A look at the music library: How much DAC intersample overhead is enough? True Peak analysis of genres in an eclectic digital collection.

In a recent article about Linn DS streamers, I touched again on intersample peaks ("True Peaks") as I measured the DAC hardware. In the comments section, there was further talk about this topic so this got me thinking over the last few weeks to spend some more time on this but from a different angle.

For those who are not aware of what "intersample peaks" are, recall that DACs implement filters to ensure accurate reconstruction of the waveforms (unlike the jagged-looking NOS DACs which some audiophiles still subjectively appear to prefer). Typically this is done through oversampling whereby 44.1kHz (ie. CD samplerate) might be increased by something like 8x (352.8kHz) while applying a low-pass filter that cuts off content beyond the 22.05kHz Nyquist frequency. The DAC will interpolate to create those extra upsampled datapoints. In this DSP process, some of the reconstructed "intersample" points might have >0dBFS values. If these "True Peak" values are too high beyond the limits of the digital filter or output level of our DACs, the samples will "clip", resulting in extra distortion in the playback.

For this post, instead of considering from a hardware measurement viewpoint what kind of overhead a DAC can handle, let's consider from the music content side by using a relatively large music library - my library of thousands of albums - and consider just how much overhead peak we might find and "need" based on the commonly-owned music genres in an audiophile's collection.

Before getting into the data, let me say that I have been fortunate to have a family of music lovers! Not necessarily hardware audiophiles of course since most family members care little about their speakers and DACs compared to me.🙂

While I have been the one collecting the majority of the library to be discussed here, mainly as CDs since the '80s then moving on to digital downloads, so too have my parents, siblings, and wife. My music library is therefore essentially the central "family music collection" with contributions from members, tagged, and available to all family members through Logitech Music Server / Lyrion Music Server typically streamed to them as high bitrate MP3 320kbps.

In 2025, this 2-channel library of lossless CD-resolution (and some 16/48 downloaded) albums number 13,354; note that this includes albums with multiple copies/remasters/deluxe editions, various "greatest hits" sets, vinyl rips, and singles so there's a little redundancy. This amounts to around 150k individual tracks/songs.

A. A stroll through the music library... and life.

Before we get into the more technical True Peak data, permit me to stroll down memory lane first to have a look broadly at the library of music accumulated based on the number of albums originating from each year across a lifetime:

As you can see, I've added some notes to the graph. The era of hi-fi stereo recordings began in the 1960s so it's quite natural to see the number of albums in my collection rising starting from the early '60s.

The peaks and valleys in the graph somewhat correlate with the years that I and the family (primarily Gen X siblings and spouses) were more active in music listening growing up and hence more engaged with collecting albums in those years of life. As a Gen X audiophile, I started working in my late teens in high school in the late '80s, got my first cheap CD player around 1988 and buying my first CDs then. The early '90s were awesome as I started university, actively listening and collecting. There was a dip after that in the mid-90's as I finished medical school and started medical residency, completing the training by 2001. From 2001-2004, I worked hard to pay off loans, bought my first house, got married so these were not years I listened to music as much. By the latter half of the '00s, with kids now at home, this actually revitalized my music consumption as I got into headphones and personal audio before resuming my interest in the larger loudspeaker system once the kids were old enough not to run into furniture and stick fingers into the loudspeakers. By the early 2010's, I was excited by the potential of hi-res audio and bought a bunch of albums off HDTracks, I then started this blog in earnest by early 2013; it was so easy to collect downloads by then with the computer music server set up with more than a terabyte of storage initially using Logitech Media Server (now Lyrion), and then Roon.

Notice what happened by the mid-2010's with the number of albums in my collection dropping precipitously! This is a result of the change in technology with streaming services and subscriptions to Spotify, Apple Music, Qobuz, and Tidal basically killing the need for buying albums weekly to add to the library. Furthermore, I suspect as one gets older, we start becoming less excited by the newer artists. I'll leave you to consider whether music quality is getting worse these days as has been discussed by others. These days I still have Apple Music and Spotify. Not just me, but also my siblings didn't buy nearly as many CDs/downloads for the family collection by the mid-2010's.

