ansaurus

Question

Answer 1

+5 A:

Robert Harvey 2009-07-12 23:56:50

Does freeverb3 provide tools for you to go out and measure the impulse response yourself, or does it just use data they measured to generate the reverb effects?

MusiGenesis 2009-07-13 00:00:29

Yes. I'm aware of that project. I'd like to see at least one other project for ideas, as I don't just want to copy code.

Nosredna 2009-07-13 00:11:06

@MusiGenesis, there are a lot of libraries of impulses. I don't know if freeverb3 includes a way to generate them, but there are plenty of tools for that. Some people have traveled to great churches to get impulses.

Nosredna 2009-07-13 00:12:45

No, Freeverb3 just processes the samples. It takes a lot of specialized equipment to produce good impulse response samples, see http://web.arch.usyd.edu.au/~basse_j/auditoria.html. But there are apparently plenty of third-party sources for these samples, see http://freeverb3.sourceforge.net/

Robert Harvey 2009-07-13 00:12:54

Have you seen this article? http://www.soundonsound.com/sos/Oct01/articles/advancedreverb1.asp If it's as easy to create algorithms for this stuff as MusiGenesis claims it is, you should do well with these ideas.

Robert Harvey 2009-07-13 00:28:06

I printed that article out about five years ago but had since forgotten it. Thanks. Printed again. :-)

Nosredna 2009-07-13 00:38:12

Great article. Thanks.

MusiGenesis 2009-07-13 00:44:17

Well, clearly I'm going to go from Freeverb3 at least in part. Any other source code pointers are welcome. For now, I'll accept this answer.

Nosredna 2009-07-13 00:53:15

Answer 2

A:

Are you kidding? Reverb is the easiest thing in the world to do programatically:

for (int i = 0; i < input.Length; i++)
{
    output[i] += input[i];
    output[i + delay] += input[i] * decay; 
}

I write this kind of stuff full time now, so maybe it just seems easy. Do you mean you're looking for more general echo or spatial effects, that might include frequency-modulated delay lines and chorusing and so on?

MusiGenesis 2009-07-12 23:57:47

Will this produce a realistic reverb sound? Most people listen with two ears; how do you spread it across the spatial field?

Robert Harvey 2009-07-13 00:00:29

A simple trick is to just slightly vary the delay and decay parameters on the left and right channels.

MusiGenesis 2009-07-13 00:01:42

The result of the above function is "realistic" in the sense that it sounds like an analog reverb device, which means it does not sound like a real space. You can create a more realistic spatial echo sound by producing a few discrete echoes (single echoes instead of reverb) at short distances and then blurring all the echoes together in a variety of ways.

MusiGenesis 2009-07-13 00:04:20

Real-world echo effects usually consist of a few sharp, relatively undistorted repeats of the original sound, followed by a muddy pitch-shifted cloud of echoes.

MusiGenesis 2009-07-13 00:06:03

I'm not kidding. Most of the algorithms I try sound inferior to good hardware reverbs. Often metallic. There are entire conversations on musicdsp and kvraudio about how hard it is to get a good musical-sounding reverb.Traditional way includes multiple allpass filters.And in my experience, you need to modulate the delay lines in various ways.Basically, I have the feeling there's black art involved in the best reverbs, and I'm looking for some hints.So I'm not looking for the "easiest reverb in the world." I'm looking for some inspiration from other people's good code.

Nosredna 2009-07-13 00:09:50

It is definitely a black art. Check out my website (the link is in my profile - I subscribe to the "no self-serving links on SO" policy). MusiGenesis has three different "reverb" settings that all use variations of the above reverb method. One of the basic tricks is to match the delay times of the reverb lines to the tempo of the music.

MusiGenesis 2009-07-13 00:17:25

I've heard that the delay taps should be prime numbers so they don't clump up. Do you do that? Do you happen to have any source code to show?

Nosredna 2009-07-13 00:19:49

Modulating the delay lines is also critical, as you pointed out. The "black art" part is picking modulation frequencies and depths that either don't reinforce each other or reinforce each other in the way that you want. The usual consquence of picking the wrong values is phase effects.

MusiGenesis 2009-07-13 00:20:06

@Nosdredna: sorry, my stuff is all proprietary, and it might not be useful to you anyway. Most people that do audio are interested in real-time synthesis, whereas my code all works on full recorded samples. I enjoy discussing this stuff in principle, though, and I'll be happy to stear you towards any resources I know about. Harmony-Central is an excellent site for this stuff (or it was 10 years ago).

