Slashdot Mirror
Concert to be Performed from Beyond the Grave
rtphokie writes "Raleigh, NC based Zenph Studios is hosting a live concert performed by two piano virtuosi long since dead. Zenph developed software which digitally transcribes performances even from the scratchy recordings. A more faithful transcription of timing, key and pedal pressure is achieved using Yamaha's high resolution version of MIDI."
did you actually read the article - or even the summary?
The piano is just the thing they play it on, no one's saying that's a great achievement - MIDI enabled pianos have existed for years.
The achievement is being able to accurately translate a recording into MIDI instructions.
Whether or not it's as good as they claim is yet to be seen.
http://www.pragmaticprogrammer.com/articles/zenph/ index.html
Agile Artisans
Disclaimer. I used to work for the founder of Zenph Studios, back when he founded a networking software company and he left when it merged with a larger one. I am still in touch.
Someone takes a recording from long ago on vinyl. They play it on vinyl. Advanced signal processing listens to the sound from the original recording and detects which note is played when keys are pressed and lifted, and apparently when pedals are in use as well. This is laid down in a high-fidelity MIDI format. This MIDI file is fed through a high-performance Yamaha piano and the concert is played live on the piano in the concert hall. The piano translates the MIDI files and hits the keys, pedals, everything to the exact timing specified in the file. The magic is in the signal processing of the original recording. The idea is to replicate the original recording, note for note, tone for tone, microsecond for microsecond, feeling for feeling.
So, this is a HUGE step beyond player pianos. We can replicate old recordings and (GASP!) re-record them using modern methods, saving old lost tapes, making old recordings available in SACD and DVD-Audio. We can replicate concerts across the globe. Piano competitions can be done remotely. This could be of incredible significance to old classical music libraries and performances.
And the CD sounds pretty darned good. http://www.amazon.com/exec/obidos/tg/detail/-/B000 005J1I/103-9970286-1975026?v=glance
Velocity is a standard part of the MIDI protocol...
There are a few MIDI things related to volume -
Channel volume and Key velocity are the most common used.
Volume just being the output volume.
Velocity is usually used to affect both volume and timbre.
There's also Chanel aftertouch and Key aftertouch, which depending on the instrument may affect volume, timbre, pitch, or nothing at all.
Controller 11 - Expression is often volume related, and there's also the Breath controler (I don't know the number off the top of my head) which can also be used to affect volume and / or timbre, pitch or whatever...
There's plenty of room in standard MIDI for a wide range of expressiveness, it's usually the instrument that falls short, not the protocol.
Advanced users are users too!
That's incorrect. Velocity is sent as part of a Note On event in General MIDI 1.
GLaDOS for President 2016! "Well here we are again. It's always such a pleasure." -- GLaDOS, 2011
First article quote: The key to the phantom concert lies in the transcription of the scratchy recordings into a high-resolution version of MIDI
Second article quote: It has successfully tried out the Cortot and Gould pieces on the Disklavier Pro, one of only a few concert grand pianos that can record and play back high-definition MIDI files.
It looks like you skimmed the article, only saw the second quote, and rushed to post an half-assed comment about how 'the author of the article just seems to think that MIDI is "high-def"'.
If you'd have read the full article carefully, you'd have noticed the first quote and understood that "high-def MIDI" denotes another format.
Nice post, but just a small remark - sometimes the score won't be available, since for example Fugues (which Glenn Gould liked to play) involve some improvisation, and I bet that even in other places he would improvise some times.
The AACS key is NOT 0xF606EEFD628B1CA427BEA93A9CA9773F
Gould's reputation was built on improvisation. His 1955 recording of the Goldberg Variations became an instant best seller and was one of, or, possibly, the first classical best seller. His refusal to play Legato (smoothly) upset the status quo of the classical music world. Leonard Berstein, legendary conductor of the New York Philarmonic, deferred to Gould's quirky redition of a work, suggesting that genius of Gould's level should be allowed leeway.
"Academicians are more likely to share each other's toothbrush than each other's nomenclature."
Cohen
The "high def" MIDI the slashblurb refers to is probably what yamaha calls "Yamaha XP MIDI".
I can't find a technical spec on this right now, but its mostly backwards-compatible with MIDI plus the addition of a few extra details about the piano performance, e.g. key stroke depth, using the MIDI controller extensions. (However I'd be really suprised if it it was actually possible to determine key stroke depth from signal analysis of an old recording).
If they did their homework it would should also have a higher clockrate. MIDI is notorious for its poor time resolution with a clock of only 1khz -- and studies have shown that virtuoso pianists can control timing down to the sub msec range, so this is essential.
The article also beats around the bush on the polyphonic transcription issue -- but since these are classical pieces, score following seems like an obvious if not relatively easy way to do it.
Last I looked, Midi over USB and Firewire (generally in the mLan protocol) still has the 31200bps limitation.
You get a lot more channels out of this running parallel (as opposed to trying to string 16 midi channels side by side -- which at 31200bps, you can hear flams if all 16 channels are sounding at the same time on the same midi bus, even more pronounced if CC data is also being sent).
But then again, its been a few years since I talked with any of the Yamaha engineers. A good frind of mine was manager of the mLan project over there, but I think he's off doing something else in the company now (Hey George!).
But all in all, for the data thats being sent through, its not that limiting if you are only dealing with one channel.
In my job as a sound engineer, I regularly edit classical piano music. High-end midi pianos have an amazing potential for recording, as no high fidelity recording can substitute for a real live sound. A recorded midi file would have the feel of an artist, and it would be easy to correct wrong notes.
