January’s Industry Update included a report on a scientific article presented at last year’s AES meeting, in which the authors used test tones and a modest audio system (albeit in an anechoic chamber) to prove that listeners can discriminate between high-rez and CD-rez audio. This is important because scientific evidence of an audible difference between high-rez and CD-rez music is considered weak by some, even as anecdotal evidence grows stronger by the day.
The response to my Update—in the article’s online comment thread and elsewhere on the web—was vigorous. Some accepted the result; others attacked it in varied if mostly predictable ways. In the real world, the noise floor will be too high to hear what they heard, said some. People don’t listen to test tones, said others, so what difference does this result make? Here’s the answer to that question:
It shows that high-rez and CD-rez audio are audibly different. If test tones sound different, then it’s completely plausible—even likely—that music sounds different, too, although differences are likely to be harder to detect with complex stimuli like music. A preference for high-rez music becomes a more defensible position, from a scientific perspective.
As I pondered this, I recalled a recent paper I’d seen in the Journal of the Audio Engineering Society but hadn’t yet read. “High Resolution Audio: A History and Perspective,” which the AES has made available free online, does precisely what the title says: reviews the history of digital audio beyond CD-rez and frames the issue of high-rez audio’s audible superiority on the basis of the available evidence.
If you’re up to speed on the issue, little in the article will surprise you, but a few ideas are worth recounting.
Some of those who reject the idea that people can hear differences between CD resolution and higher sampling rates argue that, because people can’t hear frequencies above CD’s Nyquist frequency (22.05kHz), the presence or absence of those frequencies on an audio recording cannot possibly affect what people hear. But, as author Vicki Melchior explains, that’s not really the point.
“A frequent misconception is that high data sampling rates assume the audibility of frequencies above 20kHz,” writes Melchior, who has a PhD in biophysics from Yale, runs her own consultancy focused on audio DSP, and has served as vice-chair or chair of the AES Technical Committee on High Resolution Audio since its inception in 2000. The idea that direct perception of ultrasonic sound has “a role in normal audio listening has been rejected since about the late 1990s (due to lack of evidence) in favor of the ideas discussed in Secs. 4.4 and 4.3,” Melchior writes. Section 4.3 is about changes in the dynamic range when the bit depth is changed.
Section 4.4, “Filtering and the Time Domain,” gets to the crux of the issue: whether “filters with shorter impulse responses, gentler transition bands, and less ring”—all of which are facilitated by higher sampling rates—”might sound better than steep brickwalls.” That possibility was suggested as early as 1984, by R. Lagadec and T.G. Stockham, in a paper establishing that pre-echo is clearly audible when fast and slow filters are compared (and when the cutoff frequency is well inside the audible range). “The steep roll-off filter produces a clearly [sic] ringing sound,” Lagadec and Stockham wrote. “The effects which are present in such an exaggerated form in the experiments above also exist, though on a smaller scale, in today’s digital audio systems. To which extent they are objectionable is still unknown, as the conceptual framework for such investigations has only now been established—a few years, unfortunately, after the promises of ‘perfect digital sound’ have first been uttered.” Twenty-nine years later, we’re just getting started.
Melchior concludes: “Current thought regarding the relationship of sonic transparency to high resolution is that ultrasonic frequencies are not involved in normal music listening, that the effects are not due mainly to lower hardware distortion, and that, besides dynamic range, the most likely issues are the anti-alias and anti-imaging filtering chains used to implement classic Shannon sampling.” The main evidence that such filtering chains are responsible for the alleged sonic differences is the fact that “a broad group of design approaches that reduce time dispersion are reported to improve transparency compared to the sharp brick-walls typical of CD.” “These filtering ideas,” Melchior concludes, “are amenable to scientific testing but this has only recently begun.”
Skeptics of high-rez audio, and of high-end audio in general, should note that Melchior’s article, like many others in this field, cites, in addition scientific evidence—with its listening panels, blind tests, and rigorous statistics—a second kind of evidence: the judgments of recording engineers and other audio professionals. “[M]usic professionals with access to first generation data have widely reported subjectively better sound” with high-resolution audio, Melchior writes. In the world of perfectionist music production and consumption, such opinions constitute real evidence—not decisive, certainly, but worthy of consideration and respect. Given that the subjective judgments of those same people profoundly affect our industry’s main focus—how recorded music sounds—how could it be otherwise?
Click Here: Liverpool soccer tracksuit