In Podcast 117 we discussed Exponential Audio’s Michael Carnes’ comments he made in a story we posted. Michael has since come back with a response to our discussion….
I listened to podcast 117 with a rising level of bemusement as you riffed on my ability to generate cabinet models while avoiding the use of convolution. I remember my remarks on convolution as perhaps being strong, but applying only to reverb and not in a more general sense. Let me take this opportunity to clarify. It’s true that I’m not a big fan of convolution in reverb. Besides the fact that it models a complex space with measurements taken from only a few point sources, I also believe it’s overkill in terms of the way we actually perceive space. The auditory cortex is masterful at reducing data before presenting it to the brain, and I suspect that reverberation is heard only after the vast majority of data are discarded. We hear in a sort of ‘shorthand’ and I think it’s more important to try and understand what gives us the perception of space than to try to follow a pure mathematical model. Now when we think of cabinet models as recorded, we’re listening to the output of a microphone or two. While the actual sound in the studio might be complex, the complete reality of the recording is the output of those microphones. As such we can think of the whole system—guitar, amp, transformer, cabinet, microphones—as a filter. And convolution is actually a very good starting point for that sort of filter. It is much more capable at duplicating complex response curves than a more simple IIR (Infinite Impulse Response) filter. And, up to a point, efficiency is not a huge problem. That breaking point is where the impulse response is also capturing a room in addition to the cabinet. This problem is exacerbated as sampling rates go up. You correctly identify the fact that most convolvers use a static impulse response, while the reality is dynamic and considerably more complex. Nothing much in the real world is linear, and that’s especially true of speakers, mic diaphragms, and things of that nature. And while there are issues to consider in creating a dynamic convolver, it’s certainly not impossible. It might turn out to be among the best approaches. The room part of the sound may be best handled by a reverberator of some type, but convolution is a great approach for modeling the physical chain. I’m flattered that my name came up in the discussion, but there are many people working in that field who deserve the real credit for what they’ve done and what I’m sure they’re working on.