Skip

Original Performance, New Recording
September 4, 2007 1:10 PM   Subscribe

Zenph Studios has developed a process (using high-resolution MIDI) which "re-performed" Glenn Gould's famous 1955 piano recordings of Bach's Goldberg Variations in hybrid multichannel SA-CD format.
posted by chuckdarwin (48 comments total) 9 users marked this as a favorite

 
I, for one, welcome our new robot Glenn Gould overlords.
posted by The Bellman at 1:29 PM on September 4, 2007 [1 favorite]


Ah, but does it sing to itself as it plays?
posted by felix betachat at 1:34 PM on September 4, 2007 [2 favorites]


Using high-resolution MIDI they can extract the metadata of his soul and upload it to MECHA GLENN GOULD.
posted by basicchannel at 1:58 PM on September 4, 2007 [1 favorite]


I saw a demonstration of this a couple years ago. Being 3 feet away from a grand piano playing Glenn Gould playing Bach on itself is kind of eerie. Eerier was the other demonstration, of a jazz guy (whose name temporarily escapes me) improvising drunkenly on an upright piano several decades ago. Some of my friends were kind of vehemently creeped out about the whole thing, but I thought it was pretty awesome.
posted by revfitz at 1:59 PM on September 4, 2007


More seriously, though, I can imagine no reason why anyone would by the Zenph recording given that there is an exceptional version of the 1955 recording available (and you get the 1981 recording and a very odd "interview" for free). Are the multichannel bells and whistles really so awesome that you would give up the real thing and go with Mecha Gouldzilla?
posted by The Bellman at 2:05 PM on September 4, 2007


Can I get the midi file and play it with barking dogs?
posted by StickyCarpet at 2:07 PM on September 4, 2007 [1 favorite]


@StickyCarpet

Yes, but only if your barking dogs sample has 1024 velocity layers.
posted by guywithnoear at 2:14 PM on September 4, 2007


Leave it to the Internet to develop a high-resolution barking dog soundfont.
posted by Mr.Encyclopedia at 2:26 PM on September 4, 2007


Odd that we've spent the last 10 years working on bell-and-whistling 3D APIs up-the-yin-yang but sound programming is stuck in the 80s. I've got a membership on Yamaha's pay-MIDI site . . . the XG data is pretty good, but closer to old-school MIDI than what it could be, these days.
posted by Heywood Mogroot at 2:26 PM on September 4, 2007


Fast Company had an article on this. They don't seem so impressed by the original version: "Yet the monophonic recording, marred by tape hiss, holds up poorly."
posted by smackfu at 2:31 PM on September 4, 2007


Has anyone done a double-blind "taste test" to see if Zenph's MIDI transcriptions can hold up to an original?
posted by Blazecock Pileon at 2:35 PM on September 4, 2007


That's a spectacularly idiotic comment, even for Fast Company. The 2002 re-mastered release I linked to above is stunning. See the Slate review:

The first thing one notices about it is its technical excellence. The sound of Gould I has been spruced up, and Gould II is a revelation. The sonic crudity that bothered me 20 years ago resulted from the Sony engineers' inexperience with digital technology; for this re-release, they have located analogue masters that were made simultaneously with the digital and painstakingly edited them to match the version supervised by Gould. The results are glorious.
posted by The Bellman at 2:43 PM on September 4, 2007


Yes, but only if your barking dogs sample has 1024 velocity layers.


At first I assumed you must be joking. I just finished up my last sampling project in which 16 velocity layers were required.

Then I started clicking.


Jesus. H. Christ.

Simply amazing.
posted by lazaruslong at 2:47 PM on September 4, 2007


i just checked, because i had never noticed anything wrong with these recordings, and i have the same version bellman linked to.

anyway, is sacd taking off in the usa? i was under the impression it had died long ago.

also, now i think about this a bit more. wtf?! they're going via midi? what on earth for? so that you can hear he'd sound like if he was playing a vibraphone? combining huge levels of compression (which is what encoding to midi is) with audiophile formats seems a little odd. is this just one needless technology trying to leverage another needless technology for sales?
posted by andrew cooke at 2:55 PM on September 4, 2007


Also from the Fast Company review:

A third set of "binaural" tracks was recorded with microphones positioned in the ears of a foam head perched at the keyboard--so with headphones, a listener can hear the music as Gould would have himself.