I got into vinyl between 2014-2018 but knew quickly that this was not something I would collect much of. LPs of course can sound good but not high enough fidelity for my preference, among other space and cleaning hassles. I knew my music purchases were dropping over the years, but it's quite amazing to see this quantified on a graph like this!

Furthermore, from 2021 onwards with the availability of "Spatial" multichannel audio, I consciously started to stream more of those albums when I can, than buying 2-channel stereo, furthering the decline in music being added to the 2-channel collection in the 2020's. It should be no surprise then that in my music library, multichannel additions are increasing in greater proportion. As for the much smaller hi-res (24-bit and >48kHz) side of my collection, this has been dropping since 2019 as well for reasons previously discussed.

While the ups and downs will be different, correlating to individual life experiences, I wonder if other avid, eclectic Gen X music collectors out there might be seeing a similar pattern if they were to plot out their music collections across the years. 

Let's end the reminiscing here and get back to the main technical topic today. 😉

B. True Peak Testing

I was therefore curious: using this eclectic "audiophile" library, if we analyze the "true peaks" of all this material from across the decades, how much DAC intersample overhead is needed to prevent clipping if played bit-perfectly?

To examine this, I will analyze some of the directories I have in my music library consisting of the most common genres collected over these decades. I'll use foobar and the True Peak Scanner component; current version 0.6.14 from June 2024 which implements ITU-R BS.1770-5 for the calculation of individual tracks. We can then look at a histogram of the distribution of this True Peak data.

The way I thought I will attack this one is to examine albums based on genre instead of just as a monolithic library, in part because this is also how I loosely organize the folders in my music collection.

Note that it's not always easy to judge the best genre to categorize a particular artist or album because musical influences are diverse and artists evolve over time. For example, I've categorized U2's material in the "Rock" category but one could argue that they're more pop and some of their albums might have more blues influences. Likewise, an artist like the audiophile-favorite Diana Krall I suppose is "Jazz" but because she's more a vocalist in my view, I've categorized her material in my "Vocals" collection. Just be mindful then that there will always be some shades of grey when it comes to these broad categories I'm presenting. However, over thousands of tracks, I think the results will point towards a reasonable representation.

     1. General Rock

To start, as a dad-rock audiophile who loves his classic and progressive rock, let's have a look at my "General Rock" directory which is where I have some typical stuff: Pink Floyd, Fleetwood Mac, 38 Special, Adam Ant, America, Bryan Adams, Bad Company, Blue Rodeo, Bon Jovi, Coldplay, CSNY, Dire Straits, Don Henley, Steely Dan, Genesis, Frank Zappa, Grateful Dead, Hendrix, The Beatles, INXS, Kate Bush, Lenny Kravitz, Lou Reed, Lynyrd Skynyrd, Mark Knopfler, Nick Drake, Nirvana, Oasis, Pearl Jam, Peter Frampton, Queen, REM, Roger Waters, Sheryl Crow, Sonic Youth, Steve Earl, B-52s, The Flaming Lips, The Who, The Rolling Stones, U2, as examples of the acts I lump together in this directory. Since I believe mastering is important, it's not unusual for there to be 2 or 3 different masterings of the same album.

Here's a peek at the histogram of dBTP (dB True Peak) analyzed for each song:

Notice in this first graph that each bar on the X-axis is of 0.1dB span. We can see that with rock recordings, the highest number of songs sit right between 0.0 to +0.1dBTP (at least one sample in the whole song between 0.0 to +0.1dBTP).  We see that 44% of these recordings have true peaks higher than +0.1dBTP which means that our DACs do need to have at least some intersample overhead.

Within this collection of 31.4k tracks, the highest TP I found was one track with peaks up to +5.88dBTP.

