Friday, July 7, 2017

Is 'OK Computer' really DR10 on vinyl?



Spend some time among audio enthusiasts online and you will soon be confronted with the unassailable truth that the only way to truly experience Radiohead's classic 1997 album OK Computer is on vinyl. Not just any vinyl, the original UK Parlophone vinyl pressing. This is because purportedly the vinyl cut is not as much a victim of the loudness war as the original compact disc release. Screenshots of dynamic-looking waveforms and dynamic range analysis scores are used to bolster this claim. Of course, comments of effusive praise are usually all that is needed to affirm this Truth -- often along the lines of "the vinyl master blows the CD out of the water!" I mean, just take a look at those DR scores!

What is implicit in all of this is that the program material that was used for the vinyl cut was mastered with more dynamic range than the digital CD release. This means that less peak limiting and dynamic range compression was employed for the "vinyl master" than the "CD master." But is this really the case? I have never seen specific evidence of this from the artists and engineers involved in the album's production, unlike, say with the recent remix of Sgt. Pepper's Lonely Hearts Club Band. Silence from Nigel Godrich and Chris Blair notwithstanding, vinyl superiority bias (it's analog! no digital stair steps! you can't cut loud masters to vinyl! etc...) and high DR scores and spiky vinyl rip waveforms serve to form a feedback loop of confirmation bias that clouds what we expect to hear, what we think we hear, and what we're actually hearing.

What if I told you that you can turn the original OK Computer CD (DR7) into DR10 (like the vinyl rips) without adding any dynamic range by using simple high/low pass filters to make the bass mono (such as what might be done during a standard vinyl cut)? I have shown previously that using a filter for rolling off some bass content can yield high DR scores and dynamic-looking waveforms (see also: this video). This method of making the bass mono, which I outline below via screenshots, achieves a very similar result using "The Tourist" as an example:




















Multi's Mono Bass Mix DR Analysis
foobar2000 1.3.8 / Dynamic Range Meter 1.1.1
log date: 2017-07-06 18:57:39

--------------------------------------------------------------------------------
Analyzed: ? / ?
--------------------------------------------------------------------------------

DR         Peak         RMS     Duration Track
--------------------------------------------------------------------------------
DR10      -2.78 dB   -15.27 dB      4:44 ?-01 Airbag
DR10      -2.62 dB   -16.33 dB      6:24 ?-02 Paranoid Android
DR9       -2.95 dB   -13.99 dB      4:28 ?-03 Subterranean Homesick Alien
DR9       -2.89 dB   -16.62 dB      4:25 ?-04 Exit Music (For A Film)
DR10      -3.25 dB   -15.35 dB      4:59 ?-05 Let Down
DR10      -2.26 dB   -14.49 dB      4:22 ?-06 Karma Police
DR10      -5.74 dB   -18.78 dB      1:57 ?-07 Fitter Happier
DR10      -2.72 dB   -14.17 dB      3:51 ?-08 Electioneering
DR10      -2.55 dB   -14.94 dB      4:45 ?-09 Climbing Up The Walls
DR10      -3.24 dB   -15.20 dB      3:49 ?-10 No Surprises
DR9       -3.50 dB   -14.22 dB      4:20 ?-11 Lucky
DR10      -3.14 dB   -15.86 dB      5:27 ?-12 The Tourist
--------------------------------------------------------------------------------

Number of tracks:  12
Official DR value: DR10

Samplerate:        44100 Hz
Channels:          2
Bits per sample:   16
Bitrate:           1411 kbps
Codec:             PCM
================================================================================



Without going into too much detail about the process, what I've done is created two files, rolled off the bass in one, rolled off the high frequency content in the other, made the low frequency content mono, and then mixed that back into the original file from which the flow frequency content was rolled off. Thus we end up with the same frequency content overall as the original file (perhaps a slight bump around our chosen crossover point for the new mix), but with a dramatically different looking waveform and DR analysis. In fact, it looks just like a vinyl rip! But how does it sound? I've prepared a few clips from "The Tourist" for comparative purposes (level-matched, of course):
The vinyl rip will always sound different for myriad reasons; the biggest differences here are some high-frequency roll off (thus the 'warm' vinyl sound), perhaps some pleasant harmonic distortion, and how the stereo image manifests due to the limitations of the medium and playback equipment (one may prefer these sonic characteristics, but vinyl isn't "better" in this regard from the perspective of audio fidelity). Never mind all that; instead, listen carefully to the balance of the elements in the mix.

Remember, the issue here is dynamic range. I've shown that the normal process of filtering can result in the same kinds of dynamic-looking waveforms and high DR scores associated with vinyl rips. However, I haven't added any dynamic range. It only looks like I did. Of course, the mono bass mix does sound subtly different than the original CD -- how could it not? To me, the mono bass mix sounds just as "dynamic" as any vinyl rip I've heard of this album (I also own a copy of the Parlophone pressing). 

So is OK Computer really DR10 on vinyl?

Sunday, March 5, 2017

Clipping Distortion and Digital Bass Management

This is your audio on bass management. (source)

I recently posted a discussion of my experience with digital bass management somehow introducing distortion when listening to music. It was determined that enabling the receiver's bass management process reliably led to audible distortion with some recordings. I had been listening to music with bass management for years without any trouble. But when listening to Landscape In Portrait by Marihiko Hara, it was immediately apparent that something was extremely wrong. Now I know why.

An audio signal processor's bass management system is a set of digital filters with selectable crossover points. This means that instead of sending the main speaker channels the full bandwidth signal, the receiver filters out the bass that is sent to the speakers (high-pass filter) and redirects the bass signal to the subwoofer (low-pass filter). This is not a brickwall filter where all content below the crossover point is redirected to the subwoofer; instead, the bass is gently "rolled off" on a slope to allow for a smooth transition between the speakers and sub.

