Saturday, 7 May 2022

DEMO: Software de-clipping of dynamically compressed recordings (Red Hot Chili Peppers' Unlimited Love). On "Why you can't trust audio measurements". And "Those Obscure Objects of Desire" - Utility and Luxury.

Let's get right to the heart of the issue with the diagram above!

There have been discussions among audio sites like Audiophile Style and Darko Audio that this latest album from RHCP - Unlimited Love (2022, average DR7) - is their most "audiophile-friendly". 

Flea, the band's bassist, tweeted on April 1st (hope not meant to be April Fools' joke!):

“For you audiophiles out there, the new RHCP record is mastered directly from the tape we recorded it on, no computers, no lame compression or limiters”

In the image above, I extracted the last minute from track 1 ("Black Summer"), a portion of the music which is quite loud for a peek at the waveform. One look at the original data clearly shows that the music has gone through a compression step at some point in order to create that "flat top" DR5 in the image above. Flea is wrong.

Whether this happened in the recording, mixing, or mastering steps, who knows. The clear use of dynamic compression is present in both the 24/96 version and on the CD. Clearly, music like this does not benefit from the "hi-res" 24/96 version so I would recommend saving your money if you're tempted to purchase a download. As usual, be critical when buying hi-res.

You'll notice that below the DR5 tracing, what I have done is used iZotope RX 9's "De-clip" function to restore some of the compressed portions and expand the peaks. Here's the batch process which simply consists of applying -6dB to the signal (I used the 24/96 version of the song), and then one of iZotope RX 9's De-clip default settings to restore "Extreme Analog Clipping". I previously discussed doing this a few years back.

As you can see, the de-clipped version doubles the DR value to DR10 for that 1-minute segment.

We can overlay the original and de-clipped versions to see where the software detected compression and expanded those regions:


Pretty cool, eh?

I normalized the 2 tracks to the equivalent amplitude using Adobe Audition's ITU-R BS.1770-3 loudness meter at -12.25dB LUFS. Download and have a listen here:


As usual, I present these demo tracks as a short segment for testing and education only based on the principle of "Fair Use". Please delete the tracks when you're done testing and purchase the album if you like the music.

Do you hear much of a difference with the de-clipped track?

Personally, I find the difference to be present but subtle. Over the years, I have processed a number of albums this way but the "reconstituted" effect is not anywhere near as good as a proper studio mastering job. For example, the recent Tears For Fears demo using a higher dynamic range multichannel downmixed to stereo was much superior. Once the music has been squashed, as the saying goes - "the damage is done". There is loss of resolution, nuances have been compromised, even if nominally we can apply some de-clipping to create a higher DR version. As I discussed before, I suspect this could be what happens with vinyl as those flat tops go through adjustments (eg. summing low frequencies), RIAA EQ, vinyl cutting, and eventual playback from the physical medium to reconstitute something that measures with a higher DR but in essence probably was still the original compressed signal to begin with.

BTW, other software can also do this de-clip processing. For example, check out Perfect DeClipper. I ran this track through that software as well and similarly got a DR10 output using the default settings.

Well, at least Unlimited Love is an improvement over Californication (1999, DR4), Stadium Arcadium (2006, DR5), or another of the typical Vlado Meller hot mess mastering jobs like I'm With You (2011, DR4). Would have been nice if Flea were correct that this album had "no lame compressors or limiters" applied. (It's OK to have some compression... Just leave us a few more dBs, double digit DR would be nice!)

Alas, we have to dig back over the years to 1991's Blood Sugar Sex Magik (DR14) before you see an album from RHCP with a double-digit average. Incidentally, if we look at USA album sales figures, Blood Sugar Sex Magik was their biggest selling album at around 7.7M, followed by Californication at 5.9M. Since 2000 with digital downloads, streaming, and overall worsening dynamic compression, we don't see album sales like that anymore except with some of the biggest contemporary pop like say Ed Sheeran with >9M for x (Multiply) (2014) or Taylor Swift's 1989 (2014) with >10M.

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

I see that GoldenSound made an interesting video on some of the complexities of measurements here:


Good video. Yeah, when measuring, one could do all kinds of stuff like cherry-pick results from inter-test variations, numbers change with measurement bandwidth, weighted calculations (like dBA, dBC, dBZ) will change results, application of notch filtering can improve measured performance (on the 1kHz sine tone), FFT parameters matter, and load characteristics are significant...

Sure, measurements can be tweaked to increase or decrease the results of a particular device to a certain extent. Manufacturers can certainly be sneaky about presenting results in an excessively good light in their advertising for example if they choose to use unusual settings. Nothing new here, in fact if we look at advertisements over the years we see all kinds of stuff including graphs with vague scales or results with little relevance to audio frequencies (be careful of cable salesmen using these tactics).