While a higher resolution histogram like the above gives us a good look at the shape of the distribution, let's lower the resolution of the histogram and consider the X-axis as 0.5dB 'bins':

As an audiophile, suppose we conservatively say we want our DACs to be able to handle 99% of all tracks without clipping. Based on the histogram above, we can say that a DAC that can handle +2.35dB overhead will play 99% of my "general rock" tracks without concern.

     2. Classical Music

I know many of you audiophiles are classical music enthusiasts. Without question, the care taken when recording classical music makes these some of the best, most natural-sounding albums.

Here's the True Peak Histogram of the 10.1k stereo classical tracks I own:

Notice the contrast between this pattern and the "General Rock" category above!

Classical music albums typically have higher dynamic range (commonly >DR10). Notice on the far left of the graph, many tracks have even lower than -6dB true peaks! And for 99% of classical tracks, a DAC that can accommodate just +0.79dB peaks will be enough.

     3. Jazz & Blues

While different genres, jazz and blues are historically related and I typically keep these together in the same directory. This is where I put Miles Davis, Albert King, Antonio Carlos Jobim, Art Blakey, Art Pepper, B.B. King, Ben Webster, Bill Evans, Buddy Guy, Cassandra Wilson, Charles Mingus, Chet Baker, Chuck Mangione, Count Basie, Duke Ellington, Freddie Hubbard, Gordon Goodwin's Big Phat Band, Howlin' Wolf, Herbie Hancock, John Coltrane, John Lee Hooker, Keb' Mo', Kenny G, Louis Armstrong, Lee Morgan, Ry Cooder, Stan Getz, Monk, Wynton Marsalis, and the like.

With 12.6k tracks analyzed, here's the distribution:

Clearly, these are louder recordings than classical music and closer to my "general rock" collection. Using the same threshold, a DAC needs to be able to handle up to +2.42dB intersample peaks for 99% of these jazz/blues tracks. I was actually a bit surprised that this was so similar to the General Rock directory.

     4. Soundtracks

Ah yes, soundtracks and soundtrack scores. Many of these albums can be considered as representatives of modern classical music. Within the albums we also commonly come across vocal tracks. Notice if we go to our local symphony orchestra these days, soundtrack materials are often played in live concerts intermingled with the classical numbers of olde.

Other than the usual movie soundtracks and scores (like John Williams, Alan Menken, Alan Silvestri, Danny Elfman, Ennio Morricone, Henry Mancini, Howard Shore, James Newton Howard, Jerry Goldsmith, Ludwig Goransson, Hans Zimmer, Michael Giacchino, Ramin Djawadi, Thomas Newman...), I have a number of anime soundtracks (from the likes of RADWIMPS and Joe Hisaishi) as well as videogame scores like the excellent recent Black Myth: Wukong (2024, the multichannel/Atmos version is available and very cool!).

In this directory, I have 8.9k tracks:

Indeed, like the Classical Music pattern above, these are on average less dynamically compressed recordings and 99% of tracks have true peaks below +1.70dB.

     5. Vocals

An audiophile favourite genre often found in hi-fi demos and showrooms!

Here we find the likes of Diana Krall, Aaron Neville, Alfie Boe, Aretha Franklin, Astrud Gilberto, Barb Jungr, Barbra Streisand, Bobby McFerrin, Carol Kidd, Cécile McLorin Salvant, Charlotte Church, Claire Martin, Dominique Fils-Aimé, Ella Fitzgerald, Eva Cassidy, Frank Sinatra, Hayley Westenra, Jane Monheit, Jacintha, Joan Baez, Josh Groban, Melody Gardot, Michael Bublé, Nat King Cole, Nina Simone, Radka Toneff, Rebecca Pidgeon, Smokey Robinson, Tony Bennett, etc.

With a total of 6.5k tracks analyzed:

Since many of these are "audiophile approved" recordings, they're often not as loud on average, 99% of tracks have true peaks below +1.83dB which is very close to what we saw of the Soundtrack albums above.

