The Skeptics Guide to the Universe has gone silent in response to my petitioned request to offer counterpoint to their podcast in which they asserted that anyone promoting 192k hi-res digital audio is promoting pseudoscience. Their regular guest, Rebecca Watson, says that she is not an audio engineer, which is readily apparent by her hook-line-and-sinker acceptance of, and reliance on, Xiph.orgs assertions about high sampling rates in an article called 24/192k Music Downloads are Silly Indeed” by the somewhat now renown Monty. Rebecca Watson goes on to assert that its not just that 192k is useless, its actually bad, and bases this assumption on the inability of some electronic equipment to handle higher frequencies without causing distortion. This distortion is common in a great deal of audio equipment, but that distortion is not inherent to high sample rates, as we shall see. The host then asserts that 44.1k is actually perfect, because it covers the range of human hearing - another very misleading statement that was made without further investigation. Carrying their assumptions forward, this group goes on to say that, This is whats taking the audio world in the world of wine, and continue with the idea that the unwashed masses are going to be dictated to by an elite few who will tell them whats good. (Given that their podcast came on the heels of Neil Young announcing his PONO system, which is designed to play 192k digital files, it is clear that they are, to a degree, targeting Neil Young’s assertions about the benefits of 192k audio, and they say as much.) This assertion that audio has “entered the world of wine,” of course, leads this small consortium to what I consider to be a classic skeptical adoration of the Conscious Blind ABX Tests, in which people cant make out the differences between audio playing at different sampling rates. (The sources for these tests were not given in the podcast, and we’ll deal with that below.) Finally, like any well written essay, the podcast leads the listener back to the initial idea that all of this talk of 192k audio is just “pseudoscience.” Even after presenting over 25 signatures from established audio designers, audio journalists, mastering engineers, and others with expertise in professional audio asking Dr. Novella to allow me to offer counterpoint on his show, Dr. Novella has stopped answering my emails.So, I am going to use this humble forum (not my usual megaphone at Tape Op Magazine because, ultimately, this topic doesn’t have to do with actual recording work) to offer my counterpoints to this podcast. My hope is that that Dr. Novella, Rebecca Watson and their associates will being to consider that they are relying on are a number of assumption and misunderstandings about real-world implementation of digital audio. I hope that they will soon join me in truly remaining skeptical about digital audio and higher sampling rates, because it is precisely a lack of skepticism about the core assumptions around which early digital audio technology was built that has led so many to prematurely conclude that we arrived at digital audio perfection when we settled into the Red Book 16bit/44.1k standard. And so I humbly offer my counterpoints. Counterpoint 1 - Digital Audio Equipment Must Be Well Implemented. The reproduction of ultrasonic frequencies will produce undesirable artifacts in the audible range when (a) the equipment used to amplify and digitize those frequencies during the analog-to-digital process produces those distortions, thus creating a digital file that includes the distortions, and/or (b) when the digital-to-analog conversion and amplification of a file produces those distortions. It is wrong to say that 192k is actually bad.” It is right to say that 192k (or any sampling frequency) can be poorly implemented. Lets work from analogy: my Toyota will perform very badly at speeds over 100mph (shimmy, shake and handle horribly), but my (imaginary) Ferrari will perform best and is optimized for speeds around 100mph (suspension is rigid for such speeds, brakes, gearing, etc...ready to perform at those speeds). 100mph is neither good nor bad. The cars just perform differently at 100mph, my Ferrari better implemented for that speed of travel. We are just now getting to the point where digital audio equipment is being optimized for high sampling rates like 192k. I have done tests with different converters and, yes, some not-so-old designs sound horrible at 192k and some newer converters can actually sound their best at 96k and 192k. These are measurable and audible differences *in the performance of the equipment*. One can not conclude that “192k is actually bad.” One can say that lots of, but not all, audio equipment introduces audible distortion at 192k, and we are on the cusp of having relatively affordable equipment on the market that is well implemented for performance at 192k. Counterpoint 2 - Phase Distortions, Linearity and Non-Linearity Theoretically, a reconstituted sampled waveform will be identical to the original analog waveform. However, because theoretical implementation is impossible, the reality of digital converters is that they must employ lo-pass filters *in the analog realm*, and analog filters can not be phase linear. In actual implementation of digital converters, the required filters to eliminate frequencies that are above 1/2 the sampling frequency, will “scramble” the arrival times of waveforms within the audible band. This scrambling is called “phase distortion,” and