Saturday, May 10, 2014

End of an Epic Era

I just bought another virtual instrument collection. Which it turns out I maybe don't need for the work next week but that's another story.

No, not this one. (Although I wouldn't mind having this one). Instead I got Garritan's world instrument collection, which is one of the few cheap options that includes a suona.

But faceplates like the above make me wonder if we are finally coming to the close of an era.

Using electronics to generate musical notes was not a new idea. Theramin saw his invention as a possible replacement for the violin. Classical composers were incorporating the Ondes Martenot (and the experimenters like Stockhausen were experimenting with non-traditional sound sources like untuned radios and vacuum cleaners). And something as venerable and established as the Hammond B3 is of course an electronic version of a traditional pipe organ.

By the late 60's electronica had made it out to the pop world, although pioneers like Wendy Carlos were still working laboriously, often note-by-note, with hand-soldered circuits (Delia Derbyshire created the first version of the Doctor Who theme by recording on to audio tape and physically cutting the tape to make the individual notes).

Then came disco and the various flavors of electronic pop; Kraftwerk, Tangerine Dream, and a great many bands with sound driven by a keyboard wizard like Keith Emerson. Electronic sounds were out there, and just like the fate that befell the Theramin, these sounds were celebrated for their novelty, even as their inventors continued to work tirelessly to achieve half-way realistic simulations of "real" instruments.

(Take as example the Mellotron. Which, technically, could reproduce a realistic orchestral flute sound. But as employed by George Martin, the technical limitations of the tape-drive mechanism was exploited to cut an unusual and fresh flute-like track into the album. Much as the artificialities of a device originally intended to clean up the pitch of an inexperienced singer -- auto-tune -- has become a performance tool and a style all of its own.)

Realistic synthesis has always been there as some Shangri-La of the tireless engineers and experimenters, even as the music-makers have been out there using the tone provided in whatever way they see fit.

I started playing with electronic keyboards at the height of FM synthesis, that Yamaha creation that was at the heart of their DX line, and allowed the first generation of General MIDI chips that gave the term "MIDI Music" such a bad name. When I started, samplers were cumbersome, expensive, and limited. But when I went out and purchased my first sound module, LAS had hit the scene.

LAS was really an elaborated way to stretch the limited number of samples forced by the size of memories those days into enough octaves to be playable. In the Roland sample-based synthesizers, several different looping samples and one or two initial transients were layered together to provide a tone that more-or-less disguised the looping nature and more-or-less changed timbre as you moved up and down the instrument range. With good programming, you could even simulate to some degree the changes in timbre around the chalumeau register.

But of course Moore's Law marches on. And each manufacturer -- and soon, a whole new industry selling sample libraries -- touted the sheer size of their library as compared to their competitors, or to last year's model.

Soon enough the personal computer outstripped the pace of improvement in rack-mount samplers, and the sheer size of sample libraries had forced them out of ROM cards and floppies and on to dedicated hard drives. Rack-mount samplers moved out, and virtual instrument libraries moved in.

Meanwhile the tireless elves were experimenting with physical synthesis, trying to achieve those same characteristic instrument colors by modeling the physics of the instrument itself. With some interesting results, but rarely ones that were both affordable and playable.

I of course spent most of this era one step behind. Barely able to afford used sound modules, then hanging on to them when the world had moved on. Barely able to afford a computer that would run the latest libraries, and generally unable to afford things like Kontakt (nee...well, lots of name changes over its history); the software that is like an operating system for the vast majority of modern libraries. As in you are assumed to have bought it if you are doing anything in computer-based music, and that way the libraries don't have to bundle a sampler with their sound collections.

For the usual reasons -- among them, getting tired of having all their work tied to someone else's ever-changing standard -- the big library makers did eventually start distributing their own playback engines, free with their libraries.

Which only highlights the growing insanity of copy protection. During the boom, these big libraries, and the flagship software like Kontakt, were priced in the thousands of dollars. And inevitably onerous copy-protection schemes -- including dongles that hogged your USB ports, challenge-response that required an active internet connection, MAC checkers that shut down your software if you dared repair or replace any of the hardware on your computer, and even root kits were all employed.

To protect, in many cases, an engine that was given away anyhow, and file libraries based around samples (the majority of the work) that were in plain unencrypted format.

Hand in hand with this, seemingly; the efforts to protect their work, and the desire not to be dependent on another manufacturer's playback engine, the tendency of libraries moved swiftly away from being editable sounds, to being presets. All the limitations of the old rack-mount was still there, except the patches took a lot longer to load and had a habit of turning off mysteriously in the middle of a recording session (because some line of deep-buried code had just decided your registration had expired).

And the nature of the libraries began to change.

Always, there has been a small subset of users that, like the academics mentioned previously, were only interested in simulating existing acoustic instruments. These are the people who write orchestral material one part at a time, as if they were writing for a real orchestra (no-one named "String Ensemble" sits in a real orchestra. Instead you have first violins, second violins, violas, etc., etc.)