     6. Country

This next one is a bit of a mixed bag because many country albums are really pop/country hybrids like Shania Twain, Dixie Chicks, Garth Brooks, Rascal Flatts while others are more prototypical country artists like Dolly Parton, Willie Nelson, along with traditional-sounding Alison Krauss + Union Station bluegrass or the guitar work of Chet Atkins.

I don't have a large collection. With 4.6k tracks analyzed:

99% of tracks have true peaks below +1.89dB. Interestingly these are not much higher in TP than the music in the Vocals directory above.

     7. Hard Rock & Metal

Alright folks, let's start moving into music less likely to be played at hi-fi shows 😉.

To be honest, as a "classic rock" kind of guy, my choice in hard rock and metal isn't the more extreme or brutal material like death metal, gore metal stuff. Rather, I'm good with AC/DC, Alice in Chains, Black Sabbath, Alice Cooper, Metallica, Disturbed, G'n'R, Iron Maiden, Judas Priest, Lita Ford, Megadeth, NIN, Ozzy, Queensryche, Rush, Rammstein, Scorpions, Tool, White Snake, ZZ Top kind of stuff.

With 5k tracks analyzed:

Yup, it's definitely louder compared to many of the other genres but at least true peaks were not as high as even in the "General Rock" category. 99% of tracks have true peaks below +2.03dB here. 

     8. Rap / Hip-Hop

I also don't have much rap/hip-hop in my music library beyond the more popular and radio-friendly albums; IMO, the excessively crude language and often unpleasant, coarse, topics are just not quite to my taste.

With over 2000 tracks consisting of albums by 2 Live Crew, 2Pac, 50 Cent, Beastie Boys, Dr. Dre, Doechii, Drake, Eminem, Jay-Z, Kanye, Kendrick Lamar, MC Hammer, Rick Ross, Salt-n-Peppa, Snoop Dogg, The Notorious BIG, Tone Loc, Wu-Tang Clan, and the like:

As you can see, the material in this directory is louder than "General Rock" with the 99% threshold at +3.08dB.

 

     9. Electronica

Another popular genre these days. This is my directory of electronic, techno, EDM, DJ mixes. Let's see, we have Aphex Twins, Armin van Buuren, Art of Noise, Blank & Jones compilations, Booka Shade, Brian Eno, Boris Blank, Burial, Daft Punk, David Guetta, Deadmau5, Tiësto, Galantis, Giorgio Moroder, Jean-Michel Jarre, Lichtmond, Moby, Orbital, Prodigy, Skrillex, The KLF, dubstep compilations, Vini Vici, etc.

Survey of 4.4k tracks show:

Wow, we have a winner folks! This stuff's loud and the synthetic waveforms can create some pretty wicked high intersample peaks. Unlike the other genres, only 39% of these electronic tracks have true peaks at or below 0dB!

99% of tracks have true peaks below a high +4.22dB!

     10. Pop

Finally, let's just look at my directory of Top-40 pop artists. Pop is of course just popular music for the times and comprises of all kinds of genres and subgenres including some R&B, hip-hop, soul, rock, synthpop, etc.

Over the decades, I have all kinds of stuff from obvious pop acts like ABBA, Ace of Base, a-ha, Aqua, Ariana Grande, Backstreet Boys, Bee Gees, Beyoncé, Bruno Mars, Celine Dion, Debbie Gibson, Duran Duran, Ed Sheeran, Frankie Goes to Hollywood, George Michael, Go-Go's, Michael/Janet Jackson, Jason Mraz, The Weeknd, Justin Bieber, Lady Gaga, Madonna, Milli Vanilli, 'NSync, Paul Simon, P!nk, Phil Collins, Spice Girls, Taylor Swift, The Carpenters, Tiffany, Pet Shop Boys, Rick Astley, Wham!, Wilson Phillips, etc. here.

With a healthy 21.7k tracks analyzed:

We can show that 99% of these tracks have true peaks below +2.87dB. Perhaps predictably similar to the General Rock category given similar production techniques over the decades.

