I interviewed [Eventide co-founder] Richard Factor [Tape Op #130] in 2018, but when we first met, ages ago, you were doing a rack mounted plug-in with Joe Waltz.
That’s Manifold Labs and Plugzilla. But “ages ago” fits me. I started with Eventide in ‘73.
With Eventide being one of the early makers of digital audio hardware, did you ever imagine that it would all just be in a computer?
Sure. Yeah.
Once the H910 Harmonizer came out, you were basically writing code.
To my mind, all of this is obvious. So, when computers came out, it was clear that everything someday could be in the computer. In fact, one of the first products we made was a real time one-third octave spectrum analyzer for the [Commodore] PET computer, IBM PCs, and for [Apple] Macs. You could plug a card into the computer and the computer became a test instrument. No one had done that before. And then when DSP [Digital Signal Processing] chips happened, I founded a company called Ariel and we were putting DSP cards into PCs so people could write software for them. But all of this was foretold. When I was in graduate school at CUNY, I was reading Bell Labs' journals [Bell Labs Technical Journal] and guys like [Manfred] Schroeder were saying that once you're in ones and zeros [digitized], you can do all manner of stuff. It was just a matter of the technology advancing the point where we could do things in real time at audio rate. Computers existed, but the challenge was for full spectrum audio. For 20 kHz, you want to sample at least at 50 kHz. That means every 20 microseconds, you've got a slot to do something. When Richard [Factor] designed the first shift register-based delay line, nothing was happening in the digital world. All you could do at that rate was move the bits along the chain. When I designed the first Harmonizer, RAM [Random Access Memory] had just come out, with 4 kilobytes of RAM. And all I could do at a 50 kHz or so audio sampling rate, was come up with one address to write and another address to read – two operations every sample period. When we got to the late '70s, we were still not at the point where a general purpose CPU [Central Processing Unit] could do anything at audio rate, but there were special purpose processors. There were these things called bit slice [processors]. But it came to a point where we could do about a hundred things every sample rate. And that's the point where you saw the first digital reverbs. If you have a hundred operations, real time at audio rate, you can build a bunch of delays and feedback around them – feedback networks – and simulate what a room sounds like.
Right. All the acoustics of a room.
We call that algorithmic reverb. Eventide and I did it with the SP 2016 [Signal Processor]. With the 2016 I went a little nuts, because I didn't want to just do reverb. I figured that I could do a hundred things. I could do band delays. I could do a vocoder. It was more of a general purpose processor. EMT did a digital reverb [EMT 250], the Lexicon 224, Quantec [Room Simulator] – all of those were algorithmic reverbs.
Right. They're not sampling spaces.
They're not modeling a given space; they're just giving you the sense of a real room. And what did that turn out to mean? If you built a Schroeder reverb, it doesn't sound like a room. Even if you get the early reflections right and have an FIR [Finite Impulse Response filter] for the early reflections, it still doesn't sound like a real room because he just had a bunch of delays with feedback in parallel. The echo density builds up linearly. But what happens in the real room is the echo density builds up exponentially. The way to make that happen in an algorithmic reverb is you want to interconnect delays. Today that's called an FDN [Feedback Delay Network]. I built it for the 2016, before it was called a feedback delay network, but that's what it was. I thought it was a matrix.
Sure. Technically it is.
