SDMI Watermarks Tested In Nashville

Last year the music industry was jolted from its complacency by the rise of MP3, a scheme for the quick and easy transfer of digital audio files over the Internet. Legal attempts to block the format as a form of copyright violation failed, and the industry began scrambling to find a way to prevent the wholesale piracy of higher-resolution formats to come. The Secure Digital Music Initiative (SDMI), an alliance of more than 240 hardware, software, and music-publishing companies, has been working overtime trying to develop an unobtrusive technique for preventing unauthorized copying—something that digital technology is making easier than ever.

"Watermarking," the insertion of audibly detectable signals into recordings in order to track their origins, was one solution the SDMI arrived at. The concept of altering recordings in this way has enraged many audio purists, who have decried the absurdity of intentionally coloring recorded sound when the audio industry is on the threshold of a transition into the most transparent recording and playback technology in history.

Audiophiles' misgivings and cautionary complaints have not been lost on the SDMI's technical working groups, who have winnowed a dozen or so potential watermarking standards down to a final few. For the past couple of months, SDMI engineers have been conducting listening tests of these final contenders using music-industry professionals—recording and mastering engineers, producers, and even a few audiophiles—as test subjects to ascertain the audibility of the watermarks. If music professionals can't detect the watermarks, or at least aren't offended, the thinking goes, then the general public probably won't be either.

Testing has taken place in Los Angeles, New York, and Nashville—cities with high concentrations of audio experts. On Tuesday, July 20, I was one of a group (including Chuck Ainlay, a mastering engineer who works closely with guitarist Mark Knopfler and his band, Dire Straits) that participated in SDMI-administered watermarking tests at Georgetown Masters studio in Nashville. The studio is owned and operated by Denny Purcell, a legendary mastering engineer whose work has helped propel more than 500 recordings to gold or platinum status—as attested by the many awards that line the studio's walls. Purcell's hearing is so acute that he can identify which of three rough takes has a vocal that is "up one"—mastering jargon for a 1dB boost. Ordinary music fans generally can't detect a change in level unless the difference is at least 3dB.

The test site was Purcell's largest mastering suite, a solidly reinforced and acoustically neutral room on the top floor of the four-story building. Great effort has been taken to isolate the room from outside noises, while at the same time making it reveal as much as possible of the recordings played back in it. It is a space any audiophile would be happy with—as was the system used in the tests. The digital datastream---as selected by the test subjects---was routed to a dCS D/A converter (I neglected to note which model), and from there to a Pass Aleph P preamp, whose output fed a pair of Nelson Pass X1000 power amps, which in turn drove a pair of Nova Audio Applause loudspeakers through top-of-the-line Goertz cable. The system sounded very good. Test music was an unreleased Dire Straits track, "Shut Up and Deal."

Playback was via a pair of time-aligned and level-matched hard-disk recorders, whose outputs could be seamlessly switched at the test subject's whim by a handheld ABX switching box. A primary set of buttons let us select between "A" (always the unencrypted original), or "B" (always one of SDMI's encrypted versions), or "X" (which could be either A or B). A secondary set of buttons let us vote as to whether X was A or B. After voting, the source could be reset to either A or B for another trial. I always selected A as my reference and proceeded from there.

Malcolm Davidson, Sony Music's vice president for technology, presided over the testing and dutifully recorded the results. The genial Englishman is in the process of collating all the test results for the SDMI to make its final decision on which watermarking system—or systems—will be used. Davidson explained the operation of the ABX switcher and set the playback level as we preferred; otherwise, he sat back and took notes. At no time did he attempt to influence our votes. We were allowed to listen to each version as long and as often as we wished. Only when we felt ready did we cast a vote, then reset the switcher and vote again. Each of us voted multiple times on four different watermarks, a process that took about an hour per person. Davidson must have heard "Shut Up and Deal" several hundred times that afternoon. The man has infinite patience.

At the risk of undermining whatever credibility I may have as an audiophile, I admit that I could not identify the watermarked music. In the first three trials, I averaged 50%—the same as random chance. In the last one, I thought I heard some vague difference in the vocal and scored 62.5% correct—a better, but still statistically insignificant, result. Chuck Ainlay thought he heard something "phasey" happening in the fourth trial, but I don't believe he fared any better than I did in identifying the culprit. Denny Purcell admitted with a laugh that he scored "worse than random—only 37% correct" on his first attempt, but noted without irony that one of the engineers who had designed the watermarking scheme had been able to score 100% in multiple trials. Obviously, to spot the watermark you must know what to listen for. Without knowing, you might as well flip a coin.

Which leads to some interesting observations:
1) Despite the test subjects' expertise, for the purposes of this test we were all naïve listeners. All the sonic cues we've trained ourselves to listen for—dynamic anomalies, frequency-response incongruities, timing and phase relations, microtonics and imaging—were of no help in attempting to identify the watermarks, which rely heavily on psychoacoustic "masking."
2) The fact that several "golden ears" were unable to pick out a pattern superimposed on a high-quality recording means that SDMI is sincere in trying to make its copyright-protection technology as transparent as possible. Not only could I not hear whatever it was the encryption designers built into the music, there was nothing I heard that even remotely offended my audiophile sensibilities. The clean and unclean versions of "Shut Up and Deal" sounded equally good to me.
3) ABX testing—one of the most controversial and hotly debated topics in the audio realm's vast constellation of hotly debated topics—may hide as much as it reveals. Long-term listening in familiar circumstances might reveal details (as was proven with the subjective effects of jitter) that short-term listening does not. In which case, the SDMI may bring a technology to market that will later be discredited.
4) Audiophiles—a tiny slice of the overall music market—have little to fear from the music industry's watermarking effort. The whole affair is strictly voluntary, as SDMI has emphasized many times, and the many specialty record labels serving the audiophile community, such as Chesky and Water Lily Acoustics, will go on about their high-resolution business regardless of what decisions are made by SDMI.

I don't happen to share the animosity-on-principle toward watermarking that is so widespread among audiophiles. Neither am I the paid stooge of corporate Big Brother. I'm simply a somewhat neutral observer. My take on the whole situation is much like Denny Purcell's. He told me late that Tuesday afternoon that he is "philosophically opposed to the whole notion of watermarking," but quickly added, "as a practical matter, I recognize that it is a necessity." Big Brother isn't conspiring to deprive us of our rights to copy music ad infinitum. The music business is exactly that—a business—and must respond to the threats of scammers, pirates, and thieves.

While I am wary of efforts by big, well-funded organizations to control anything, my brief time in Nashville reassured me that SDMI really doesn't want to damage the music. That would be, ultimately, counterproductive. Watermarking as I heard it—or, more precisely, as I didn't hear it—was inoffensive and unobtrusive. The nearest thing I can compare it to is an optical illusion that you can't see—no matter how hard or long you look—until someone points it out to you. The spot of light that appears in the upper-right-hand corner of a movie to tell the projectionist to change reels is a similar phenomenon. You've probably seen hundreds of movies and never noticed. Neither have you noticed the watermarks on fine stationery unless you knew where to look, and exactly at what angle to hold the paper to the light. In normal use, they're invisible. That's why SDMI calls its copyright-protection efforts "watermarking."

I look forward to seeing Davidson's final tabulations, after which I may be able to put my own limited experience in the greater context of the experiences of hundreds of others—most of them more expert than I. Until then, please hold the hate mail.