C. Summary:

Let's summarize then the results for these genres after analyzing thousands of tracks sourced from CDs and lossless stereo downloads; this might be one of the largest analyses published with more than 107k tracks examined in total. Obviously, depending on what we have in our collections, the exact numbers will vary, but I suspect the results I have here represent a reasonable distribution for those of us with eclectic tastes in music.

If we want to ensure absence of intersample clipping for 99% of tracks in each of these genres based on the library, here are the dB peaks the DAC/CD player should be able to handle - sorted from genre with lowest to highest value:

Classical                  +0.79dB

Soundtracks                +1.70dB
Vocals                     +1.82dB
Country                    +1.89dB
Hard Rock/Metal            +2.03dB
Classic & Progressive Rock +2.35dB
Jazz & Blues               +2.42dB
Pop                        +2.87dB

Rap / Hip-Hop              +3.08dB
Electronica                +4.22dB

As you can see, Classical Music tracks by nature of their recording of acoustic instruments and higher dynamic range do not place much demand on DACs when it comes to pushing intersample peaks. In fact, a DAC that can provide just +1dB intersample overhead will be more than enough for 99.5% of the 10,000 classical music tracks I analyzed.

Contrast this with the synthetic waveforms of electronica and dance music where more than 60% of the tracks will have intersample peaks above 0dB and in order to accommodate 99% of all tracks, you'd want a DAC with about +4.25dB intersample overhead! These are the extremes between the genres.

Out of interest, of all the thousands of tracks analyzed, the one with the highest True Peak was from the electronic album Making Orange Things by Venetian Snares & Speedranch, called "Meta Abuse" with peaks up at 7.72dBTP! I think I got this "hardcore techno" punishment track from my brother's collection years ago. Note this is clearly not "audiophile" music. 😱

I suppose if you're the ultimate obsessional audiophile, you're welcome to look for a DAC with +7.75dB headroom to essentially guarantee never to run into issues with your music. 🙂

Here's a look at "Meta Abuse". We can see the extensive intersample overload through most of the track above 0dBFS:

Notice that the extremely high peaks are relatively brief even though intersample overs occur throughout. This track measures a horrific DR0. And beats the 1997 remix of Iggy & The Stooges' Raw Power with its average DR1. 🤣

Imagine spending good money on a DAC and hi-fi system to reproduce "music" like this! An extreme example of why we don't need to spend much money if the quality of the content simply cannot be improved regardless of hardware. 

My recommendation of +3dB discussed a few years back for DAC intersample overhead I think remains fair. This will easily cover the vast majority of music out there (including 96% of my Electronica tracks). However, if you're very much into electronica/techno/EDM, maybe a DAC with +4.25dB headroom would given you a bit more assurance - or simply attenuate your DAC digital volume by a few dBs, or even better, use ReplayGain with something like -18 LUFS target.

While these >0dBFS calculated intersample peaks might exist in 'good' sounding recordings, I see them mainly as a reflection of loud productions that are the victims of the Loudness War. So while I think it's useful to consider whether our CD players and DACs can handle them, to me it's actually not that important. When possible, the prudent audiophile is simply advised to look for recordings that are not so "hot", that are more dynamic, and in turn more "natural" sounding (more like classical music production quality).

--------------------

BONUS: MULTICHANNEL!

Before ending, it would be incomplete of me given recent discussions to not include some data on multichannel material.

Here's a directory of 5.1/7.1 FLACs converted from Dolby TrueHD and EAC3 5.1 to give you an idea of what intersample peaks look like in the world of modern "Spatial Audio" (this analysis did not include old SACD or DVD-A mixes from the early 2000's which I know can also be quite dynamically compressed). No volume changes or normalization were applied to the conversion process from the Dolby encodings to straight PCM.