Crudely drawn illustration of a crossover filter slope by the author

In the last few months, I have been applying high-pass filters to digital audio recordings in order to properly compare them to vinyl rips (i.e. needledrops). It was immediately apparent that I needed to attenuate the signal before applying the filter. The whole point of comparing vinyl rips with filtered versions of retail digital releases was to demonstrate the "false peaks" that resulted in more dynamic-looking waveforms. So if I ran the filter without first creating headroom, those newly created "peaks" with run up against the -0dBFS ceiling and be clipped.

While listening to an album in Direct mode (no bass management) last night, it occurred to me that the very same "peaks" I encountered in my filtering tests could be the source of the bass management distortion problem. After all, the bass management process is not unlike what I was doing in the digital audio workstation, except that the receiver isn't compensating for the potential clipping issues that may result in filtering.

As soon as I thought this could be the issue I was experiencing, I was certain it was indeed the case. But a hunch will not suffice: A good pretend scientist must make an experiment! Using the track "Yoru Ga Mita Keshiki" from Landscape In Portrait, let's take a look at some waveforms.

"Yoru Ga Mita Keshiki"
As we can see, this minimalist piano music is mastered in such a way that it often peaks near the digital maximum. Played in Direct mode or through headphones, it sounds perfectly clean. It is only when the bass management filter is applied that distortion becomes an issue.

"Yoru Ga Mita Keshiki" high-pass filter applied at 80Hz
Those alarming red lines are Audacity's way of telling me that the software is detecting clipped samples. This is effectively what is being routed to the amplifiers for the main stereo channels during bass management (another signal that is low-pass filtered is routed to the sub at line level). When I play this file with my receiver in Direct mode, I hear the exact same 'tearing' distortion sound that I encountered with bass management. Problem: Identified! But before going further, let's take a look at what the filtering looks like for this song if I "turn it down" first:

Same song, attenuated by 4dB

High-pass filter at 80Hz after attenuation

We can see now that the filtering results in peaks higher than what was represented in the attenuated file. These were clipped in the version above with all the red lines. This clipping is precisely what is happening when the "full volume" digital audio is hitting the receiver's processor and the bass management filters are being applied. When I listen to my filtered version in Direct mode, the distortion is audible during the same passages as when listening to the unaltered file with bass management enabled.

Let's zoom in to see what's happening in a little more detail:

Before filtering...

...after filtering

When we apply the filter to the unaltered file, we can clearly see the red lines indicating clipping. We now have square waveform tops in place of smooth curves. Squares do not sound very good when your speakers try to reproduce them. Some recordings are intentionally clipped, but this one wasn't.

Attenuated before filtering...
...after filtering
If we lower the volume of the file before applying the filter, we are creating enough headroom to allow for the peaks to resolve. There are no square wave tops or red lines in the filtered version above, and thus no distortion. 

Update: Listen for yourself with the clips below. They were recorded from my Oppo UDP-203 playing the FLAC file of "Yoru Ga Mita Keshiki" using the stereo analog outputs of the player. I've bypassed the receiver, but the results are the same. One clip is the audio played full range, and the other is with the players bass management system enabled (front left/right channels set to "Small"; crossover at 80Hz).

Merely by using the bass management ubiquitous in home theatre receivers, we are potentially adding a not insignificant amount of audible clipping distortion. This will potentially occur in any digital recording without sufficient headroom for peaks to resolve after filtering. In reality, this is happening all the time. It just took this special case with minimalist piano music mastered in such a way that some peaks ran into the digital ceiling after bass management filters were applied. There were a few times in other recordings where I thought I'd heard something similar, but couldn't be sure. Whenever I went looking for the offending sections monitoring carefully in headphones, I couldn't find them. Now I know why: No filters are applied with the headphone output.

So what's the solution? I mostly play music directly from my computer via optical cable connected to my Onkyo TX-SR706. Given that all is required here is a bit of headroom in the recording before it hits the receiver's processor, I should be able to create it by adjusting the output volume in my music playback software. Sure enough, by turning down the volume in foobar the recording sounds perfectly clean in a bass managed configuration. It doesn't take much, -6dB should be plenty if you want to play it safe. It's that simple. (Of course, this means you are sacrificing 1 bit of resolution. Given this filtering-causing-clipping thing is only an issue because digital audio is mastered at such high levels, I wouldn't worry too much about effectively elevating the noise floor from -96dB to -90dB.)

But what about playing a disc or file from a non-personal computer source? I have a standalone Pioneer CD/DVD-Audio/SACD player for playing various music discs. To my knowledge, there is no way to adjust the volume of the the digital output before it reaches the receiver. Therefore, any audio signals that peak near -0dBFS will potentially be subject to clipping distortion due to bass management filtering. The same goes for bitstreaming Dolby or DTS-encoded audio (lossless or otherwise) from Blu-rays via my Playstation 4. I think the PS4 has a volume option, though it would require decoding everything to LPCM before sending it to the receiver digitally for bass management. Though movies tend to be mastered with plenty of headroom compared to most music, so the likelihood of encountering clipping due to bass management filtering is much lower.

There is also the option of some players like those available from Oppo which have volume controls on their analog outputs. They also feature bass management on their analog outputs, which could be sent to a capable receiver and run in Analog Direct mode. However, there's no guarantee the Oppo's filters wouldn't clip the audio. (Update: I have since purchased the Oppo UDP-203 and yes, using the player's bass management system for analog output also results in distortion.) One might choose to send use the Oppo's analog output without bass management and lower the volume before it hits the receiver for bass management, but the receiver would then need to do an analog-to-digital conversion before its processor could apply the filter. It wouldn't be an ideal solution, but it's a workable one if you don't mind the extra conversion.

