15 Aug
2007
15 Aug
'07
10:43 a.m.
I have now written a vocal reverb. It needs a pluginization.
An example would do fine but the example must be exactly
what I need, otherwise it may just confuse me. Latest plugin
stuff, of course.
Unlike GVerb, the new reverb has a preset structure. The reverb
is initialized this way:
effect = effect_new(config_ptr,preset_ptr);
The preset structure can be modified during the processing
by switching the pointer or by modifying the parameters
in the structure.
The audio is processed by calling
effect_do(effect, input_array, count, outleft_array, outright_array);
I'm also glueless with the following:
1. Denormalization. Feeding noise at the reverb input would
save me trouble of finding all places where the problem
occurs, but is this wise? How to generate the proper noise?
2. Effect modulation is done by two extra arrays at
effect_do( ..., mod1_array, mod2_array, ...)
They should be filled with slowly varying sin() signals, but
how to do it efficiently? I have used earlier the sin() call.
3. The audio is divided to blocks of N samples. Apparently
I should interpolate the tweaked parameters over the N samples
inside the reverb, or I should provide arrays for all parameters
and leave the interpolation to the plugin wrapper.
Juhana
--
http://music.columbia.edu/mailman/listinfo/linux-graphics-dev
for developers of open source graphics software
15 Aug
15 Aug
11:39 a.m.
Hi Juhana,
On 15 Aug 2007, at 15:43, Juhana Sadeharju wrote:
1. Denormalization. Feeding noise at the reverb input
would
save me trouble of finding all places where the problem
occurs, but is this wise? How to generate the proper noise?
You don't have to use noise, you could use a fs/2 signal at very low
amplitude, eg.
sig = 1e-30;
for (i in input)
x[i] += sig;
sig = -sig;
it's a little ugly, but works.
2. Effect modulation is done by two extra arrays at
effect_do( ..., mod1_array, mod2_array, ...)
They should be filled with slowly varying sin() signals, but
how to do it efficiently? I have used earlier the sin() call.
There's a trick to calculating sinus signals where you calculate the
sin delta value for the starting point (once) and cross sum them to
calculate the sine and cosine. Works fine as long as you don't need
to change the frequency: (100 is the frequency of the LFO)
#include <math.h>
#include <stdio.h>
int main()
{
float fs = 48000.0;
float sinv = 1.0;
float cosv = 0.0;
float delta = 2.0 * sin(M_PI * 100.0 / fs);
int i;
for (i=0; i<1000; i++) {
sinv -= delta * cosv;
cosv += delta * sinv;
printf("%f\n", sinv);
}
}
If you do you can just use a smallish table and interpolate it, or
use an approximation function.
3. The audio is divided to blocks of N samples.
Apparently
I should interpolate the tweaked parameters over the N samples
inside the reverb, or I should provide arrays for all parameters
and leave the interpolation to the plugin wrapper.
It's probably easier to do it inside your code, some of them will not
need interpolating and for the ones that do linear is generally fine.
- Steve
12:04 p.m.
On 15 Aug 2007, at 16:39, Steve Harris wrote:
Hi Juhana,
On 15 Aug 2007, at 15:43, Juhana Sadeharju wrote:
Hm, I've just realised that this is a bad idea if you do it to the
effect when its 100% dry. If someone else does it and the plugin are
chained it will either be inphase in which case it will build, or out
of phase in which case it will cancel and you'll have denormal
problems again :-/
Needs to be done with care. But as long as the effect is even
slightly wet it should be fine.
To avoid any risk there are cheap psuedorandom number generators you
can use, eg. take a large prime unsigned integer and keep multiplying
it by another large prime integer, then divide by UINT_MAX/2 and
subtract 1.0 to get a floating point value in [-1.0,1.0].
- Steve
1. Denormalization. Feeding noise at the reverb
input would
save me trouble of finding all places where the problem
occurs, but is this wise? How to generate the proper noise?
You don't have to use noise, you could use a fs/2 signal at very
low amplitude, eg.
sig = 1e-30;
for (i in input)
x[i] += sig;
sig = -sig;
6132
days inactive
6132
days old
linux-audio-dev@lists.linuxaudio.org
2 comments
2 participants
participants (2)
-
Juhana Sadeharju -
Steve Harris