AES Technology Update, Part 3

By Craig Anderton

Sometimes technology is about a dramatic breakthrough that changes everything: multitrack recording, FM synthesis, digital signal processing, and the like. But sometimes it's more about reaching the tipping point, where what once was esoteric and expensive is now standard. So let's look at convolution reverb now that it has truly reached critical mass, the evolution from single core to dual-core processing, and the ultimate...uh..."noise reduction" system. You'll see what we mean.

The Convolution Revolution

The process of convolution - where two spectra are multiplied together - has been around for some time. I remember doing convolution processing on an E-mu Emax when it first came out; sometimes it would take over an hour to do all the calculations, but the wild, wacky sounds were often worth it. And when it wasn't, well, another hour down the drain.

So why does anyone care, anyway? Convolution-based reverb is to standard digital reverb as sampling is to synthesis: Higher realism, with the tradeoff being slightly less programming flexibility. Convolution reverb takes a "sample" (called an impulse) of an acoustic space, and multiplies that with a signal so that the acoustic space becomes part of that signal's "sonic signature." Nor are you limited to taking impulses of acoustic spaces; impulses are available of classic spring reverb units, amplifier cabinets, mics, whatever...the sky's the limit, and the resulting sound is exceptionally realistic (given a well-recorded impulse, of course - the law of "garbage in, garbage out" applies to impulses, too).

Over the years, as processing got faster, the number-crunching needed to do convolution became more practical. Convolution became a part of programs like Cool Edit Pro, but while pretty speedy by older standards, it still wasn't a real time process.

Well, it still isn't...but latencies have shrunk from hundreds of milliseconds to as low as the single digits. And convolution reverb is being built in to programs as a highly desirable freebie: TASCAM's GigaSampler 3 and Native Instruments' Kontakt 2 are examples of instruments that include convolution. Furthermore, Cakewalk Sonar 5 is the latest DAW to add convolution processing to its bundled arsenal of signal processors.

Bottom line: Convolution used to be an expensive, slow, processor hog; now it's everywhere. Better yet, convolution reverb is becoming more editable - almost to the point of matching conventional, synthesized digital reverb. Indeed, technology marches on.

Advertisement

Clean Up Time

Noise reduction is another technology that has gone from the esoteric to the commonplace. But what Zenph Studios showed at AES was a dramatically different take on cleaning up old recordings. You've heard of re-mixing and re-amping? Well, get ready for "re-performing."

The Zenph process uses software to analyze a piano recording - even one with noise, hiss, or other poor recording quality - then converts the audio into digital data capable of driving Yamaha's Disklavier Pro. As a result, you hear a piano performance that is not as much a re-creation as an actual "re-performance" of the original piece, at the moment of its creation.

Cleaning up archival recordings electronically can only go so far; while it's amazing how today's noise reduction algorithms can surgically remove the noise while leaving the "meat" alone, Zenph's approach takes the recording back to the source itself. In the demos at AES, the ability of the Zenph process to extract music from even pretty bad recordings was remarkable.

There are other important points. Mono recordings, by being played back over a real piano, can be miked in stereo or even surround (5.1 Glenn Gould, anyone?). The process even captures pedal movements and intricate, "humanized" timings. John Q. Walker of Zenph sums up the process as follows: "If you 'clean up' a 1920s recording, it's still in 1920s monaural sound quality. You couldn't 'start over' - until now."

Dual Core Rules

Fast is good. Faster is gooder. Fastest is goodest. Which is why dual core processors are getting so much attention: They're like dual processor systems, only faster.

This is because with two processors, each one has to know what's going on with the other one's cache - otherwise, they might end up searching for redundant data. With AMD's dual core chips, this communication happens within the chip; with a dual processor system, this communication needs to take place over a bus. (Interestingly, Intel's Pentium D chips are dual core, but communicate over a bus and therefore aren't quite as fast as the AMD offerings.)

Nor are you limited to PC offerings: You can get to the core of the Apple, too. Their new Power Mac G5 Dual uses a single dual core processor, and the Power Mac G5 Quad uses two dual core processors. That's right - a dual core, dual processor system, which means eight double-precision floating point units per computer, along with four velocity engines. Can anyone say "video rendering"? Apple can: They claim 40% faster rendering with Final Cut Pro SD compared to a dual 2.7GHz Power Mac G5, and 69% faster rendering for After Effects.

This isn't tomorrow's technology, either; dual core is readily available. For example, Open Labs used AES to debut the NeKo 64 Gen2 ($5,995) and NeKo LE Gen2 ($2,295), the industry's first 64-bit dual-core keyboards, based on Athlon 2.0GHz chips. Okay, that's impressive. Also impressive: Their RunSilent function, which dynamically controls CPU speed to reduce temperature (and fan speeds), and the complementary PowerRush technology, which detects system load and increases CPU speed on the fly as needed. Nice, as are the other specs.

Also at AES, Gibson showed their digital guitar playing with Cakewalk Sonar 5 running on HP's xw 4300 workstation - which is, natch, powered by Intel's dual core processors. Prediction: Dual core will be the nest big computer buzzword. In fact, maybe it already is!