reading tolerance | +0.2/-0.4 dB | EBU Tech 3341 (2023) |
| Expected reading of the worst-case EBU true-peak test tone | +3.0 dBTP | EBU Tech 3341 (2023) |
| R 128 true-peak measurement tolerance | ±0.3 dB | EBU R 128 (2023) |
| Electronica tracks at or below 0 dBTP | 39% | Archimago’s Musings (2025) |
| Inter-sample overs in one Steely Dan CD track | 1,129 | Benchmark Media (2017) |
| Crest factor decline since the early 1980s | 3 dB | Sound On Sound (2011) |
| Limiter in front of lossy encoders | -2 dBTP | EBU Tech 3344 (2016) |
| Ceiling for MPEG-1 Layer II and Dolby AC-3 | -2 dBTP | EBU Tech 3343 (2023) |
| Spotify ceiling for masters louder than -14 LUFS | below -2 dBTP | Spotify for Artists |
| Spotify Loud-mode limiter attack and decay | 5 ms, 100 ms | Spotify for Artists |
| Apple headroom recommendation | at least 1 dB | Apple Digital Masters (2021) |
| Limiter ceiling suggested for MP3 test encodes | -3 dBFS | Sound On Sound (2016) |
| Disc players that squashed or distorted a +3 dBFS inter-sample tone | 8 of 9 | Lesurf, Over The Top measured results (2012) |
Methodology and Sources
- ITU-R BS.1770-5, Algorithms to measure audio programme loudness and true-peak audio level (November 2023): true-peak algorithm, oversampling and under-read table (Annex 2 and its Attachment 1).
- EBU R 128, Loudness normalisation and permitted maximum level of audio signals (version 5, November 2023): definition of maximum true peak level, -1 dBTP production ceiling and ±0.3 dB tolerance.
- EBU Tech 3341, EBU Mode loudness metering (version 4, November 2023): true-peak test signals 15 to 23 and tolerances.
- EBU Tech 3343, Guidelines for production of programmes in accordance with R 128 (November 2023): -2 dBTP for MPEG-1 Layer II and AC-3, 4x meter under-read margin, legacy -9 dBFS permitted maximum level.
- EBU Tech 3344, Guidelines for distribution and reproduction in accordance with R 128 (version 2.1, July 2016): encoder limiter, AC-3 decoder levels, D/A overshoot.
- Spotify for Artists, Loudness normalization: true-peak recommendations and playback limiter settings.
- Apple Digital Masters technology brief (April 2021): headroom recommendation, afclip inter-sample detection and example readout.
- Benchmark Media, Intersample Overs in CD Recordings (updated February 2017): per-track over counts and converter headroom (vendor source).
- Archimago’s Musings, A look at the music library: how much DAC intersample overhead is enough? (August 2025): genre-level true-peak distribution from an independent library scan.
- Jim Lesurf, Over The Top: measured results (July 2012): Hi Fi News player tests with a +3 dBFS inter-sample tone.
- Sound On Sound, Dynamic Range and the Loudness War (September 2011): 4,500-track loudness and crest factor analysis.
- Sound On Sound, reader Q&A on inter-sample peaks and mix distortion (December 2016): codec testing ceiling suggestion.
- Data watch: no peer-reviewed, catalog-wide study of inter-sample over incidence was found, so the genre figures in section 2 rest on one listener’s library scan, which is large but not a random sample of released music; the per-track engineering figures come from a converter vendor and cover hand-picked tracks; the Hi Fi News player tests date from 2007 and current converters may behave differently. The AES TD1008 streaming recommendation (2021), which discusses true peak at the codec input, could not be retrieved from aes.org during research and is not quoted here; the Nielsen and Lund AES papers on 0 dBFS+ levels (2000, 2003) were also unreachable. The afclip clip counts in section 3 are a worked example, not catalog data. The EBU distribution guideline has not been revised since 2016, so codec-specific ceilings may tighten or relax in its next edition.
- Last updated: October 24, 2026. We update this roundup quarterly, and the next refresh is expected when the EBU publishes the next revision of Tech 3344, its distribution and reproduction guideline.