It must be so obvious to the engineers designing this equipment that bass management is a source of potential distortion. Ideally, the receiver would have an option for lowering the level of a given digital input prior to processing the signal. My receiver doesn't have this option, though I'm not sure if more recent models have some equivalent feature for avoiding the issue.

Given how loud modern recordings are, it's a wonder why this problem never required my attention sooner. The simple reason for this is that the distortion is masked by the music and would only be noticeable during quiet passages. But quiet passages don't usually contain peaks near -0dBFS... unless all the music is quiet and is therefore allowed to be mastered that way. A quick Google search about this issue doesn't seem to come back with many discussions of the problem, and what few results there are tend to specifically deal with subwoofer clipping. That's not the issue I encountered: The distortion is present in the main channels. Hopefully this post will be helpful those those who like to listen to their music in a "2.1" bass managed stereo configuration like me:

If you are encountering clipping distortion when using digital bass management, the solution is to lower the volume of the player's output before it reaches the receiver's processor for filtering.

Sunday, February 26, 2017

Stereo Bass Management & A Distortion Dilemma: Toward An Acceptable Compromise (Update: [SOLVED])

Image by Big Daddy

 Update: Solved*


A few nights ago, I turned out the lights and turned up the volume on my Onkyo TX-SR706 to listen to Landscape In Portrait by Marihiko Hana. As you can tell from the video below the style is mainly solo piano, and a fair minimal one at that:



Very soon after starting up the album, I was greeted by some rather non-musical distortion emanating from my speakers. It sounded like brief tearing noises that were produced when a piano note was played loudly with some attack. Initially I thought it may just be an issue with the recording, perhaps it was clipped and was meant to sound that way. However, the distortion kept returning and was terribly non-musical. I switched over to listen via my pair of Audio Technica ATH-M50x headhones and the distortion disappeared.

That's odd, I thought, what was going on here. Were the headphones not revealing enough? Are my speakers blown? Is there an issue with the amp? Needless to say I was worried and that some investigation was surely in order.

Googling "distortion" and "piano music" I came across some forum threads where users argued that piano music is notoriously difficult to capture and reproduce. The instrument is capable of tremendous dynamic range and can easily overload an improperly configured microphone. The dynamic range can also cause issues with amplifiers that are not up to the task of delivering clean wattage for the highest peaks. To me, the recording seemed fine (no issues in the waveform that I could detect) and I was still hearing the distortion at low volumes. This suggested to me that it wasn't a problem with driving the amp or speakers too hard.

Still, I was concerned. I was able to quickly test the same files on a wholly different receiver/speaker setup and lo and behold: distortion! Now I could be reasonably sure my speakers aren't damaged (phew). Was the amp the culprit? Why did it sound fine in my headphones?

A little more deep googling alerted me to another thread where an audio enthusiast was having trouble with his signal processor.  Their distortion issues went away after replacing it. I have no other means of using another processor with my A/V receiver. Then I got the idea to try the "Pure Audio" mode, which passes the signal from the input stage to the amplifiers with as little processing as possible.

It worked. The music was crystal clear with no audible distortion. Everything sounded as it should. Except... no subwoofer.

The Pure Audio mode (and Direct mode on this particular receiver) disables the bass management filters and processors. I am quite fond of listening to stereo music in a 2.1 configuration with a subwoofer appropriately crossed over with the speakers (I actually use a 5.1 setup, though I rarely choose to upmix stereo to surround using built-in Dolby or DTS algorithms). I like my subwoofer and I like hearing (er, feeling) the lowest frequencies of the program material. Engaging the receiver's crossover filter immediately introduced the awful distortion, and bypassing it instantly solved the issue. But what about the subwoofer?

Fortunately, this receiver has a Double Bass feature. This means in Stereo mode, which is usually used to engage crossover and other settings, I can run my front left and right speakers in Full Range mode and have to receiver also send bass information to the subwoofer via the LFE output. Instead of some bass being routed to the sub away from the main channels, the speakers get the full bandwidth signal and the sub gets some love too. And there is no distortion!

I have no idea why this distortion is introduced in Stereo mode with bass management engaged. It does not distort with bass management engaged when I'm using Dolby Pro Logic II or All Channel Stereo modes. This is bizarre.

Whatever the cause, I was on my way to a solution. I had been using the receiver's bass management system to control information routed to the subwoofer and the settings I used performed really well. Using Room EQ Wizard, I performed low frequency sweep tests to measure the response from the listening position. Here's what I'd be using all along:

Bass management on receiver, crossed over at 80Hz
Using the built in microphone on a laptop (I know, not an ideal measurement tool), I can get a good idea of how the system is performing. These graph is useful for confirming what I'm hearing when I do the bass tests. It's fairly smooth in the prime listening position, which is more than can be said for elsewhere in the room...

But if I want to avoid the distortion caused by the stereo bass management in the receiver, I must deviate from this setup. First, here are some other measurements to provide context for performance and settings:

Running the two main front speakers full range; no subwoofer or bass management
Bass management on receiver, crossed over at 50Hz

As you can see, running my SVS Ultra Towers at full range leaves a little to be desired when it comes to achieving flat response in my somewhat large room. The speakers are rated at 28 Hz-32 kHz (+/-3 dB), so they are certainly capable of delivering serious low end response. However, the laws of science be a harsh mistress. The SVS PB-1000 subwoofer helps pick up the slack when it comes to exciting the room in the lowest frequency ranges. Crossed over at 50Hz, the sub helps bring up the levels and results in a flatter curve; but the 80Hz point is really tough to beat, though some may deem it a little high for use with these particular speakers.

Now it is time to dial in the Double Bass settings to see if I can get a roughly comparable curve. The subwoofer has it's own dial for setting the crossover point. The filter can be set anywhere between 50-160Hz. For the first test, I ran the front stereo speakers at full bandwidth with Double Bass engaged and the filter on the subwoofer set to it's maximum height (160Hz, or "disable/LFE").