Having said this, for those who have not done measurements themselves, let's just say that while there are many variables, it's not that complex either to the point where one needs to feel overwhelmed nor should this discussion trigger major trust issues. Parameters for measurements these days do follow a typical range of settings whether it's with Audio Precision or alternative gear. Have a look at the Audio Precision manuals online for some great discussions and guidelines if you're interested. Likewise, have a look at AES17 (here's a PDF to AES17-1998 r.2009, latest AES17-2020 can be purchased here) for formal specifications like THD+N measurements done at 997Hz, notch filter, bandwidth to 20kHz, etc... Those of us doing these measurements appreciate that slight changes such as using 1000Hz or a good ADC even without notch filtering can still provide very reasonable results better than the limits of human hearing; we'll get a chance to talk more about this ahead - see the preview image at the bottom.

My sense is that while the GoldenSound video is correct factually, the initial "tone" of the video seems overly dramatic with the scary title of "Why you can't trust audio measurements" as if there is anything surprising here. Anyhow, I assume the dramatic title was chosen as click-bait without anything all that surprising within the video contents; I see a lot of YouTube videos doing this or using some dramatic picture to catch the viewer's attention these days.

While anyone can make mistakes when doing measurements, the fact that many people are doing measurements means you'll see replication of results from around the world by folks acting independently. That ability to replicate, finding either similar or different results is part of the empirical nature of objective testing. Unless there are huge measured variations found for the same device (which would speak either to undocumented changes in specifications over time, or potentially poor quality control), a low-noise high-fidelity device will measure similarly using different test techniques. Different objectivists coming to similar conclusions using various instruments with various approaches isn't a bad thing so long as everyone understands not to directly compare numbers inappropriately using different methodology, and having many test samples will improve the "power" of the findings. This is how we empirically discern facts from opinions.

Comparatively, what is the "inter-rater reliability" of purely subjective assessments of sound quality? Certainly I've seen my share of wildly different opinions based on all kinds of subjective preferences. A few years ago (late 2013), I posted on the case study of the Wadia 121 Decoding Computer with subjective reviewer impressions among the magazines and online and then the subjective/objective review from Stereophile which showed anomalies and questioned the subjective accolades. So has anyone made a YouTube video titled "Why you can't trust audio subjective impressions"? This "confession" from Steve Guttenberg might be something of that sort.

I agree with GoldenSound that many DAC measurements these days, especially with AP gear, there is a focus on the 1kHz SINAD/THD+N measurement (not just the number but make sure to have a look at the FFT graph itself to see the pattern of the spurious noise and possibly hum). This makes sense as the simplest way to demonstrate non-linear distortions in the output of a high-resolution device; these non-linear distortions being the difference between the output and input signals, with output frequency components absent in the original.

Don't be too critical of the simplicity of the SINAD/THD+N though. Those harmonics are a reflection of non-linearities in the system and when we do a more complex test with the superposition of multiple frequencies, we end up with different patterns like intermodulation using two or more tones. While intermodulation is non-harmonic and potentially more audible, it's still part of the same non-linear distortion system captured in the SINAD/THD+N. Then there's the +N component which adds in the noise level and other spurious content.

As with any single measurement, it is important not to hang too much emphasis on just this result. Ranking DACs using this single metric is fine so long as we see it as a quick "screen" of potential resolution and engineered precision. IMO, it's always best to remember the limits of human hearing and cognition and contextualize these results. Make sure to appreciate that many of these quantities are already multiple times better than human hearing thresholds. For me, SINAD 100dB is more than good enough to enjoy the full beauty of music already. The fact that many DACs perform at least at this level if not better is a testament to the bountiful high-resolution DACs out there.

Distortion + noise are but a couple of parameters we need to be aware of and which can change with output level and frequency. The others of importance include frequency response (in DACs, variation here would include impact from filter settings) and timing accuracy (wow & flutter, jitter, phase). If all those parameters objectively can be shown to be "healthy", then I don't think there's anything to worry about in terms of sound quality. As usual, make sure to have a listen and enjoy the music. ;-)

Remember that nobody ever said human preferences need to follow strict "ideal" patterns like ruler-flat frequency response or zero distortions. There's nothing wrong with preferring the sound of a lower resolution device (like say a tube amp versus cleaner solid-state amp). Some amps sound great subjectively perhaps because of the distortions. Ruler-flat headphones can sound "boring" as well (I agree!). And so on and so forth. An objectivist is allowed to acknowledge these subjective idiosyncrasies without shame, and even with pride in understanding oneself!