I can only assume they are referring to the Neumann Binaural Dummy Head. And I mean...the heads sound interesting, for sure. But go take another listen to the Pearl Jam album Binaural (many of the tracks recorded entirely with binaural mics) and tell me it really sounds like you are "right there".

To my ears, it sounds like you are right there with a wet sheet draped over your head. Just sayin.
posted by lazaruslong at 2:56 PM on September 4, 2007


also, now i think about this a bit more. wtf?! they're going via midi? what on earth for?

I may be able to take a stab at this.

The process they are using is pretty unique. They wrote software that "listens" to the original recording. It utilizes 1024 velocity layers for both noteOn and noteOff, as well as 1024 layers of "touch". As the software "listens" to the original, it quantizes what it hears into MIDI data and recreates it as exactly as possible. Then it compares what it extracted to the original for error correction.

Then all that incredibly complex midi data is fed through a digital piano that "plays" the performance, which is then recorded.

I'm not entirely sure what you are driving at with the compression bit. Sure, MIDI uses compression, if you want it to. But I highly doubt there is any compression being used in this process. If they have the bucks to write and develop software like this, they probably have the hard drive space to recreate the performance without using compression.

So I guess the real question is, what else could they use? I suppose they could use a sampler built from several thousand samples of an actual piano being played and then attempt to arrange the samples properly to recreate the original recording. But you want to talk about time consuming....hoo boy. Since it's sampled data and not MIDI, I don't know that you could write software to intelligently choose the right velocity layer from a wave file. Having the different "samples" arranged via MIDI makes the process automatic.
posted by lazaruslong at 3:03 PM on September 4, 2007


and if you are gonna call MIDI a needless technology then it's gonna be swords at dawn, sir.
posted by lazaruslong at 3:04 PM on September 4, 2007


what i mean by compression is that they take a huge amount of information - the recording - and represent it by a midi stream that is much, much shorter. even with an extra three bits (which is all you need to go from 128 levels to 1024) that's a *huge* amount of compression.

i believe i understand the technology. i certainly understand why midi is useful. but what they are doing here is completely pointless and, from an "audiophile" pov, amounts to the most amazing level of filtering.
posted by andrew cooke at 3:33 PM on September 4, 2007


Well, they are trying to simulate what it would sound like if he played it on a piano today, with current recording standards.

All you really need to encode his physical performance (as opposed to the sound waves resulting from it) is a list of what keys he pressed when and how fast, and how long he held the key down. This is eminently compressible using a higher-precision MIDI representation. I'm not really sure what part you're objecting to, unless it is the original goal.
posted by dfan at 3:48 PM on September 4, 2007


Maybe you didn't realize that they're using the MIDI to drive the performance of an actual piano?
posted by dfan at 3:50 PM on September 4, 2007


Holy grail? Not *quite*.

This is great for solo instruments in itself: but the really fantabulous and scary technology will get it working for ensembles.

And then there's the human voice.


posted by Twang at 3:53 PM on September 4, 2007


Ahh, I see what you are saying about the bitrate. But here's my take on it:

They take the original recordings, which depending on the age are probably pretty low-fidelity. So they start off with something significantly less than a badass bit rate / sample depth recording.

They quantize the orginial into 10- bit midi, with additional bits (256 layers = how many bits? my brain hurts.) of pedal position information.

Then they play that back through a badass piano and re-record it, probably at the highest sample rate and bit depth they can get. For the examples sake let's say they have a rig capable of recording at 24-bit 88.2k.

THEN they take that and convert it into Scarlett Book SACD, probably at the mastering house. That uses a whole shit ton of oversampling* and DSD, a 1-bit Delta/Sigma conversion scheme, which basically means it's bit depth is 1.

SACD also obviously play in full multichannel surround with lossless compression, as opposed to the lossy shit you get on redbooks.

So I don't know. I've listened to both the mp3s, and it certainly doesn't sound like they are losing any fidelity.