Stereo speakers full bandwidth, double bass active, highest crossover setting on subwoofer (160Hz/disable)
It's a start, but that's a nasty trough at 60Hz. There's a slight peak beyond that, but nothing to be overly concerned about. Let's keep trying...

Stereo speakers full bandwidth, double bass active, lowest crossover setting on subwoofer (50Hz)
Turning the crossover dial all the way down to 50Hz yields a nice curve. There's still a dip between 50-60Hz, but it's not nearly as dramatic as the previous setting. I think we can do better than this. After all, we already have a good idea of what the appropriate setting might be given the positions of the speakers/subwoofers haven't changed, nor have I adjusted any gain levels.

Stereo speakers full bandwidth, double bass active, crossover at ~80Hz on subwoofer
Now this is much better. It ain't perfect, nor quite as smooth as the original setting with the receiver's bass management enabled, but it's fairly close. We still have the slight dip and peak in response in the 50-80Hz range, which could probably be targeted if I were so inclined to use some precision equalization. But let's not get carried away here.

Just for fun, Room EQ wizard can overlay all the response measurements for comparison:


The two outliers are the Full Range with no subwoofer active and the Double Bass with the highest crossover setting on the subwoofer curves. The other graphs are actually fairly similar, especially beyond the 60Hz point. Judging by these measurements, it seems that enabling Double Bass and running the speakers full range with the subwoofer's crossover setting at about 80Hz is a perfectly acceptable compromise. It's not ideal; I'd much rather have a true crossover between the main channels and the subwoofer, and hopefully this configuration is only temporary. I certainly do not want to hear that distortion ever again, and I certainly do want to continue using my subwoofer with stereo music. I mean, have you heard the new Dirty Projectors record!?

Update 

Upon further consideration and some general listening, the Double Bass option is untenable. While the graphs look good as far as room response is concerned, the sound is just too muddy. I guess I'll just run everything in Direct mode until I find another solution, or go back to the original bass management settings and switch it off if I begin to notice distortion. The full bandwidth option with no bass management is fine for piano music, and I've only noticed the distortion once or twice before on albums with full band arrangements (though I originally thought the recording was at fault, I have since confirmed Direct mode clears up the trouble spots I'd heard elsewhere prior to this glaring issue).

Update The Second

Tuesday, January 31, 2017

Vinyl Is Bullshit, or: How I Learned To Stop Worrying And Love Digital Audio

The truth: it is out there, maybe?
I am not an audio professional. I am interested in music and how it is delivered and reproduced. There was a time when I didn't care about such things. I bought compact discs. I made compilation tapes. I ripped discs, downloaded poorly encoded MP3s and burned CDs. Little attention was paid to sourcing in the early days of discovering and collecting music files via the internet. Over time, formats and bitrates became important.

Also along the way, I began to notice that some songs and albums sounded louder than others. Having tried my hand at some recording, I was proficient enough to be able to load these recordings into a digital audio workstation to analyze the waveforms. Taking this cursory peak behind the auditory curtain, I was thusly introduced to what is popularly knows as the Loudness War.

A history lesson is not necessary here, suffice it to say that through further mining of the netspace for audio goodies led me to what appeared to be the solution to the problem of excessive loudness in digital audio mastering: Vinyl. Or perhaps more specifically, vinyl rips. A vinyl rip, or needledrop, is a digital capture of the analog sound that is reproduced from a vinyl record. You hookup the output from your turntable and phonograph preamp to your computer and record it. Voila! You have captured the sound of your record. Some recent turntables have built-in ripping capabilities over USB, but the result is more or less the same (quality assurance notwithstanding).

First of all, why rip vinyl? Well, you may want to have a copy on your computer or portable device. Perhaps you'd like a digital copy of an old record you own that isn't available on CD. Or maybe you'd like to run the digital rip through a click-removal process that eliminates surface noise, providing you with a "clean" copy of a record to listen to anytime. Or perhaps you want a digital copy to play so that you don't wear out your record! Whatever the reason, vinyl rips are a thing and I got sucked into both collecting and making them.

When I discovered vinyl rips, I found the waveforms and amplitude measurements intriguing. They didn't look like the digital versions with all their peak limiting and dynamic range compression. Some rips sounded better than others, of course, as their quality depends on the skill of the ripper and the equipment used to create the rip. So it appeared that through ripping vinyl, one might gain access to a different master with healthier peaks and more dynamic range compared to the digital version of the same release.

But before going any further, it may be helpful to provide a few visual examples of what I'm talking about. This is what a loud digital master looks like:

"Drunkship Of Lanterns" from Deloused In The Comatorium by The Mars Volta
For comparison, here's what the same song looks like ripped from vinyl:

"Drunkship Of Lanterens" ripped by me from the Music On Vinyl reissue of the album

One need not be an audio expert to infer the difference. The CD version has uniformly loud peaks at the digital maximum. The vinyl rip looks "spiky" (the aforementioned peaks) and you can easily differentiate the "louder" sections of the song. It doesn't look an audio brick, or like too much meat is being stuffed into a sausage casing. When ripping vinyl, you have to set the recording level low enough so that the peaks to do not clip (i.e. reach a digital maximum level that cannot be exceeded, and so the wave is "clipped"). Without getting too much into it, this means that a direct listening comparison cannot be made between the two files because the volume levels are different. They need to be level-matched before you can switch back and forth between them to compare. If I lower the volume of the digital version to sound about as loud as the vinyl rip (lowering the levels of the file itself, not the volume control on your playback device) it looks like this:


Digital version level matched against the vinyl rip
I had to "turn down" the file by 8.3dB in order for the CD version to sound as loud as my vinyl rip. Only now can the files be listened to back to back for comparison without needing to change the volume on your player or device. There are various ways to level match various files, but this is my preferred method so I can see what I'm doing.