Finally, I think the real power of objective analysis is that as audiophiles it allows us to say when devices basically make no difference. This is how we differentiate what's worthwhile from snake oil. For example, we can with good confidence say that cables make little difference outside electrical parameters that purposely distort sound. A computer server makes no difference when sending unaltered data to a good USB DAC for example. I see GoldenSound measures external DDCs - Digital-to-Digital Converters - which also based on the data and logically will make no difference worth spending much time or expense on unless there's some other function you need. Let's face it, the measurable differences in videos like this are audibly meaningless. I bet if a controlled, blinded listening trial were done, there would be no significance among these devices.

Yes, basically "Bits Are Bits", and jitter is boring to talk about these days with decent devices as it makes little audible difference. Unless there's good evidence otherwise, let's keep it straight forward everyone.

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



To end, I thought I'd address a recent article from Michael Lavorgna on Twittering Machines; entitled "Those Obscure Objects of Desire".

It's fair to say that since the days of the now-defunct AudioStream, my opinions on audio and the value of many audio "objects" appear quite different than what Lavorgna champions (some discussions here and here with Herb Reichert as the main author). Heck, I doubt we'd even enjoy the same albums or visual art pieces. (I trust despite the exchange of words over the years, this isn't mean-spirited and we can't chat over dinner and enjoy a beer talking about life and audio if the occasion ever arises!)

I don't think there's anything wrong with writers like Lavorgna equating luxury products like Patek Philippe watches and Hermès handbags with audiophile gear as in that article. However, I believe this kind of comparison acknowledging non-utilitarian functions of luxury products is incomplete and doesn't get at the heart of why many audiophiles are unhappy with what's often published. The issue is not that we must be cheap as audiophiles, the issue is about being honest with the consumer. With luxury products at luxury prices, I think we have to respect the basic understanding that:
The value of luxury products go beyond the utilitarian purpose. To argue that these products improve utilitarian goals primarily would be a fallacy.
What this means is that when Omega advertises their fancy wristwatch:


We appreciate that James Bond isn't saying something utilitarian like: "This is easily the most accurate watch I have ever used." Rather, the message is: "This is a very cool watch, and you too can be just as affluent and debonair as this handsome gentleman." That non-utilitarian "coolness" is really what one is paying for; even though I'm sure it tells time very well. The Omega Seamaster by the way is "only" around US$10,000 (many variants out there).

Let's look at a couple examples in the audiophile world.

Is the advertising message we receive from audiophile companies and reviewers like this one recently for Crystal Cables starting at a luxurious US$15,000/m for power cords, similar to the Omega watch?

No, of course not. They dare suggest that this luxury wire (which I agree looks great) also are "the best wires I’ve used or heard", that the "designer Edwin Rijnveld claims a universal improvement in performance", and that "Edwin and Gabi’s new cables, interconnects, and power cords are simply better than other wires I’ve heard — closer to the sound of the real thing, when the source permits."

The argument here is not that it's merely "cool" or handsome to look at, but that the utilitarian purpose - that of sounding even better - is central to the value of these wires. No evidence of course is ever provided in reviews like these that the sound is actually different compared to anything else; it's just faith that you accept the words of the "TAS Staff" that authored this or Mr. Rijnveld's claim that there has been a "universal" improvement (where?). Unlike the subjective eye-of-the-beholder judgment of how "cool" it looks, whether these wires actually transform the sound produced is not a subjective debate but one that surely should be measurable if it's true given the superlatives offered to the potential customer.

Obviously one could, and arguably even should, buy an expensive watch, fancy car, or très chique handbag if desired! That is the free choice of consumers. I champion the rights of individuals so long as the product isn't unethical or illegal in some way of course. To be clear, I also do not believe we must cater our purchases to only objective measurements. As I discussed above in the previous section, I'm actually quite happy with "merely" SINAD 100dB DACs, and non-flat frequency response might be preferable.

Let's use another concrete example to push the point. If an audiophile really wants to buy an expensive Taiko Audio SGM Extreme as discussed a couple weeks back, then sure, go for it! I think many audiophiles will love the non-utilitarian benefits it brings because it looks nice, weighs 100lbs, and many will be impressed by one's devotion to "the best" digital computer audio product on the market (allegedly).

