pouët.net

switch(type)
{
case LowPass:
if (cutoff < 10.0f) { cutoff = 10.0f; }

c = 1.0f / tanf ( pi * cutoff / (float) Constants::Audio::SampleRate );

coef1.a1 = 1.0f / ( 1.0f + ( resonance * c ) + ( c * c ) );
coef1.a2 = 2.0f * coef1.a1;
coef1.b1 = 2.0f * ( 1.0f - ( c * c ) ) * coef1.a1;
coef1.b2 = ( 1.0f - ( resonance * c ) + ( c * c ) ) * coef1.a1;
break;
case HighPass:
if (cutoff > Constants::Audio::SampleRate / 2.0f) { cutoff = Constants::Audio::SampleRate / 2.0f; }

c = tanf ( pi * cutoff / ( float ) Constants::Audio::SampleRate );

coef1.a1 = 1.0f / ( 1.0f + resonance * c + c * c );
coef1.a2 = -2.0f * coef1.a1;
coef1.b1 = 2.0f * ( c * c - 1.0f ) * coef1.a1;
coef1.b2 = ( 1.0f - resonance * c + c * c ) * coef1.a1;
break;
}

And then..

output1 = ( coef1.a1 * input ) + ( coef1.a2 * in1 ) + ( coef1.a1 * in2 );
output1 -= ( coef1.b1 * out1 ) + ( coef1.b2 * out2 );

// Denormalise
output1 = undenormalize(output1);

if (changed) {
output2 = ( coef2.a1 * input ) + ( coef2.a2 * in1 ) + ( coef2.a3 * in2 );
output2 -= ( coef2.b1 * out1 ) + ( coef2.b2 * out2 );
output1 += (output2 - output1) * ((float) Constants::Audio::StepSize - s) * (1.0f / (float) Constants::Audio::StepSize);
}

// Write output
data[s] = output1;

// Write context
out2 = out1;
out1 = output1;

in2 = in1;
in1 = input;

Your topology is a direct-form I biquad with the input coefficient normalized to 1.

it works! the zero bug was a dumb casting error.

And yes.. well spotted.. you get the "I know my filters" trophy :-)

b1 = -2 and b2 = 1 when c = infinity (cutoff = 0), if I did my math correctly.
Then, both the poles of the filter are _on_ the unit circle at z=1 and form two perfect integrators. These have _very_ high gain at frequencies close to 0 Hz. Round-off noise will accumulate causing a ramp (?) at the output.

My guess is that you're not clipping the audio to the soundcard and your float->short conversion will continuously overflow, causing a full-scale saw to appear.. *ouch* :)

Am I getting close?

Can I have my "I know my z-transforms" trophy too? :)

I'll have to make one.. here.. I've got a cigarette lighter stuffed into a potato.

It's yours.

A few notes:

1. If you can live with slightly higher noise when you change the filter coefficients (ie. a filter sweep), Direct Form 2 will save you some CPU time.
2. If you use SSE math, setting the FP flags that flush denormals to zero will be a lot faster than any undenormalise routine you can write yourself.

I avoid denormals for lowpass filters by adding 1e-20 to the input.
For highpass filters, you can do the same trick, but you should change the sign of 1e-20 every sample.

I prefer to make resonant filters using cascaded integrators. They have good noise properties and the coefficients are generally easier to calculate (and interpolate) than those of biquads.

@Sesse the cpu problems were down to memory lookups on the wavetables (even though I got them small enough to fit the cache), so I'm going to leave it alone for now, but thanks for the advice.

@trc_wm: excuse my ignorance.. you're obviously a synth coder.. point me at a demo with your synth in it! I want to listen!

@Sesse: see above :-)

Buy a Waldorf Microwave II, XT, or XTk :)

or download this: http://www.scene.org/file.php?file=/parties/2006/breakpoint06/executable_music/newschool/can_i_have_my_pills_now_trc_mw.zip&fileinfo

Only synth related demoscene prod I did; I wrote the "music" in one day.
It's VA with lots of karplus strong stuff and 1 voice sample.

FFFFFUUUUUUU

MeteoriK: There's been exactly one intro released with “my” synth in it, namely Nemesis from TG04. (I say “my” because most of the credit should go towards titanstar.) There's been quite a lot of bugfixing (taking a degree with DSP in it in the meantime surely helped) and some new features since that release, but I believe I've complained enough about musicians not being interested already :-)

One day I'll understand why "FFFFFUUUUUUUUUUUUUUU" makes me lol.

I can't explain it and I'm the one who tried to make the whole of breakpoint laugh at the sundown quiz. At least there was logic to my humour.

Stephen Wright is a master of, er, taking drugs and then saying what he thinks :-)

@Sesse: nice! I reckon pouet should set up a separate forum so that rabid eczema hairy bastards like maali can't accidentally click on a link about synth programming and post pictures of attractive babies he wishes he looked like in his dirty long-haired childhood.

I'm guessing from the samples that the drums were samples? I'm not giving up on my white whale - the fucking snare drum.. if I can't do it this time, then it's a 1k BRR compressed sample.

MeteoriK: No, the drums are synthesized, save for the little shaker which is indeed a sample (compressed with the same LPC-based system as the sampled George W. Bush in the beginning).

basically, maali's parents had a problem.. dummy in one hand.. maalii in the other..

which is which? they both look like rubber.

Sesse: well your drums are better than mine! lets fight! (in a cooperatively competitive and mostly loving hell lets get gay and have a hug way)

Can't run Nemesis on my Intel MAC :(

trc_wm: get a computer!

I can't, I'm gay.