The point of all this is to get a sense of how appearances can be deceiving. So much conventional wisdom surround vinyl rips rests on the claim that a heavily limited and compressed digital master cannot be used to cut a vinyl lacquer. The medium simply cannot handle the levels without forcing your stylus out of the groove upon playback. Therefore, a more dynamic master with less peak limiting and compression compared to the crushed digital version must be used for the vinyl release. And this is what we appear to see in vinyl rip waveforms and amplitude measurements compared to the CD release of the same album.

I accepted this. I thought when I bought an album on vinyl I was necessarily listening to a different master entirely. As such, I was willing to overlook the medium's flaws (more on this later). I was confident in the belief and am now embarrassed to admit that I spent little time seeking out potentially disconfirming evidence. When presented with a commonly linked to page about vinyl myths, I was dismissive. That vinyl was sourced from a "better master" just had to be true. I mean, I could see the waves and dynamic range scores for myself!

Then I decided to be more critical. There wasn't a specific moment when this shift occurred, though it was certainly around the time of the release of the deluxe vinyl version of In Colour by Jamie xx. The pressing was a mess and I returned my copy, not even bothering to get a replacement. I haven't purchased a vinyl record since. Instead, I started looking for evidence that vinyl was cut from a different master than what was used to prepare the CD. Except for a few specific releases (like this one), there just isn't a lot of evidence to support the claim that wholly different masters are used for modern releases on CD and vinyl. What you think you're hearing or seeing in the waveform analysis is not evidence. And then I stumbled upon this thread.

It is my view that for most albums today, one master is prepared. This master is used for the various delivery formats, except the program material is further processed for vinyl by the cutting engineer. This does not constitute a "different master" in the way I am talking about it. What we have here is, in effect, the CD version being tailored to the vinyl format. And this tinkering and processing is why we end up getting dynamic-looking waveforms and measurements from vinyl rips.


I will provide quotes from relevant sources about the kinds of tinkering involved in getting music to play back properly on vinyl, but perhaps the most important point for our purposes here is that sub-bass content is attenuated when the lacquer is cut. Very low frequency content is not vinyl-friendly, so it must be reduced accordingly (if not, there may be serious tracking errors). The amount of reduction depends on the source material and the decisions of the cutting engineer, but it is a brute fact that the deepest bass is a problem for vinyl. This is crucially important for understanding the differences between vinyl rips and digital releases of the same music. 

If you take a loud CD master and roll off a bit of bass using a high-pass filter, you get a dramatically different waveform. You also get a better "score" when measuring a peak-to-RMS ratio, a crude measurement for assessing dynamic range. What this analysis tool does is look at the difference between the peak level of an audio file and compares it against the average level measured in RMS (root mean square). This roughly corresponds to the "area beneath the curve" on a graph, so think of a "louder" recording taking up "more" area on a time vs. amplitude plot. 

So if there is less distance between the peak level and average level of a digital audio file (absolute vs average level), then the file is accorded a lower Dynamic Range Score. This is bad. For a sense of why peaks and average levels matter, take a look at this classic video:



When we understand that removing a little bit of deep bass, while not necessarily glaringly audible (those frequencies are relatively difficult to reproduce and perceive), affects the waveform shape and the average levels of a recording measured in RMS. This makes sense, since we are lowering the amount of content in a certain frequency range. And this frequency band tends to take up a lot of space on the graph, as it were. So lowering the bass for vinyl without changing the peaks, we get a higher peak-to-RMS ratio, or "better" dynamic range score.

Hopefully it is clear by now that the dynamic-looking waveforms and high Dynamic Range Scores that are purported to be evidence of vinyl being sourced from a different master compared to the digital release can be explained by a simple roll off in the sub-bass region (somewhere below, say, 50Hz). This filtering out of the deep bass energy can easily be done in a program like Audacity. All you have to do is load up a file and find the High Pass Filter in the list of effects and set your parameters. I will illustrate the steps using the same song as above:

"Drunkship Of Lanterns" from the CD, untouched
Turned the file down by 8dB for headroom so that it roughly level matches the vinyl rip
Applied high-pass filter at 50Hz shelf with 12dB slope
My vinyl rip, untouched.
Oh my, those last two images look awfully similar, don't they? There appear to be peaks in the high-pass filtered version that appear magically when a bit of bass is rolled off. But what about the Dynamic Range Scores that measure peak-to-RMS ratios?*
  • Original CD file: DR4
  • High-pass filtered: DR9
  • Vinyl rip: DR10
Voilà! By simply attenuating the deep bass we get dynamic looking waveforms and higher DR scores that match quite nicely. They also sound similar, except for the sonic colourations inherent to the idiosyncracies of vinyl playback. No separate master required. And this Mars Volta song is just something I cherry-picked for this post to illustrate the process -- I see this phenomenon wherever I look.

So why do the waveforms looks so different? Let's zoom in to find out...





By filtering out some of the bass the square tops of the waveform become slanted when represented visually. The uniform peaks have become less, well, uniform. We no longer see the trademark continuous block of sound that characterizes a "brickwalled" waveform. Without altering the true dynamic range, we seem to have created what appears to be a more dynamic-looking shape when zoomed-out. They may not line up exactly (the filter shelf and slope settings may have been different, there may be phase rotation due to the RIAA curve and other EQ adjustments, low-pass filters and de-essers may also have been used, there's crosstalk and other distortions in the cartridge, etc.) but the similarities are clear. Again, the filtered version of the CD sounds nearly identical to the vinyl version, except for poor high frequency response and image shift associated with vinyl playback. For a little more in depth discussion about why vinyl rips result in imaginary peaks, read this.

