DSP Robotics Support

The concept. I need a pair of filters, that will separate certain part of spectra for further processing: at given "Q" factor, generally good separation (but not aggressive/sharp in terms of sounding; rather pleasant smooth mix), and not too much of CPU consumption. The separation itself (mix with no processing) should generally give - "flat" audible results so to speak. My guess is, that bandpass and bandstop butterworth should do the job. I used filters reworked on SM forum for tests.

I checked green 6 pole variant (enough per octave? too much?) and such sort effects I would describe as fine (but I might be wrong if there something else is going on with phases?). I'm trying to rebuild them into blue version. Now the thing is - I can't get at least similar flat response. Plus bandstop part produces bad artifacts. One of my guesses is, that detuning part produces differences (I used one, that I use for high Q in one of my projects). Other guess - these filters are just different from green ones (too much for me to recognize just by looking at them).

Can someone help with this part?

Found something on SM/FS forums called Linkwitz-Riley xover with 3 bands / 2 cut freqs, and it appears that this is the direction for such flat band extraction. Converting into blue version and testing how it behaves in practice.

Puff... I think this is it.

Question 1.

Can this be improved (filter section)?

Question 2.

I made also higher order (96dB) in similar way, but blue version produces audible artifacts (while green not). It was rather expected. I guess this is due to precision differences between green and blue things, because this sort of thing was mentioned in description info of source filters I used. Can this sort of higher order filter be done without artifacts? I haven't played yet with mid versions (between 24 and 96dB), so I'm not sure where is the optimum limit for such cross-over, to get "nice sounding" results during further processing and mixing. But maybe someone has experience with it?

Question 3.

Having somewhere else a bandpass filter made of LP/HP BW's, capable of producing very narrow bands, I'm somewhat confused here. I noticed, that above certain Q factor (i.e. "external Q" recalculated for extracted band which will be processed by other stuff and then mixed in), like over 10 - there is no audible difference on extracted midband. Either I did something wrong, or this is normal for such type of crossover? I would rather expect, that LP/HP will affect general gain of the midband, crossing the center frequency?

Question 4.

How to "translate" Q into something (+/- blue spread around center value; will be modulated), that will produce "linear" audio response within usable range? I guess it has to do with logarithms and decibels, but still not sure of that...

*

Yep, I feel dumb... Maybe less than yesterday, but though...

Stupid me!

1. Bandpass (or whatever else filter for band extraction).
2. Substract filtered part from the stream to get the rest
3. Process the bandpass part
4. Mix processed badnpass part with the rest

Perfect mix, no crossover issues.