I think the coolest potential application for the software they wrote (which I would love to play around with) is the ability of the computer to "learn" signature styles of composers it analyzes. The way a specific pianist uses the pedal to acheive a dynamic, and other types of stylistic choices that make eac musician unique. They could take a "set" of frequently heard stylistic choices and then have the piano reproduce a different work of music with those variables implemented.

*from 88.2kHz in our example up to 2.822 MHz
posted by lazaruslong at 3:57 PM on September 4, 2007 [1 favorite]


I've been listening to the RoboGould version on regular CD for a few months now, and although I'd really like to be able to say that I can detect some sort of absence of soul or humanity or something in the playing. But I can't. It actually sounds really nice. Okay, sometimes I get a little creeped out over a robot piano emulating a dead man's hands, but that feeling usually passes.

In summation: a perfectly fine recording, as long as you don't end up plagued by nightmares of Glenn Gould's soul trapped for eternity in a robot body, doomed to endlessly repeat the exact same performance of the Goldberg Variations, making the same mistakes each and every time.
posted by thatswherebatslive at 4:04 PM on September 4, 2007 [3 favorites]


In summation: a perfectly fine recording, as long as you don't end up plagued by nightmares of Glenn Gould's soul trapped for eternity in a robot body, doomed to endlessly repeat the exact same performance of the Goldberg Variations, making the same mistakes each and every time.

You're missing the best part! Now that RoboGould has been made, they can have him assimilate more and more of his orginial recordings until they have enough information to use his style and have him play pieces he never would have had the opportunity to play!

Just around the corner is: Gould Does Coldplay. It'll sound perfect.
posted by lazaruslong at 4:06 PM on September 4, 2007


here's a typical argument i have against this, as an illustration for the kind of thing that i think is wrong here.

gould was not a machine. he was *there* when the recording was made. it was made on a specific piano - a piano that gould both used and could hear. so there's a feedback process - to some extent his performance is going to be controlled by how the piano behaves. he's not going to play exactly the same way given another piano. and yet that is what this process assumes - that you the performance and the piano are orthogonal components that can be separated. the reconstruction is what you'd hear of a robot copied gould in precise (but limited) ways on a different piano.

as i said, that's just an illustration - a specific example of how you lose information through compression. i'm sure i could think up others given time, but i don't think it would make the point any clearer.

do they encode the humming?!
posted by andrew cooke at 4:14 PM on September 4, 2007


[just to clarify i am sure it is a perfectly nice recording. i just find it at odds with the kind of aims/ideals/prejudices/aesthetics that i always assumed sacd was targetted at. although perhaps part of the failure of sacd is, in fact, that i am wrong and sony do think this is consistent...]
posted by andrew cooke at 4:17 PM on September 4, 2007


gould was not a machine. he was *there* when the recording was made. it was made on a specific piano - a piano that gould both used and could hear. so there's a feedback process - to some extent his performance is going to be controlled by how the piano behaves. he's not going to play exactly the same way given another piano.

Yes, independent of the compression issue, I totally agree with this.
posted by dfan at 4:48 PM on September 4, 2007


The inner asthete is confused: this IS his performance of the piece from 1955, in every detail.

How should one react this ultimate homage? It's not like anyone is COVERING his version. It's HIM, and HIS keystrokes... in a really human way.

Fuck. I'm sorry that Bill Hicks and Phillip K Dick aren't available for comment.
posted by chuckdarwin at 5:14 PM on September 4, 2007


Maybe you didn't realize that they're using the MIDI to drive the performance of an actual piano?

That's obviously the money point of this whole thing. That you could go to a concert hall and hear ghost Glenn or ghost Oscar Peterson play on an actual piano in an actual room. If you're talking about sitting at home listening to a recording, then what's the point? You can listen to non-ghost Gould or Peterson play the piano, on any number of their recordings.

I noticed there's a schedule of upcoming live performances (and calling them "live" is pretty funny, eh?) at their website. Actually, they call them "Re-performances". This is all pretty damn interesting!

Thanks for the post, chuckdarwin.
posted by flapjax at midnite at 5:14 PM on September 4, 2007 [1 favorite]


And you have to wonder: if legendary prima donna Keith Jarrett's MIDI ghost ever "Re-performs", is he gonna get up and storm off the stage should there be an audience member talking during the performance?
posted by flapjax at midnite at 5:16 PM on September 4, 2007


Thanks for the post, chuckdarwin.