Let's take a look at another example. The vinyl release of Beck's album Morning Phase came with a digital download code for what is called "The Vinyl Experience EQ". This is, in fact, a vinyl rip by mastering engineer Bob Ludwig of a test pressing cut by Bernie Grundman. As per Michael Fremer at Analog Planet:
In addition, Bob [Ludwig] told me that he used his turntable, Dynavector cartridge and Manley Steelhead phono preamplifier to transfer a Bernie Grundman cut test pressing for the "vinyl experience" MP3 download that is included when you buy the LP. It's a good playback system, Bob told me, but contrary to what's been written online, it does not increase dynamic range! Bob told me that Chris Bellman was supposed to cut it but he was very busy and when Bernie heard the master he liked it so much he asked to cut it.
So what we have here is a very interesting case. The album was mastered to DR6 by Bob Ludwig. Bernie Grundman then cut the lacquer using Bob Ludwig's master. Ludwig then ripped the test pressing using his own equipment, resulting in a vinyl rip that measures DR9. So without adding any dynamic range, we've turned a DR5 master into a DR9 vinyl rip, just through EQ and filtering during the lacquer cutting stage. Here are some images using "Heart Is A Drum" that follow the process I've been using above (and yes, they sound very similar apart from EQ changes and anything imparted on the rip by the equipment used):

 
 

A proponent of vinyl rips on a music site posted a CD vs Vinyl comparison image for a song from The King Of Limbs by Radiohead. They cleverly added an image of Thom Yorke in apparent distress over the loudness of the digital file. I added a lower third to the image with my high-pass filtering trick and a picture of Thom that I think is appropriate:

[caption joke not found]

The lesson here is that using waveform plots and dynamic range scores to assess the "quality" of a "vinyl master" versus the digital release is flawed. Claims of vinyl's superiority based on the dynamic-looking waveforms and DR values do not hold up to scrutiny and can be the result of simple filtering at the cutting stage rather than evidence of a wholly different master being employed.




I hope I have sufficiently articulated my point about what might explain the differences between digital releases and their corresponding vinyl rips. Again, without clear evidence to the contrary for specific releases, I think this applies to many of the records being cut and pressed today. An album is mastered, and files are sent along to a cutting engineer who then processes the mastered files for vinyl. 

But vinyl is bullshit in other ways. Through the course of my readings on the subject, I've collected a few articles that I have found particularly interesting and informative. This is how I learned about what kinds of processing goes into creating a vinyl cut, and why the high-pass filtering is so important to understanding how a vinyl rip looks and measures. I didn't just make this stuff up, I swear! The reason I got into vinyl (and I suspect the same may be true for many others) was because I had become convinced vinyl was the solution to the Loudness War. It is my view that this is not, in fact, the case. In light of this, the myriad well-known problems with the vinyl medium are no longer forgivable.

Therefore, the processing that goes into cutting the lacquer and the distortion inherent to playback are the real reasons why vinyl is bullshit:

The Signal Path of Shame
Assuming the use of plain stereo, the job of the hi-fi system remains simply to ensure that the two channels that the record producers have created for us make it into the room as close to intact as possible. When this is done properly it can sound pretty spectacular.

However, at every stage in the recording and replay process, distortions are added to the signal. It would seem fairly straightforward that we would like to minimise those distortions. But while audiophiles and the hi-fi industry may pay lip service to this idea, they then simply rip it up and start from the position that pimped-up copies of technology from circa 1950 are the best way to achieve this. This ancient technology contributes quite a few distortions that modern technology would not. There is no prospect of fundamental improvements.

Here is an incomplete list of those distortions:


Noise and distortion


  • basic noise floor of -70dB (A-weighted) if we’re lucky – stylus scraping along a groove in plastic, vinyl has finite grain size
  • pops and clicks: scratches and dust
  • electrical hum and noise: cartridge produces a tiny signal, and high gain pre-amplification is needed
  • rumble: bearings, motor
  • warped records cause various problems
  • stylus wear
  • stylus contamination: dust, dirt, vinyl particles
  • stylus misalignment – may vary as arm moves across record
  • record wear
  • record contamination: dust, dirt, vinyl particles
  • fundamental limitations in linearity of vinyl cutting/replay system
  • diameter loss: speed of groove decreases throughout LP, increasing noise and distortion and reducing upper frequency response
  • pre-echo: adjacent groove modulation
  • microphony: sound from speakers feeds back into the pickup
  • Channel separation: varies with frequency and typically only 20-30 dB at maximum
  • Record may be pressed towards end of life of the stamper, resulting in increased levels of various distortions

Arbitrary processing needed for vinyl mastering


  • compression (raises the quietest sections in volume to make them audible above the background noise, reduces the loudest sections to economise on groove spacing)
  • de-essing (reduce treble response for high amplitude, high frequency sounds)
  • mixing stereo bass to mono (otherwise the needle jumps out of the groove)

Wow and flutter


  • off-centre pressing
  • motor speed, belt etc.

Inaccurate frequency response


  • RIAA record and/or playback curves are often only approximate
  • cumulative effect of factors above causes imprecise frequency response (arbitrary processing when mastering, diameter loss etc.)