has been known since the very early years of digital audio to have accounted for the harsh sound of the higher frequencies in digital systems operating at 44.1k. There are different ways to deal with phase distortion caused by the filters in converters, and the most common way now is to use oversampling techniques at both the input and output of a converter, allowing for more “relaxed” filter slopes set at very high frequencies which cause less phase distortion in the audible range. In a sense, oversampling is like “faking” high-sampling rates in order to benefit from the inherent advantage of pushing the anti-aliasing filters to higher frequencies so that phase distortions are continually moved out of the audible range. Another solution is simply to run the whole system at higher sampling rates (this is what PONO proposes, for example). A reduction in phase distortion is likely why *well implemented* 88.2k audio was found to be consciously distinguishable in ABX Tests from 44.1k audio in this study published in the AES Journal (aes.org/e-lib/browse.cfm?elib=15398). In essence, every time we double the sampling frequency, we push the phase distortion up by an octave, and the filter artifacts in the audible range are also pushed up an octave. When well implemented, digital audio systems operating at 192k minimize phase distortions and associated artifacts in the audible range. (Because there are other solutions to this problem the topic is more complex than this, but this explanation should suffice to counterpoint the assertions made in the podcast.) Counterpoint 3 - What Equipment Were Those Tests Done On, And Who Did Them? Because we are arriving at a moment in which digital audio technology is finally being well implemented at higher sampling rates, it might be time to return to the Conscious Blind ABX Tests on which the early assumptions about human perception of audible differences between different sampling rates were done and do them using different equipment. One of the big problems with relying on non-refereed sources like that at xiph.org - and especially other blogs who have conducted ABX Tests via the web (just google Blind ABX audio test - yikes!) - is that we don’t know for sure that they’re using best practices, or even adequate equipment, with which to conduct their tests. Hopefully, as skeptics, Dr. Novella, Rebecca Watson and associates will turn to refereed journals for their scientific evidence, rather than un-cited sources from unrefereed publications. And, let’s stay open to the notion that as digital audio technology continues to improve that we may need to revisit older tests conducted using less developed technology. Counterpoint 4 - The Nyquist Theorem Doesn’t Tell Us About Real-World Implementation of Digital Audio Systems. It’s important to remember that the Nyquist Theorem does not tell us whether, how, nor why the presence and behavior of frequencies above 20khz affects the behavior of frequencies below 20khz in real-world digital audio systems. Monty seems to have jumped from the Nyquist Theorem to assumptions about real-world implementation of high-end digital audio systems. Hopefully a good skeptic will be on the alert for these jumps, and I ask that Dr. Novella and his associates take a closer look at that. Counterpoint 5 - Very Few Are Claiming Publicly That People Can Hear Above 20khz. And Neil Young isn’t either. More and more we are coming to realize the role of frequencies above 20khz in the execution of well implemented digital audio systems. Affiliating one’s claim that well implemented digital audio systems running at 192k sound better with the idea that that person also believes s/he can hear above 20khz is merely an assumption. Many smart people with good understanding of these issues will tell you that hearing above 20khz is impossible but that implementing sampling rates that allow for the presence of frequencies above 20khz can affect what we do hear in the audible range. Counterpoint 6 - Audiophiles Are Not All Snake-Oil-Swallowing-Quacks I used to hold the same prejudice against audiophiles that was expressed in the podcast: the notion that they believe there are special people with special ears that can hear things most people can’t hear. As I’ve come to know more and more self-proclaimed audiophiles, however, I’m learning (as one tends to do when actually interacting with members of a subculture) that there is a great deal of variety in what audiophiles do believe about the role of ultrasonics in audio. In fact, audiophiles seem, as a group, to be more interested in the implementation of excellent audio systems than they do in the possibility of hearing beyond 20khz, and many audiophiles seem to have a cutting edge grasp of what goes on inside digital audio systems. Rather than rejecting this subculture as snake-oil-swallowing-quacks because a few loud members of that group go on and on about hearing ultrasonics, I think it’s far better to turn to this concerned and fairly well informed group for a better understanding of what is going on in cutting edge digital audio systems. It’s likely that we will find some highly skeptical allies with expert knowledge in the area of digital audio. I offer these counterpoints with the humble hope that we can all help each other in developing more incisive skeptical attitudes toward our world. I applaud both Dr. Novella’s and Rebecca Watson’s agendas and am, truly, on your side. Sincerely, Allen Farmelo
Posted on: Tue, 06 May 2014 16:09:09 +0000