You're most welcome, as ever. I was so worried that I was late to this party, but it seems like I've done a rare thing and scooped the blue.

I had no idea that piano rolls had gotten this advanced...
posted by chuckdarwin at 5:21 PM on September 4, 2007


Here an interesting page showing the tonal differences of the original piano, a Yamaha C9 and the C7 Disklavier.
posted by doctor_negative at 5:36 PM on September 4, 2007


gould was not a machine. he was *there* when the recording was made. it was made on a specific piano - a piano that gould both used and could hear. so there's a feedback process - to some extent his performance is going to be controlled by how the piano behaves.

This point is well made by Zenph Studio's own sample of the process. Cortot's original version of Chopin's 3rd Prelude is interesting & worth listening to. Probably because of the specific piano & room he was playing in, the particular balance & voicing he chose, works.

On the "midi clone" recording the running left hand 16th notes become muddy & overbearing & the balance is a mess. Undoubtedly this is because of the thicker sound of the modern piano in that register, greater reverberance of the room, microphone placement choices, and all the rest.

I'm sure with some experimentation with pianos/rooms/recording techniques a better result could be obtained.

But the point is--any musician worthy of the name doesn't just push down the right buttons in the right order with a pre-determined degree of loudness. It's more a matter of listening, responding, adjusting to the actual sound an actual instrument is making in an actual room.

(And a guy like Glenn Gould wasn't just recording in random rooms at some recording studio. He chose the space where he made is recordings, as well as the instrument he used, very carefully . . . you just can't separate the final results from choices like that.)
posted by flug at 6:18 PM on September 4, 2007


ghost Oscar Peterson

I just had a minor heart attack.

OK, all better now.
posted by emelenjr at 7:44 PM on September 4, 2007


I think this is a really interesting technology. And I don't really understand this talk about compression. The fact that it's MIDI is immaterial. They're recording 1024 levels of touch and 1024 pedal positions. Do you think the difference between level 986 and 985 of hitting C#5 is going to be audible?

I can only assume they are referring to the Neumann Binaural Dummy Head. And I mean...the heads sound interesting, for sure. But go take another listen to the Pearl Jam album Binaural (many of the tracks recorded entirely with binaural mics) and tell me it really sounds like you are "right there".

Binaural recording is a technique, like stereo recording. One recording using that technique sounding bad doesn't necessarily mean anything about binaural recording generally.

And thanks so much for the post, chuck. Listening to Gould play the Goldberg Variations just now brought tears to my eyes.
posted by ludwig_van at 10:34 PM on September 4, 2007


i think people are confused by talk of compression because they are used to a certain kind of compression and certain kinds of artifacts as a result.

i am not saying for a minute that this recording will sound like a low-bitrate mp3. it's obvious that a well-made recording using this technique will not "sound compressed" in that kind of way at all.

but still, you are taking a lot of information, throwing away most of it, and then reconstructing the result from what remains. you simply *cannot* expect to get an identical result out of the end. the results *must* sound different to the original.

if you listen to the example on the web pages that flug linked to, this is obvious. the two recordings sound nothing like each other in many ways.

in some cases, that may be what you want. for example, there's a huge reduction in noise. but the fact remains that this technique cannot - as a mathematical fact - reproduce exactly a recording. because the data are compressed - the sound gos through an intermediate (in this case midi) representation with a lot less information.

from people's responses here, i guess that "compression" as used in popular language refers to the kind of thing you hear with mp3s. but it's also a technical term, as i have tried to describe.

as far as whether or not 1024 levels is interesting - a cd uses about 16bits depth. that's equivalent to 65,536 levels. mastering often uses 24bit, which is 16,777,216 levels.

now the reasons these levels are much higher than the 1024 used in the midi representation here is to largely to do with keeping noise levels low. but even so, again, you can see that there is a loss of information. a cd can effectively record 60 thousand different intensities of note, while this system can handle only a thousand.

again, the way the system works, this won't sound "bad in the way an mp3 sounds bad". but the data are most certainly being compressed.
posted by andrew cooke at 4:16 AM on September 5, 2007


one final comment. it just struck me why this kind of technology is being pushed now.

for a long time the "holy grail" of audio was accurate reproduction - getting the "authentic copy". companies have made money from this by re-visiting master tapes etc, as described above.

but with digital media, the internet, etc, this is losing value. i read just recently that labels are now putting increasing emphasis on concerts and merchandising. even here in chile, where median wage must be a couple of hundred dollars a month, we get big bands coming over and charging 100 dollars for tickets.

so how can labels continue to make money from people like gould in a market that is changing like this?