This directory contains mixed genres including rock, pop, classical, EDM, vocals (yes, including Diana Krall of course!) and jazz. A few artists represented here include ABBA, ABC, Alan Menken, Andrew Lloyd Webber, Ariana Grande, Armin van Buuren, Avril Lavigne, Billie Eilish, Billy Idol, Bob Dylan, Bon Jovi, Carrie Underwood, Chicago, Daft Punk, Dua Lipa, Fleetwood Mac, George Michael, Hans Zimmer, Miles Davis, Joe Hisaishi, John Mellencamp, Joni Mitchell, Lady Gaga, Madonna, Maroon 5, Michael Jackson, Miley Cyrus, Olivia Rodrigo, Roxy Music, Shania Twain, Sting, The Chainsmokers, UB40, Whitney Houston:

With more than 7,000 tracks analyzed, notice that there were ZERO tracks with true peaks above 0dB and 99% of tracks have true peaks below -0.44dB!

This is what healthy digital mastering looks like without strong dynamic range compression! As discussed, multichannel audio sounds and measures with higher dynamic range. Let's hope that the mastering studios continue to respect limits on output levels as defined by Dolby and Apple's target of integrated loudness value not exceeding -18 LKFS. To be honest, I would rather listen to a lossy high bitrate EAC3-JOC Atmos than the typical DR8 stereo lossless version stripped of dynamics and nuances.

As an audiophile desiring high-fidelity sound, I wish we had enforced loudness limits for our 2-channel audio since the mid-90's! For digital stereo music, it's too late at this point to do anything other than hope that going forward, younger audio engineers and artists recognize the problem, and remastering efforts can still find dynamically uncompressed versions of older albums to re-release.

I was watching a mixing video with Andrew Scheps the other day and he clearly noted the increased dynamic range of his mixes in multichannel compared to his usual 2-channel process due to the Atmos integrated loudness limit. Over time, he expressed the possibility that many artists might transition to creating albums intended for multichannel/Atmos playback as their default mix. When this happens, the multichannel mix becomes the definitive presentation with the stereo version a tweaked derivative; perhaps similar to the transition from mono to stereo in the late '50s and early '60s. A recent album like Roger Waters' This Is Not A Drill - Live From Prague (2025) is an example where the downmixed multichannel-to-stereo sounds amazingly dynamic (don't bother with the 24/96 stereo mix, it's just upsampled but at least has DR11). Although it will take time, certainly this is very possible given the availability of tools and it'll be interesting watching studio productions evolve over the coming years!

I think the ultimate conclusion to all of this is simply this: instead of insisting that our DACs have high intersample overload headroom, it would have been much better if the music were created respecting 0dBFS as the limit. I would argue that music with high intersample peaks - let's say more than +1dB which is higher than 99% of my classical music collection is likely not great sounding "hi-fi" material anyways.

At the end of all our obsessive audiophile hardware explorations, we might just come back to the earnest realization that one of the most important factors in sound quality is simply the quality of the music library we listen to. Certainly when it comes to low-priority hardware (like cables and tweaks as discussed last year), at some point I think once we understand the nature of the hardware and what it cannot do as opposed to the hype that some want us to believe in, it becomes much more interesting to focus on which next album to acquire than what hi-fi hardware we need to buy.

With that, it's time to end with some recent music I've been listening to. Here's a track ("Mystery") from the excellent Turnstile album GLOW ON (2021):

A good example of hard rock done well, available also in multichannel/Atmos.

South African Abel Selaocoe's Hymns of Bantu:

Well-recorded cello, complex instrumentation, songs drawn from African hymns. This also makes for a nice multichannel system show-off for those wanting a world classical crossover album.

Finally, for the summer of 2025, the movie KPop Demon Hunters has been surprisingly popular on Netflix as the most-watched animated online stream ever achieved in 5 weeks, plus songs on the pop charts. If you like K-Pop styled music, definitely check this out! Here's "Golden", David Guetta's Remix:

A couple of the songs (this one and "Your Idol") have made it on the Billboard Top 10 already. The Guetta remix is available in Atmos with quite an active mix. I'm a bit surprised that the full soundtrack isn't available on Atmos at this time.

Hope you're having a great August, everyone! Enjoying the music of course...