Recording of September 2007: J.S. Bach: The Goldberg Variations

J.S. BACH: The Goldberg Variations, BWV 988
Glenn Gould, piano (1955); Yamaha Disklavier Pro piano, "Re-Performance" by Zenph Studios Sony Classical 8697-03350-2 (SACD/CD). 2007. Howard H. Scott, prod. (1955 mono sessions); Steven Epstein, prod. (2007 stereo, multichannel, and binaural sessions); Peter Cook, Richard King, eng. (2007); Dennis Patterson, asst. eng. (2007); Marc Wienert, piano voicer (worked with Gould); Ron Giesbrecht, calibration. Zenph project team: Mikhail Krishtal, Anatoly Larkin, Peter J. Schwaller, John Q. Walker. DDD. TT: 77:02.
Performance *****
Sonics ****

I vividly remember my first reading of the short story "A Work of Art," by James Blish (1921–1975), because it combined two of my passions: science fiction and classical music. A team of 22nd-century neurosurgeons has brought back to life the mental processes of composer Richard Strauss and implanted his revived personality in the brain of a hapless volunteer. Herr Dr. Strauss gets to work composing a posthumous opera, which is greeted with acclaim at its premiere. But, even as the last note dies away, the composer realizes that it has been a pointless exercise: the sound, the notes, may have been all authentically Straussian, but the music was bereft of substance (footnote 1).

I was reminded of the Blish tale when I first heard about the work being done by Zenph Studios: analyzing a piano recording in order to create an expanded MIDI instruction set (footnote 2) to drive a Yamaha Disklavier Pro reproducing piano. Assuming that every aspect of the performance—which keys struck when, and how hard or soft, along with the pedaling—could be reproduced, a historic performance captured in less-than-stellar sound could thus be "re-performed" on the Disklavier and recorded in a more optimal acoustic environment.

I have some experience of this concept, having in 2000 recorded Robert Silverman's performances of all of Beethoven's piano sonatas on a Bösendorfer reproducing piano (developed, like the Disklavier Pro, by Wayne Stahnke). But Bob had captured the performances by playing the same piano in the same hall where I was to make the recording—a very different matter from beginning with a recording of a different piano in a different hall. The relationship among performer, instrument, and hall acoustic is complex, and what might be the optimal combination of tone, attack, and tempo for one situation will not be transportable.

I had other doubts. The process might perhaps be feasible for music in which one piano note is played at a time; but polychromatic music, in which many notes are struck simultaneously, each with its own loudness, would seem way too complex for such analysis—I remember a piano master class in which Alfred Brendel devoted an entire hour to the different possible implications of just the opening chord of a Beethoven sonata, depending on which note was to be played loudest. Nevertheless, reports of Zenph's preliminary demonstrations—the company is named after senf, the German word for "mustard"—were very positive. (I am told that a lot of manipulation by hand goes into the transfer to MIDI.)

Then, last September, Robert Deutsch attended a live concert in the Canadian Broadcasting Corporation's Glenn Gould Studio concert hall in Toronto, with—shades of James Blish—the late Glenn Gould performing J.S. Bach's Goldberg Variations. The performer was actually a Disklavier Pro, controlled by MIDI data compiled by Zenph from Gould's 1955 debut recording (see "Industry Update," December 2006, p.15). Bob was impressed, as was the audience of over 300, who, he reported, "exploded into thunderous applause."

The same week as the concert and in the same hall, Sony recorded the Zenph "re-performance" direct to DSD, using first a surround array of DPA 4006 omni mikes placed conventionally, then a Neumann KM-100 dummy-head binaural mike placed where Gould would have been sitting (though a little higher than the pianist's famed slouch, according to the photos). The result of those sessions is this fully loaded hybrid SACD. Not only does it include high-resolution 5.0 surround and two-channel mixes, it also repeats the entire performance in binaural stereo for headphone listening.

There's no point in my discussing the performance. As Hyperion Knight discussed in his 2003 "Records To Die For" entry, Gould's 1955 original is a peak of the recorded literature, though I do find that Rosalyn Tureck, from 1999 (CD, Deutsche Grammophon 289 459 599-2), and Murray Perahia, from 2000 (SACD, Sony Classical SK 89243), spend more time in my player these days. But the sonic differences between the original and the "re-performance" are fascinating.

I have two earlier releases of Gould's 1955 performance of the Goldbergs on my CD shelves: the 1992 Glenn Gould Edition (Sony Classical 7464-52594-2) and the 2002 State of Wonder (Sony Classical 9699-87703-2, which also includes Gould's 1981 performance, transferred from the backup analog tape rather than the flawed early digital used for the original release). Looking at the WAV files for the closing Aria from each release using Adobe Audition indicates some significant differences, not only between the Zenph re-performance and the transfers of the original but also between the two transfers.

Yes, the original was mono, of course, but the recorded levels are also different. The 1992 transfer of the Aria is quieter, for example: it has a total RMS level of just –35.01dBFS compared with 2002's –32.08dBFS and Zenph's –30.29dBFS. In addition, measured from the transient of the first note to that of the final note, the 1992 transfer is 2 minutes, 5 seconds, 2 frames long (one frame = 1/30 second). Both the State of Wonder transfer and the Zenph measure 2:04:23—a quarter of a second shorter. This difference is inconsequential compared with the timing of Gould's 1981 reading of the Aria, an almost glacially paced 3:36:08, but it reinforces the idea that, with historic recordings, getting the analog playback recorder to play at the correct speed is not an insignificant issue.

It looks as if Zenph used the State of Wonder transfer as their reference, and, in fact, playing back the two simultaneously on the PC, lining up the files on the waveform of the Aria's opening note, revealed a remarkably consistent synchronization, with no phasing effects audible.

And the Zenph sound is superb. The balance in stereo—I don't have a surround rig—from just the front two DPA omnis is reminiscent of the piano recordings Bud Graham used to engineer for Sony: warm, rich, and involving. The acoustic of the Glenn Gould Studio is a little drier than Columbia's 30th Street studio, but the tape hiss of the original is missing in action, along with Gould's infamous humming.

I auditioned the binaurally recorded version of the perfomance with both Ultimate Ears UE-10 in-ear headphones, and conventional Sennheiser HD-600 cans. As is almost always the case, I couldn't get a soundfield that extended in front of my head with either set of 'phones, though the subtle sense of hall ambience did extend to the sides and rear. The piano sound was naturally balanced, if a little mellower than the more conventional recording, though, as it was recorded with the dummy-head microphone at the piano bench, the treble notes come from the right and the bass from the left, the opposite is what is usually heard from a position in the audience.

Overall, this is a superb-sounding SACD, but what about the Blishian implications—that, in the details of this technological tour de force, the music has been forgotten? The recording venues are sufficiently similar that I don't believe it has been lost. Of course, I know this performance backward, but I still reveled in the Zenph re-performance. And it is not as if the original is no longer available.

Well done, Team Zenph!—John Atkinson

Footnote 1: Reader William Murphy reminded me that the greater point of the Blish story was that the "new" Strauss opera was not the intended work of art; it was the reincarnated composer himself.

Footnote 2: Basic MIDI assumes 256 levels of loudness quantization; the Disklavier Pro multiplies the possibilities by a factor of 8, and increases the number of MIDI events per key press from two to seven, which should be sufficient for acoustic piano.