aceslowman

I am wondering, I have a system of a number of cascading filters, and the amplitude of the signal becomes very small, very fast. What advice might some of you have for correcting the gain appropriately? Should, for example, I get the signal back up to -1,1 before my signal goes down another branch of filters? Or should I just wait until the output?

correcting-gain-or-normalizing-after-filters

Also, in the case of a filter attenuating the signal to below 0, is it appropriate to just multiply it by a negative number?

And on top of that, is it important to consider the DC offset before each filter, or should I just correct that at the end of the series? Or if using gen, could dcblock be used after each filter?

there is no formula for gain correction after filters because the max peak output is dependent on the input material, too. so an autogain system is your best bet. i personally would prefer a user controllable gain regulator in most cases.

i´ve never heard (of) signals smaller than 0.0 and i have also not yet seen them (probably because they are too small to see them)

about DC... what filters are we talking about? biquads and similar designs will not cause something like a DC offset. and DC removal is nothing more than another highpassfilter itself. so if it really matters, i would put it at the beginning, and you need it only when no other highpass is in the chain.

(i am not sure if DC offset can come to play when you setup a -96 db/A lowpass at 20 Hz. but if you would now DC-correct it using a highpass at 5 Hz, you will also kill the wanted part of the signal.^^)

(i am not sure if DC offset  can come to play when you setup a -96 db/A lowpass at 20 Hz. but if you would now DC-correct it using a highpass at 5 Hz, you will also kill the wanted part of the signal.^^)


My filters are based off of some specs from a paper outlining a Klatt voice synth. I pasted the gen code for it. And sorry for the confusing wording, I didn't mean the signal is smaller than 0.0, it just has a negative offset (all signal is negative). What I am putting in as input though, might be the root of some of this. It is from 0 - 1, a series of impulses. Should I scale the signal to fit -1,1?

Here is a better illustration, using the same impulse train as the synth I'm working on.

do you want to filter the impulse train only? then you should of course not move it and leave it unipolar.

Correct, just filtering the impulse train, to give it that particular contour. Am I mistaken to try to push that impulse train through similar filters in the same way after this?

I found that adding a dcblock right before my codeblock (for the resonant filter) is fixing this. I was skeptical at first because I watched my unipolar signal become bipolar, but as far as tone goes... it seems to be working. Is it appropriate to put a dcblock before all of these gen filters? Or is it going to come down to their purpose along the signal flow of the synth?

Maybe something like [tap.adapt~] would help?

Correcting gain or normalizing after filters