Why CDs Sound Better Than Vinyl
When a piece of program enters the cutting chain, it gets split to two different places. One split goes to a level attenuator, some filters, an elliptical equalizer, and ends at a very rudimentary and basic computer. The computer tells the lathe how far apart to put the grooves. The second goes to the attenuator, the filters, elliptical equalizer, a high frequency limiter, then the cutter head, which cuts the actual groove in the record. The groove, if looked at under a microscope is actually a complex sine wave. There are variations in depth, it is not straight (it’s actually quite wavy), as well as variations in the width. All of these variations are program dependent. If an experienced cutting engineer looks at a groove under a microscope, he (or she) will have a pretty good idea as to what is happening in the music at that particular spot. The groove needs more room to go back and forth the louder the program is. The longer a record is, the lower the volume will be to accommodate the longer grooves. The more bottom end a piece of music has, the deeper the groove needs to be. Filters are usually put in around 35 Hz, but can go much higher for longer sides [emphasis added]. Finally the more stereo a track is, the wider the groove has to go. It’s actually a V shape and the left and right sides of the audio are on each side of the V, with the center being the point. The wider the stereo, the wider the V needs to be. The elliptical equalizer will take the program and mono all the signals below a certain frequency. Stereo bass can be a disaster to cut, as can any out of phase program. The Neumann electronics (the industry standard) are preset at 150Hz and 300Hz. Cutter heads also have a huge problem with high end. Most engineers will put a high frequency filter in the program as well as use a pretty aggressive de-esser to prevent any problems. Another physical limitation of the medium is “inner diameter distortion.” As the record needle travels toward the center of the disk it becomes more difficult to reproduce high frequencies. The frequency response of a vinyl disk is drastically different at the outer section than the inner section. Cutting vinyl is a constant compromise.
 Loudness Wars: Is Vinyl The Answer?
It took a little bit of research and some confirmation from other mastering engineers to find that earlier vinyl was mastered for vinyl, but new vinyl was mastered for CD and then modified for vinyl... Excessive high end or low end on the lacquer master wreaks havoc with playback. A high-pass filter is added to keep the grooves from getting too wide during cutting. All that clipping creates false harmonics, so a low pass filter is added as well. There's still a little too much bass, so the low band is reduced slightly. All this filtering shifts the spectral balance and causes phase rotation... What's commonly called an "acceleration limiter" is added to tame sibilance and other high frequencies to avoid distortion. If there's too much stereo spread in the bass, that can be reduced as well. ...The analysis tools have been fooled into thinking there's dynamics where there aren't.
When your music arrives at the plant, a mastering engineer (in-house or outside) first creates an acetate lacquer with grooves on it from your master. This mastering does not produce a straight 1:1 copy; limitations to the vinyl medium require lots of processing before the signal can be sent to a lathe for cutting. ...
You can get away with a lot of stuff pressing a CD that just won’t work cutting an LP. If relatively inexperienced operators add anything to the stereo mix that is not perfectly balanced between channels, they can cause channel imbalances that wouldn’t even be noticeable on a CD pressing, but which can make it much more difficult to cut an LP. If you use an equalizer on both channels and have a slightly different setting on one of the channels, you wind up adding level to the L–R difference signal, making it that much more difficult to cut for that reason.

This is why the LP mastering guys have equalizers and compressors with calibrated detents. They also have phase meters which provide an actual plot of the phase differences between channels on an oscilloscope screen, and often some gadgets to work around these problems (like crossover networks that allow them to collapse the whole bottom end to mono if necessary, and compressors like the Fairchild 670 which can compress the mono and L–R channels separately).
 Mastering For Vinyl
There is a lot of poking and prodding that is often done to get the stereo signal to fit into place, because the LP has less information on it than the original master tape does. Often, you’ll see mastering engineers roll off a lot of the very low bass and add a false bass peak around 200 Hz or so, just to compensate for the mechanical limitations of the equipment. The other alternative is to reduce the running time per side radically. ...
There is a lot of poking and prodding that is often done to get the stereo signal to fit into place, because the LP has less information on it than the original master tape does. Often, you’ll see mastering engineers roll off a lot of the very low bass and add a false bass peak around 200 Hz or so, just to compensate for the mechanical limitations of the equipment. The other alternative is to reduce the running time per side radically. ...

The number two rule is always try to mix sources with heavy bass content into the middle. If there is a difference between channels on a signal with a lot of bass, you get bass in the L-R channel and the groove on the record becomes very shallow at points. It will have a tendency to pop out of the groove, unless the mastering engineer either limits the bass or reduces the level. If you want the disc loud, keep all the bass in the center. ...

There are a lot of judgment calls in the LP mastering and pressing process because there are a lot of variables that have audible effects. This is a real shock to someone who is used to cutting CDs, where one can send off a tape with directions to cut it flat and expect something back from the duplication house that sounds identical.

Why CDs May Actually Sound Better Than Vinyl
In 1968, a 23-year-old audio engineer named Bob Ludwig at New York's A&R Recording was asked to create a test pressing of The Band's debut, Music From Big Pink, so that the producers could hear what it would sound like on LP. During the process, he especially tried to preserve as much as possible of the deep low end of the band's sound, which he believed was critical to its music.

But when he heard the final LP that was released, he was stunned. "All the low, extreme low bass that I knew was there, was chopped right off." ...

Of vinyl's inherent deficiencies, reproducing bass is one of its most glaring. The other is that the last track on each side of a record sounds worse than the first, due to the fact that the player's stylus covers fewer inches of grooves per second as it gets closer to the center. ...

As labels seek to capitalize on a physical medium that is gaining momentum, some marketing efforts offering superior sound are downright misleading. Most notable among these is "audiophile-quality 180-gram vinyl," which consumers assume is superior because it is heavier. [Mastering engineer Pete] Lyman, however, says the added weight offers no musical benefit at all.

"It increases shipping costs and sales cost of the record. That's about it," he says. "It's the Super Big Gulp of vinyl, but you're not getting more [sound quality], really, you're just getting more vinyl."