MusiGenesis 2009-07-13 00:23:09

Do you use all-pass filters to correct phase changes that come from the first stage?

Nosredna 2009-07-13 00:23:43

I've tried the prime number thing, although my algorithms are more like what I mentioned in my 3rd comment - with single echoes, it doesn't matter whether they're prime or not. My usual approach to designing reverb is like most of what I do: I try stuff out randomly until I find something that happens to work.

MusiGenesis 2009-07-13 00:25:14

That's OK. My ears are open to hints and experiences, despite my preference for working code to play with. :-)

Nosredna 2009-07-13 00:25:45

"I try stuff out randomly until I find something that happens to work." OK. That's what I suspected. I've heard that some people have found ways to choose the values to get good sounds. Especially important if you support multiple samplerates.

Nosredna 2009-07-13 00:26:50

I generally avoid using FFT wherever possible, so filters are sort of rare in my programs, which are meant to run on Windows Mobile smartphones. Most of what I do involves fast "tricks" on time-domain data.

MusiGenesis 2009-07-13 00:27:39

I'd love to share my code, but I'm trying to make a living from it, so... Maybe I'll open-source my project and rake in millions from Google ads. I actually have this fantasy once in a while, along with winning the lottery.

MusiGenesis 2009-07-13 00:29:09

+1 for a good discussion, albeit no source code (boo hoo) :)

Robert Harvey 2009-07-13 00:29:58

Actually, filters don't have the use FFT. Most DSP filters are IRR for speed. Example: http://www.musicdsp.org/showone.php?id=24

Nosredna 2009-07-13 00:31:52

A simple thing I figured out for designing a chorus effect (which is multiple delay lines with different modulating frequencies and depths) is to just plot the modulators on a single graph in an additive way, so that the graph is bright if all modulators are near 1 at the point, and dark of all modulators are near -1. If the combined graph is relatively smooth (without a lot of peaks and valleys) the chorus will sound good.

MusiGenesis 2009-07-13 00:32:50

@Robert Harvey. I'm counting on you to pull out 30 lines of angelic C code at any moment. :-)

Nosredna 2009-07-13 00:32:56

@Robert: hey, there's a little bit of source code up at the top, here. And it does produce an undeniable reverb effect. :)

MusiGenesis 2009-07-13 00:35:10

@Nosredna: that might be the best link anyone has ever sent me. I've been trying to emulate the Moog (with limited success), and I can't wait to see what effect that method has. Tomorrow should be a fun morning.

MusiGenesis 2009-07-13 00:36:44

@MusicGenesis: Skim through the whole musicmsp archive. You'll love it. And hang out at kvr with the dsp programmers... http://www.kvraudio.com/forum/viewforum.php?f=33

Nosredna 2009-07-13 00:42:57

Ah, the joys of the occasional random down-vote. :)

MusiGenesis 2010-05-14 19:07:24

Answer 3

+3 A:

You could do a lot worse than read John Dattorro's paper on the subject found here, on his homepage. Dattorro worked at Lexicon and the paper I've referenced includes extensive discussion on the design of high quality reverb.

Outside of this, the various links at musicdsp, and scant reference in the literature, the design of great reverb is shrouded in secrecy. The finest reverbs are designed either by people who have worked with the designers of the last generation of great reverbs, or by obsessives who invest extraordinary quantities of time into the subject. In either case, the designers seem to become quite tight-lipped regarding their methodologies.

Dave Gamble 2009-07-13 01:08:53

Yeah. That one I already have in a notebook.

Nosredna 2009-07-13 17:06:36

Answer 4

+2 A:

realistic reverberation algorithms are a bit of the 'holy grail' of audio DSP programming... there are two basic approaches in the pro-audio market today:

convolution reverb (using impulse responses)
delay/feedback/dampening networks

the main challenge behind impulse response convolution has been the efficiency versus quality tradeoff (incl. latency!). whereas the main challenge behind delay matrix networks has been generating vast lattices of delays with little harmonic re-inforcement.

professionals pay vast amounts of money for realistic sounding reverbs... a "good" sounding reverberator can retail for $2000+, and "really good" ones for much more.

welcome to the pro-audio industry...

kent 2009-07-14 11:51:06

I guess that's the answer I expected. It's a hard problem.

Nosredna 2009-07-14 14:48:31

ansaurus

tags:

views:

answers:

Good audio reverb source?

related questions