Without evidence however, don't insist that Emile Bok is correct about "matched" RAM modules sounding better. That his special USB card makes a substantial difference. That he somehow has custom USB drivers that will make a sonic difference with bit-perfect data transmission. That he can hear a real difference from the DAC output between PCIe and SATA SSD storage. All such claims without any evidence are questionable based on basic understanding of computers and experience with digital audio. This is true for budget devices, and even more important when you're linking alleged sound quality benefits with very high prices! That again is the heart of the criticism; there appears to be a dissociation between the advertised reasons for why we spend money and what exactly we're getting in return. Snake oil could not perform miraculous cures as advertised back in the day. And likewise, it does not appear that throwing money at many of these luxury products will buy improved sound.

For a consumer to buy based on his/her desire is not a moral judgement for others to make (in this I agree with Lavorgna). If an audiophile friend calls me up one day and says: "Hey Arch, I got a Taiko computer in the system, ya wanna come for a listen?" He would still be a friend even if I believe he could build a superior fanless computer at 1/10 the price. I'd happily pack up some measurement gear and pop over to his place to have a listen and even test it out if he'd let me.

Moral obligation is an issue when manufacturers create claims to entice consumers, especially when they fly in the face of reality and reviewers amplify such claims to make money. There are laws and expectations of "truth in advertising". It's not OK to perpetuate lies and delusions as manufacturers. And it is important for reviewers to balance the needs of the Industry with those of hobbyists to maintain integrity. It would have been nice to see a little bit of that balance and "calling out" the manufacturers who spew nonsense back in the days when Lavorgna was writing AudioStream. Maybe doing that would have garnered more respect among hobbyists for that site. Things got even worse when Rafe Arnott took over both InnerFidelity and AudioStream between 2018-2020. Where are all those articles now? Will articles like "Those Obscure Objects of Desire" at Twittering Machines withstand the test of time, or yet another dead-end attempt to justify the unjustifiable? Think about it...

Happy month of May, dear audiophiles. I've got a few things I need to get done this month at work so might not be around as much. Speaking of THD+N/SINAD measurements, a picture of something I've been playing with over the last few evenings:

E1DA Cosmos APU, pre-production unit.

Hope you're enjoying the spring and the music!

13 comments:

  1. Did you record the tested file from the record (ie a black vinyl disk) or did you use downloads and CD only?

    ReplyDelete
    Replies
    1. Hi Edwin,
      This is the 24/96 hi-res download from HDTracks.

      I don't have a vinyl copy of this album (not really interested). I have seen some vinyl rips out there by others and "Black Summer" typically gets a DR10 result like the software de-clipped version in the demo.

      My suspicion is there's no special vinyl mastering used.

      Delete
  2. Interesting experiment! To me this software declipping is in the same vein as "euphonic enhancement" achieved by tubes. If you think about similarities: first, we add something that wasn't there in the recording, and second, we do it in a non-linear fashion (because manipulating the signal based on it's content is always a non-linear transformation). And, in both cases we do that to make the recording to "sound better." But--it's fun!

    On measurements--I think 's great that AP equipment have essentially become a standard. Now you can see lots of Chinese manufacturers who routinely use AP for R&D and QA, because there are lots of bloggers and thematic forums who also use AP for ranking these devices. As a net result, the quality of audio gear has dramatically improved.

    ReplyDelete
    Replies
    1. Thanks Mikhail,
      Yeah, good observation about the idea of "euphonic enhancement". Indeed, we can doing something here that wasn't in the original; in this case having the software analyze and extend those peak levels up by about +6dB. Yeah, non-linear extension as well. Whether folks hear this as "euphonic" I'll leave it to the "ear of the beholder"!

      Yup, I agree on the measurements and using that as feedback that ultimately improves the quality of devices like DACs. Overall, very good change to see in the last decade as the hobby evolves!

      Delete
  3. Hi Arch-
    Lavorgna gets credit for basically admitting with some of the expensive equipment you are paying for prestige/status and looks.
    Nothing wrong with that.
    Of course, he tends to review expensive equipment and always says it's better subjectively. If you show him terrible measurements (like with his beloved Totaldac units), he will say it's irrelevant.
    In audiophilia there's a massive FOMA factor. The audio establishment promotes that, and audiophiles will buy the next best thing or some expensive tweak just to make sure they have the best.
    I have NO issue with super expensive electronics, as long as people make no general claim for them having superior SQ, that is nothing outside of their sighted comparisons.
    I get that some people like to buy "the best" and then they actually do hear "the best" SQ when using it; but it It still amazes me how many audiophiles refuse to accept that it's just their expectation bias and that bias works on ALL of us.

    ReplyDelete
    Replies
    1. Hey there Unknown,
      Thanks for the comment. Yeah, I gather that's the role Lavorgna (and similar folks like that) plays in the whole audiophile machinery. The power of his opinions around prestige and status is only as good as his reputation as a reviewer. Not sure how strong his reputation is these days.