These users have always to the rest of the users of virtual instrument libraries a bit like Live Steam enthusiasts are to the Model Railroad community; admired for their technical perseverance, but rarely emulated. The vast majority of patches then and today are more likely to be labeled "Fanfare Brass" than "Eb Flugelhorn with straight mute," more likely to be properly characterized as "Psycho Strings" than "4 vlns spicc. con sordino."

There was a time during the golden age of sample libraries where a computer-mocked up orchestral score might be used in a television show, or a Hollywood movie. But that is changing. There are fewer symphony orchestras, more musicians looking for work, and the total price of properties (like big name games) has gone up to the point where it makes more sense to hire a real orchestra instead of paying someone to spend months tweaking synth patches.

As far as I can tell, the major commercial use of these things now is in mocking up score that will later be performed by actual musicians, or in overdubbing and padding out scores that are largely realized with real musicians. Which has implications as to how they are being used.

You can get half-assed "realistic" results with fairly basic work. You can add a few more percentile by going into hand-editing of expression curves and attacks and portamento rates and adding slurs and doits and grace notes...but it isn't worth it to the working users.

So the majority users aren't doing this kind of extensive editing. There is no call for it. And if there is no real call in most modern applications for extensively hand-editing a flute line, there is no real call for writing out each chair of a string section independently. Not unless you are testing a mock-up that you mean to have a string section perform, or unless you are working in a style that really requires all that internal movement.

And for the casual users, it is an even stronger impetus away from the most realistic simulation of individual instruments possible, and towards sections, pre-made swells and crescendos, and even entire canned phrases.

Back in the day one of the options was Band in a Box. Think of it a bit like a MIDI version of loops. The flexibility lay in the way the material was created, and that you were in the MIDI domain; you could assign the parts to the patches of your choice, as well as alter the root and tempo however you needed.

What is lost of course is the realism a player imparts on a performance. I tinkered a bit with an idea that I was going to include in a story I was writing; something I called "Intellisynth." The idea was a virtual instrument with software intelligence and a library of abstracted performance capture. Say you had an Intellisynth violin. You'd play the part from the keyboard. The Intellisynth would determine where to use up-bow or down-bow, which string made the most sense for each note in a sequence (obviously this is not done in real time, as it requires look-ahead), and attempts to detect phrases and lines to give each phrase an overall contour.

This is being done a bit; I have a freeware VSTi called "Spicy Guitar" that auto-detects when a chord is played and executes a strum. I believe it also detects the intervals being played and choses to throw in some fret noise as appropriate, but I could be mistaken!

The direction the real world went, though, is in canned loops. In entire sampled performances (or MIDI-sampled; akin to the old player piano performance capture rolls). Which are presented to the modern player as complete material -- which includes as is the nature of the beast those elements of tonic and harmonic development and even motive that the classically-trained composer would want to have under their control.

But the nature of the beast has changed. There are still music schools, there are still people composing and still sometimes even sponsored to write a concerto or two. But the vast majority of the people making music are now coming back much closer to a 19th century performance model; creating for their small circle of friends (that is, to whatever segment of the intertwined social media outlets they are presenting music within).

For these, loops are nothing; even remixes are completely acceptable. And that, I think, is one of the factors leading to what seems like an increasing desperation on the part of the big sample-based library producers.

We saw prices already falling before the crash. They have not recovered, although the price of libraries seems to have finally stabilized. The engines have also gone into Ballmer mode where they keep bringing out new and different versions in hopes of recapturing some sales. Of course the driver is also a move away from the laptop and MIDI keyboard controller, and into iOs. Because when most of your market is plugging loops together, or performing sections or pads, they don't need an 88-key semi-weighted controller and a DSP box capable of playing back all the parts of a symphony orchestra in real time.

And the picture above would make a nice poster boy for that kind of flailing. Except that the library in question is actually quite decent. It is marketed like some "cash in quick" collection of random synth pads and Lord of the Rings-sounding sections, but it is actually an honest and very playable library of medieval and renaissance instruments.

And that's another thing. The heyday of the giant sample libraries was also a nadir in playability. I am of the opinion that the stretching and smoothing of LAS -- that was forced by the relative sparsity of RAM in those days -- traded realism and character for playability. An 8-gig sampled library can give you a fairly realistic flute, but the sound is thinner and it is more awkward to make a lilting line with it. As soon as you realize that realistic simulation is not the only goal, you realize you could make a piece of listenable music faster with the older libraries.

This, at least, seems to be finally changing. Stretch synthesis, better legato smoothing, and sample libraries that depart from realism to make up for some of the deficiencies in the real instrument (do you want a lute sound, or do you really want a lute that is as likely to slip out of tune, and has the same limited range, as the original?)

And this essay is way too long.

No comments:

Post a Comment