Extracting a usable midi file from a recording is very sophisticated signal processing. If the pedal is down a new attack can get lost among the wash of notes. The musical score would help if the program knew what notes to expect. Even so, I suspect they chose Glenn Gould because he was very sparing with the pedal when he played Bach.
I’m old enough to remember 16K of memory being described as “whopping”
No this is wrong. Yamaha has a form of high def MIDI for recording grand pianos. They abandoned 7-bit MIDI because it couldn't reproduce all the nuances they wanted to capture. This is the data that is captured. However, there is no such this as high-def MIDI really, it's an analogy (which is strange for a digital idea).
spoonerize "magic trackpad"
An acoustic piano doesn't have a volume control. The only control over the volume is the key velocity, as a performer has almost no control over the sound after a key has been hit.
"Lo-def"MIDI would include keystrokes with their velocity, key releases, and when the pedal is engaged and released. I believe Yamaha's hi-def MIDI treats the pedals as continuous controllers so every move can be accurately recorded. They probably log the pressure and speed of release of the keys as well, though this would have only a marginal effect on the sound in all but exceptional circumstances.
I’m old enough to remember 16K of memory being described as “whopping”
Believe it or not, there isn't much more to it than pressing buttons. The striking of a particular note contains few variables: the time at which the note is pressed, the velocity at which the hammer is propelled, and the length of time during which the note is sustained (some piano effects involve the speed at which the key is lifted, but this is rare in most performances). Pedals are a simple question of displacement over time.
The opinions stated herein do not necessarily represent those of anybody at all. Deal with it.
Yamaha Disklavier Pro
Powerful Pro MIDI Data Format Holds Expanded Performance Data
To achieve its superior recording and playback performance, the Disklavier Pro generates extra MIDI data to accommodate precision parameters. All essential data such as hammer speed, key release speed and pedaling is recorded and reproduced within the standard MIDI format, enabling full compatibility with other MIDI devices. For recording piano performances with even more subtlety and detail, you can select the expanded Yamaha XP MIDI format. This utilizes general-purpose controller numbers in the MIDI specification to record additional performance data including key speed and stroke depth, for recordings of enhanced sensitivity and precision. Multi-mode flexibility allows the Disklavier Pro to interact smoothly with existing MIDI configurations and reproduce data generated in either XP or enhanced mode, with automatic selection of the appropriate playback format.
Ahhh, tuning. There are subtleties there.
Fans of "Donald Duck in Mathmagic Land" know that a "perfect fifth" interval has two notes, with a 3:2 ratio of frequencies. 1.5000000000000. And a major fourth interval has frequencies with a 4:3 ratio. 1.3333333333333....
Intro to music theory (or a little piano experience) notes that an octave is a doubling in frequency, and contains 12 half steps. A major fourth is five half steps, and a perfect fifth is seven. From an octave (2:1), one can compute the frequency ratio for a half step by solving for x:
x^12 = 2. I get 1.0594630943592952645618252949463
Five of these gets a frequency ratio of 1.3348398541700343648308318811823, not 4/3. A touch sharp. Seven mathematical half-steps yields a frequency ratio of 1.4983070768766814987992807320264, not 1.5. A touch flat.
A piano tuner (or the designer of any instrument) can use the mathematical equal-tempered scale, and have fourth and fifth intervals sound a bit off. Or the tuner can make some, but not all, of the fourth and fifth intervals work perfectly. Choose which ones. Make too many perfect, and the other notes will sound off, and may sound horrible in chords.
It's all about tradeoffs.
First of all, the concert is not only given by 2 dead pianists. I'm the performer of the recital. The 2 dead pianists are present during the middle of the recital where we will demo this new technology.
For all of you who has reservations about the results, i suggest that you come and hear the demo. This is not about getting a piece of music performed the way a composer wants it performed. It is also not about taking an old recording and "cleaning it up". What this technology will do, is completely recreate the original performance. And before we can legitimately question the accuracy of this re-performance, we need to look at the chain that constitutes a recording. In the original recording session, there are the following variables:
1. The performer, who is playing the piece on a piano whichever way they preferred.
2. The actual acoustic recording, going from mics to preamps to tape/dat/cylinder/whatever the method maybe.
3. The acoustics of the hall, room, space that the recording is taking place in.
In the re-performance cycle, there are the following variables:
1.Deciphering the original performance from the other variables, such as recording noise, reverb, out-of-tune notes, etc. However, the original performance's qualities to be preserved include nuances of tone, minute timing changes for every single note, loudness of every single note, pedal usage (including different gradations of pedal), and many other things.
2.Playing back this "essence" of the performance on a Yamaha Disklavier Pro.
3.Accurately matching the Yamaha Disklavier to the sound and tone qualities of the piano used in a recording, and adding whatever acoustic ambience variations to the performance (space, hall, room, etc).
The purpose of the re-performance is to throw out all of the variables during the recording, except for the performer. The ability of the Yamaha disklavier pro to reproduce this is not to be questioned - When it is accurately calibrated, I cannot tell the difference between my own performance and a played back version. In fact, if I were to acoustically record my own performances, and the played-back performances of the Disklavier, i would not be able to distinguish between the two (with a few minor exceptions, at least as far as the Mark III pro is concerned).
What this means, is that there are 2 variables still to be controlled in the playback cycle. The piano and the acoustics, and the actual "decoding" of the performance. The demonstration on May 19th will be about the decoding. All of the other variables, save the piano matching, has been take care of.
It is an amazing piece of technology, and I say this as someone who has critiqued the process every step of the way, and has seen the results firsthand, multiple times. Come on out, enjoy the recital, and see the tech demo. Mei-Ting Sun