      A purely subjective "review" IMO basically cannot be held to account. For example, what accountability is there for Jonathan Valin (I see the Crystal Cable review is no longer written generically by "TAS Staff") to say that these are the "best sounding" cables he has ever heard? Wait a few years, another expensive cable company might send him another set of cables, and then we begin again with the new "best". And so it goes, never ends, never any feedback or verification of anything SQ-wise.

      I assume you mean FOMO (Fear Of Missing Out) as a psychological driver of these upgrades and tweaks?

      Right, there's expectation bias bigtime when we express opinions about these devices. A good idea to be mindful of this in all domains of life! It does take some humility to accept possibly being wrong and to acknowledge sometimes to not know or that our opinions may be inaccurate - especially with vague subjective ideas comparing otherwise hi-fi sound devices!

      Delete
  4. I would go further and say often people can be wrong about their own preferences. For example, everyone's preferred headphone target is some variation of Harman 2018, but some still insist on rejecting the Harman target, or worse, rejecting the use of EQ at all. In the case of distortion there might be some scope for individual preference variations though.

    ReplyDelete
    Replies
    1. Hey Kirby,
      Interesting comment especially on the Harman headphone curve(s). Likelihood is that listeners will probably enjoy the Harman results although I'm sure there will be folks outside the norm.

      Hopefully folks will have a good listen and then can say with authority whether they like it or not. I can imagine there are some contrarian "conformity is death" people who might just reject the norm because they're "special" rather than having actual knowledge.

      Delete
    2. Yes. Sighted bias and a contrarian impulse often combine to form a rejection of the harman target.

      Delete
  5. Another great article Arch!

    Wrt to your before and after diagram, there is not only dynamic range compression being applied, but also peak limiting applied. The latter is responsible for the clipped tops. A quick explanation here.

    I use the same tools as you to restore a bit of DR. In the case of RHCP's killer tune, "Black Summer" which I listened to the samples, one can hear a bit more "crack" on the snare drum and a slightly brighter presentation overall. But unfortunately, does nothing for the dynamic range compression.

    From the studio production side, compression/limiting can be applied on each track of a multi-track recording, before it is recorded (tracking) and after it is recorded (mixing). Sub groups, like the drums or vocals can also have their own set of compression/limiting and of course, during stereo mixdown, often a compressor/limiter is applied. This is even before "mastering"

    Given the number of opportunities to apply compression/limiting, each with their own threshold, attack and release times, and compression/limiter ratio, it is impossible to reverse these accumulated effects. While declippers can do an excellent job in reverse engineering some of the lost peaks, it can't undo the dynamic range compression.

    Mixing is an art using the science to produce a good sound. Unfortunately, almost all DAW's have factory presets like Mastering1 and Mastering2 where the mixer at a press of a button can engage sophisticated compression/limiting without understanding what is going on. Just sounds louder, which is what I want, but most of those presets are pretty "heavy handed" as there is still the belief that louder is better without technically understanding that it is actually the opposite: loud wimpy sound!

    Keep up the great writings!

    ReplyDelete
    Replies
    1. Thanks Mitch,
      Much appreciate the background from the studio production side. And good to hear about the complexity there as well with potential for compression/limiting to be applied across the process and the potential for subgroup application. Great point about how impossible it would be to reverse the compression in the context of all these variables. No matter what, it's gotta be done "right" in the studio!

      Scary though with presets to basically apply the same amount of processing universally. Seems way too easy to do this without mindfully taking the time to customize the sound these days...

      Thanks and looking forward to reading more of your work as well!

      Delete
  6. For declipping, I use Stereotool/Declipper. There are two sections that must be activated and set up: Declipper and Natural Dynamics. Through years and years of tweaking I managed to create a pretty much universal preset which decongests compressed audio quite nicely. Archimago, if you have the Stereotool VST plugin, try my preset: https://www.dropbox.com/s/s9wp6stpdnuhs0c/Stereotool_Mine.sts?dl=0
    I would be interested to hear your impressions.

    ReplyDelete
  7. Hi Αρχιμαγε!
    I listened to the RHCP 'Black Summer' samples, and i'm surprised to say that i prefer the original DR 5! Your DR 10 sample, sounds like a 'loose' mix, it's not 'tight' anymore,if i'm allowed the musicians terms.
    I believe the damage in the mix, can't be undone.
    My JRiver analysis, shows that both samples, have an R128 dynamic range of 1,8! I guess R128 says also many things about compression. I would like to hear more from you,or Mitchco about R128.
    The thing is that in my 24/96 downloaded files, 'Black Summer' has a DR6 and 10,8 R128!
    Cheers from Greece!

    ReplyDelete