While playing around with fffb~, I noticed that it alters the sound quite a lot even when all the bands are set to flat.

With musical input, you can hear the difference when putting fffb~ in the signal chain.

Further experiments are demonstrated in the patch below. Ideally, the signal going through fffb~ and the signal bypassing it should be identical, meaning they should cancel each other out.

The patch shows that this is never the case for the full spectrum: at different amplitude ratios of the bypassed and filtered signals, different parts of the spectrum are canceled out.

Is there a way to improve the linearity of the frequency response of fffb~?

While playing around with fffb~, I noticed that it alters the sound quite a lot even when all the bands are set to flat.
With musical input, you can hear the difference when putting fffb~ in the signal chain.

Further experiments are demonstrated in the patch below. Ideally, the signal going through fffb~ and the signal bypassing it should be identical, meaning they should cancel each other out.
The patch shows that this is never the case for the full spectrum: at different amplitude ratios of the bypassed and filtered signals, different parts of the spectrum are canceled out.

bad-quality-non-linearity-of-fffb

Your problem is with phase response, not frequency response. Unfortunately, there is no way to have a flat phase response with an infinite impulse response filter such as reson~ or fffb~ (or biquad~, fwiw).

The only way to achieve it is using a finite impulse response filter such as buffir~, which in turns is much more computationally expensive and introduces latency.

Thanks andrea. I'll look into buffir~, computer power or latency are not that big of a problem.

@ Roman: Can you explain what you mean by a symmetrical setup?

Of course if I put a flat fffb~ on the 'bypass' channel as well, the two sides will cancel each other out just fine.

But this still leaves me with the unwanted change in phase response.

Would an fft-based approach as in the forbidden planet example yield better results? (I'll try it out anyway, but maybe someone can shoot down the idea before I try).

What I'm trying to do is build a patch that corrects the frequency response of a speaker system by taking a 1/3 octave full-band noise measurement and applying the inverse with a 1/3 octave EQ.

@ Roman: Can you explain what you mean by a symmetrical setup?
Of course if I put a flat fffb~ on the 'bypass' channel as well, the two sides will cancel each other out just fine.
But this still leaves me with the unwanted change in phase response.

What I'm trying to do is build a patch that corrects the frequency response of a speaker system by taking a 1/3 octave full-band noise measurement and applying the inverse with a 1/3 octave EQ.


Bad quality / non-linearity of fffb~