On Saturday 29 January 2005 00.00, Jan Depner wrote: Well, using a recorded impulse response with some sort of convolution algorithm works for real rooms, so why not virtual ones? However, you'd probably get some rather substantial "tweak latency", even with various shortcuts taken... And either way, at the cost of the kind of CPU power required to do this in real time, you might as well get a few more h/w reverbs! Maybe that'll change eventually, but since the expectations on quality tend to track the state of the art, the net result is that faster CPUs and DSPs just make everything harder to implement! ;-D //David Olofson - Programmer, Composer, Open Source Advocate .- Audiality -----------------------------------------------. | Free/Open Source audio engine for games and multimedia. | | MIDI, modular synthesis, real time effects, scripting,... | `-----------------------------------> http://audiality.org -' --- http://olofson.net --- http://www.reologica.se ---

On fre, 2005-02-04 at 15:56 +0100, Christian Henz wrote: I think this method will not catch the mild doppler effect arising from turbolence? (.. nor the *wild* turbolence, if house is on fire!) Again, will this work at all? The (well-working) designs I have seen, all use time-modulated delays in the 'tank'. Every sampled impulse response would be slightly different from the previous. -- ( ) c[] // Jens M Andreasen

On Friday 04 February 2005 19.52, Jens M Andreasen wrote: reverbs... ;-) convolution I guess not, because that's kind of equivalent to the room changing shape all the time, I think - and you can't capture that without a constantly "morphing" impulse response. No idea how you would go about recording something like that, though. I *do* have an idea about what kind of problems that would pull in, though: It would be like looping a sampled waveform, except each sample is an impulse response... So, you need a nice, click free, non repetitive sounding loop. :-) that outboard I'm quite sure it can work really rather well. This is a well known way to "cheat" to save cycles. Most traditional reverbs use a small numbers of allpass filters/delay lines, hooked up in various ways. Some or all of these form feedback structures, to generate the reverb tail. With a low number of filters, there are only so many possible feedback delay permutations, so you'll almost certainly get some strong resonances. (Usually *exactly* where you don't want them, of course... *heh*) If tuned properly, this modulation "trick" reduces the metallic sound caused by these resonances, so you can get away with an otherwise insufficient number of filters and/or a suboptimal tuning. (*) Now, convolution based reverbs don't need this (at least not unless you want to simplate a room with moving air, that is :-), because they don't have feedback loops, and thus, they don't have these undesired and hard to tune resonaces in the first place. If the impulse is nice, so is the reverb, basically. WYHIWYG - What You Hear Is What You Get. :-) (*) If there is such a thing as an optimal tuning at all, that is. To me it seems pointless to define, since it depends entirely on what type of sounds you're going to put through the reverb, and what kind of results you want. Real rooms are not "perfect", so what's desired is not a "perfect" reverb, but rather a perfect simulation of an imperfect but nice sounding room. //David Olofson - Programmer, Composer, Open Source Advocate .- Audiality -----------------------------------------------. | Free/Open Source audio engine for games and multimedia. | | MIDI, modular synthesis, real time effects, scripting,... | `-----------------------------------> http://audiality.org -' --- http://olofson.net --- http://www.reologica.se ---

On fre, 2005-02-04 at 22:09 +0100, David Olofson wrote: [...] Neither do I .. and I think it would literally take forever. But since the wigglings and wobblings in the reverb-tail are all slight variations on a common theme, I might like to try some of the following: Use a 'wobble-generator' to crossfade between two (or more) samples taken at different locations. Cheat and 'Wobble' the input and 'wiggle' the tail slightly. Good thinking! Click free loop can be automated. Quantifying 'nice' is harder ... Non repetitive ... Say we morph between two 'nice' loops: Loop A is 2.00 seconds Loop B is 2.71 seconds Crossfade 3.14 seconds (full circle) ... then we would get in the right ballpark for a simulation of the sameness, though everchanging sound of complex sustained notes. This is for the sustained part though, and might not translate verbatim to the decay of a room? [...] Or perhaps it is: A simplified model of a rooms characteristic decay combined with a simplified model of the turbolence, sounds better than without the simplified turbolence. In a parallel universe it could argued that, a speech-synthesizer singing 'Lieder' is much more convincing if you have the appropiate Bontempi Piano to go along with it. -- ( ) c[] // Jens M Andreasen