(i thought i was kind-of repeating what someone said up-thread, but i can't find that now)
posted by andrew cooke at 5:05 AM on September 5, 2007


but still, you are taking a lot of information, throwing away most of it, and then reconstructing the result from what remains. you simply *cannot* expect to get an identical result out of the end. the results *must* sound different to the original.

as far as whether or not 1024 levels is interesting - a cd uses about 16bits depth. that's equivalent to 65,536 levels. mastering often uses 24bit, which is 16,777,216 levels.


The point isn't for the results to sound like the original. The point is to reconstruct the performance of the original, which is separate from the sound of the original. Most of the information from the original analog source (the sound of the original recording) needs to be thrown away to isolate the performance. i.e., room sound, harmonics, etc. All they need to encode is which keys were pressed and how they were pressed.

Comparing CD bitrate to this is missing the point. This is nothing like a CD recording. Audio data on a CD needs to represent actual sound waves - the MIDI data here only represents the performance, not the sound. And the digital MIDI data is brought back to the analog domain when the piece is actually re-performed. It's going analog (original performance) digital (MIDI encoding) analog (new performance).
posted by ludwig_van at 10:10 AM on September 5, 2007


we seem to be talking past each other. i was trying to explain why i used the word "compression" in a way that confused people, apparently.

i agree that from one pov the "point" of this is to make "performances" available. i also think that from a commercial pov the "point" is to make money by shifting from recordings to performances.

for me personally the "point" remains "pointless" - i grew up under the assumption that the recording is what was interesting, not another recreation of it. but i imagine there are indeed people who will pay to see this (someone here recently called the blue man group "experimental music" - there's clearly a market for "seeing something happen").
posted by andrew cooke at 10:29 AM on September 5, 2007


Well I understand what you mean by compression, but I think encoding is perhaps the more appropriate term here. They're encoding a performance, and if the performance can be subsequently decoded without any audible loss of information, I don't see what's to criticize about their method.

Disagreeing with their philosophy is different. But I'm not sure what you mean when you say that the recording is what's interesting. When you listen to an early recording of classical music, it's not enjoyable to listen to because of the way it was recorded, is it? We listen to hear the performance, and we often do our best to mentally filter out the peculiarities of the recording (tape hiss, room resonance, if someone in the audience coughs, etc.) If we can encode the performance, we can separate it from the recording and experience it in whatever medium we desire, even watch it happen before our eyes. Sure, it might not be as good as having been there to witness the original performance, but it seems like it could be the next best thing.
posted by ludwig_van at 11:32 AM on September 5, 2007


Well I understand what you mean by compression, but I think encoding is perhaps the more appropriate term here. They're encoding a performance, and if the performance can be subsequently decoded without any audible loss of information, I don't see what's to criticize about their method.

Ah, but that's the point — information is being lost. Without a time machine and a set of motion sensors to hook to Mr. Gould's fingers in '55, we're never going to have all the information about his performance. We can reconstruct it, but there will necessarily be places where we're guessing or fudging.
posted by nebulawindphone at 1:21 PM on September 5, 2007


Yeah, fair enough, but I think that makes an assumption about how good their technology is or could potentially be -- the reconstructed performance won't contain exactly the same information as the original, but the question is when does that difference stop being audible? I'm not convinced that encoding each keypress with 1024 discrete gradations of attack and release is insufficient for capturing all the audible nuances of a solo piano performance. If not, then it's a question of how accurately they can encode the original performance, not whether the reproduction mechanism is sufficient.
posted by ludwig_van at 1:45 PM on September 5, 2007


technically, in the context of your narrow definition of performance, what they are doing is perfectly ok (and i am not sure why you think the word "compression" automatically means "bad").

