Envelope follower with different attack and release

References : Posted by Bram
Notes :
xxxx_in_ms is xxxx in milliseconds ;-)
Code :

attack_coef = exp(log(0.01)/( attack_in_ms * samplerate * 0.001));
release_coef = exp(log(0.01)/( release_in_ms * samplerate * 0.001));
envelope = 0.0;


tmp = fabs(in);
if(tmp > envelope)
    envelope = attack_coef * (envelope - tmp) + tmp;
    envelope = release_coef * (envelope - tmp) + tmp;

from : jm[AT]kampsax[DOT]dtu[DOT]dk
comment : the expressions of the form: xxxx_coef = exp(log(0.01)/( xxxx_in_ms * samplerate * 0.001)); can be simplified a little bit to: xxxx_coef = pow(0.01, 1.0/( xxxx_in_ms * samplerate * 0.001));

from : kainhart[AT]hotmail[DOT]com
comment : Excuse me if I'm asking a lame question but is the envelope variable the output for the given input sample? Also would this algorithm apply to each channel independently for a stereo signal? One more question what is an Envelope Follower, what does it sound like?

from : yanyuqiang[AT]hotmail[DOT]com
comment : What's the difference between this one and the one you posted named 'Envelope detector'? Different definiton? What's the exact definition of release time and attack time?

from : scoofy[AT]inf[DOT]elte[DOT]hu
comment : Here the definition of the attack/release time is the time for the envelope to fall from 100% to 1%. In the other version, the definition is for the envelope to fall from 100% to 36.7%. So in this one the envelope is about 4.6 times faster.

from : dtpietrzak[AT]gmail[DOT]com
comment : What if you want 0ms attack with > 0ms release? 0ms attack in the current equation results in attack_coef = exp ( log(0.01) / 0 ) ) which is not a number due to the division by 0. Would using attack_coef = 0 instead result in an accurate algorithm? Thanks.