Sound Reproduction: Loudspeakers and Rooms
Auditory masking is a natural perceptual phenomenon, operating in live concert situations just as it does in sound reproduction. It has assisted our musical enjoyment by suppressing audience noises during live performances and, over several decades, by rendering LPs more pleasurable. If we talk here about compressing data, it would be fair to say that LPs perform “data expansion,” adding unmusical information in the form of crosstalk, noise, and distortions of many kinds. More comes off of the LP than was in the original master tape. However, because of those very same masking phenomena that allow perceptual data reduction systems to work, the noises and distortions are perceptually attenuated. So successful is this perceptual noise and distortion reduction, that good LPs played on good systems can still sound impressive. ...
While one segment of audiophiles has reverted to LPs and tube electronics for stimulation, another component looks to expanded digital data space to provide enhancements to the listening experience. ...

Advocates of the new formats are probably unconvinced by scientific evidence suggesting that the expanded amplitude and bandwidth dimensions are not audible in music in the form it is delivered to consumers. If there are advantages to such systems in the recording studios, minimizing degradations in multiple layers, and generations of processing and mixing, that is a separate matter entirely, but that is not being raised as the prime argument. ...

I recall testing phono cartridges as part of the effort to improve playback quality. This exercise is substantially a test of test records, which really is a test of the entire LP mastering, pressing, and playback process. I participated in the creation of a test record. It was impossible to replay from an LP the signal that was delivered to the (carefully chosen) mastering lab. When using pure tones or bands of noise, distortions were easily measured and easily audible. They registered in whole (sometimes high) percentages much of the time, and this applied to both harmonic and intermodulation versions.

 tl;dr : When you buy a record, you are almost certainly buying an adulterated version of the CD. And then you are adding further distortion upon playback. If you care about accuracy in sound reproduction, you would do well to avoid vinyl and learn to love digital audio. 

None of this is to say that you should never listen to music on vinyl. That would be a silly claim. If you likes it, give 'er. There are plenty of albums that aren't available digitally. Some digital releases of older albums may have been converted from worn out master tapes, so earlier vinyl cuts may have been sourced from healthier tapes. And there's always euphonic distortion, where all the extra stuff that gets added to the sound when playing a record is actually pleasant.

I'm also not saying that vinyl is never cut from a wholly different master these days. It just seems to rarely be the case. Mastering is a not insignificant cost in record production. If a master (or remaster) is prepared specifically for a vinyl release, this may be something a record label may want to advertise with a hype sticker on the shrink wrap. Otherwise, check liner notes and discogs for mastering credits. When in doubt, you can always try getting in touch with the mastering or lacquer cutting engineer on twitter or via e-mail.

There are reasons to prefer vinyl in some specific cases (subjective/delusional preferences notwithstanding). But from the perspective of accuracy in sound reproduction, the evidence suggests that delivering music to consumers digitally is always going to win over vinyl. Even if your favourite band recorded an album on tape, and the mixing and mastering was all analog, and if a "flat cut" was done to prepare the vinyl release, one would still be better off with a digital transfer of said analog master. If your aim is to reproduce recorded sound with as little distortion as possible, a properly implemented digital system is the way to do it. Once an album is finished and is ready for delivery to consumers, a digital copy is the way to get closer to the original artistic intent.

Addendum

Here it may be prudent to provide some other examples of digital releases and their vinyl counterparts. First, let's take a quick look at a confirmed example where the vinyl was cut from the same "clipped" master from which the CD was prepared.  The following is from the album In Your Honor by Foo Fighters. The first image shows the DR6 CD file (RMS-matched to the vinyl rip 'loudness'), and the second shows the DR11 vinyl capture. So we can see how a digital master with typical dynamic range compression is magically transformed into a more dynamic-looking waveform in the vinyl capture.

CD master RMS-matched to vinyl, measured at DR6

Vinyl rip, measured at DR11

Next, here is the title track from Icky Thump by The White Stripes. The vinyl edition was given a separate master directly from the "straight from a one inch master mix." The CD file measures DR7, while the vinyl rip comes in at a whopping DR16. These are actual factual dynamics you can really hear. But releases like this are very rare, with another notable example being Consolers Of The Lonely by The Raconteurs.

CD files RMS-matched to vinyl 'loudness', measured at DR7

Vinyl rip, measured at DR16


_________________________________
* Original CD:

foobar2000 1.3.8 / Dynamic Range Meter 1.1.1

log date: 2017-01-31 18:20:14

Statistics for: 05-Drunkship Of Lanterns
Number of samples: 18783660
Duration: 7:06

Left Right
Peak Value: 0.00 dB --- -0.09 dB
Avg RMS: -6.56 dB --- -6.53 dB
DR channel: 4.24 dB --- 4.12 dB

Official DR Value: DR4

Samplerate: 44100 Hz
Channels: 2
Bits per sample: 16
Bitrate: 1015 kbps
Codec: FLAC


High-pass Filtered:


foobar2000 1.3.8 / Dynamic Range Meter 1.1.1
log date: 2017-01-31 18:19:42

Statistics for: 05-Drunkship Of Lanterns
Number of samples: 18783660
Duration: 7:06

Left Right
Peak Value: -3.62 dB --- -3.57 dB
Avg RMS: -14.86 dB --- -14.85 dB
DR channel: 8.93 dB --- 8.69 dB


Official DR Value: DR9

Samplerate: 44100 Hz
Channels: 2
Bits per sample: 16
Bitrate: 1411 kbps
Codec: PCM


Vinyl Rip:

foobar2000 1.3.8 / Dynamic Range Meter 1.1.1
log date: 2017-01-31 18:19:59

Statistics for: 05-Drunkship of Lanterns
Number of samples: 40887754
Duration: 7:06

Left Right

Peak Value: -2.86 dB --- -2.93 dB
Avg RMS: -14.68 dB --- -15.07 dB
DR channel: 9.44 dB --- 9.88 dB

Official DR Value: DR10

Samplerate: 96000 Hz
Channels: 2
Bits per sample: 24
Bitrate: 2976 kbps
Codec: FLAC