the word "performance" can mean many things. for the restricted sense that you are using perhaps we can use the term "restricted subset of the mechanics of the performance".

my problem, then, is that i am very suspicious of "fetishising" this restricted subset of the mechanics of the performance. it seems to me that we already put too much emphasis on particular "great" recordings. from the way you are behaving i can imagine a future where people go to concerts to see playbacks of restricted subsets of the mechanics of a performance, and i think it's just one more example of the way in which we, as a market, are manipulated to have an interest in whatever it is that can be sold to us.

it's a cute technical hack. i think it has the promise to become something much less...

maybe a completely different example will help. when the riaa or whoever it is blocked me from listening to pandora (i am not in the usa so i now have to tunnel through a us-based server to register for the stream) i decided to stop buying music from the big labels. i love music - i am no expert, but it's a big and important part of my life; i am nearly always listening to it - so i made my decision with some worry - i don't download music, so how could i continue to explore this rich source of pleasure?

i found that there are a bunch of local labels. chilean bands that press their own cds, or hobbyists that run small distributions. it turns out that the music is just as rich and interesting as the stuff i thought was being hand picked from across the world. and this is in santiago - hardly a hot-bed of musical culture.

i don't know if that helps. i'm getting tired of typing. sorry if i seem to be changing my argument, too - i'm working this through as i go.

but it seems to me that you're impressed by a narrow technical gimmick and building it into much more than it really is. and i think that's easy to do - i don't think you are particularly stupid or gullible - and i don't like it. i don't like the manipulation and i don't like way this same process has made us lose track of so much that is good but not sellable.
posted by andrew cooke at 1:53 PM on September 5, 2007


bleagh. that was pretty wanky wasn't it? sorry.
posted by andrew cooke at 3:28 PM on September 5, 2007


technically, in the context of your narrow definition of performance, what they are doing is perfectly ok (and i am not sure why you think the word "compression" automatically means "bad").

I don't think "compression" means "bad," but I think it has connotations that make it less here fitting than some alternatives. I think that term belies the fact that the audible end result of this process is not compressed digital data but analog sound (a piano actually being played, not a digital sample of a piano being played).

the word "performance" can mean many things. for the restricted sense that you are using perhaps we can use the term "restricted subset of the mechanics of the performance".

my problem, then, is that i am very suspicious of "fetishising" this restricted subset of the mechanics of the performance.


What do you mean by that? What's missing from the performance, assuming for the sake of argument that this technique works as it's meant to? Actually seeing the performer before your eyes? Between-song banter? Or some kind of non-physical (non-audible?) aspect? Again, I don't think anyone would claim that something like this is superior to having attended the original performance, but if it's between this and listening to a technically-poor recording, this seems potentially a lot closer to the ideal experience.

I don't really follow your last anecdote - I'm a big fan of independent music, too, but what does that have to do with this technology? I don't see the manipulation involved in folks wanting to experience the music of someone no longer living or performing as they and those at the time might have heard it, unmolested by the technical peculiarities of a certain recording. If you think people should just be more forward-looking about music, then I guess I can accept that, but what can I say, I'm something of a traditionalist/formalist when it comes to these things.
posted by ludwig_van at 5:56 PM on September 5, 2007


There must be some way to involve vinyl in this process.
posted by smackfu at 6:06 PM on September 5, 2007


There must be some way to involve vinyl in this process.

Maybe John Cage prepared piano pieces, using bits of boken records as material for preparing the piano?
posted by flapjax at midnite at 6:54 AM on September 6, 2007


Someone near the top quoted something about tape hiss. There's a little bit of tape hiss on those original recordings, yes, but most of the audible hissing is Gould himself making shhhhshhhshhh noises.

Now Andrew Cooke, since we're talking about what the real Glenn Gould sounds like, riddle me this: Does a Glenn Gould record sound anything like Glenn Gould would sound live in a concert hall? I'm willing to wager a hell of a lot of money that no, it doesn't.
posted by Reggie Digest at 10:54 AM on September 15, 2007


« Older A Picture Counts   |   FILM IS ABOUT TO START... Newer »


This thread has been archived and is closed to new comments



Post