DSP Robotics Support

Well - I was in Dozius filter collection, but somewhat did not found what was looking for (or did not understood what I heard on output?). The concept I work on is thought not for instrumental designs, so there are totally different challanges in the game. With production, I'd like to go beyond my older projects. More spatialization and location, more individual character of effects, greater variety of shifts between sounds, etc. FFT, node-based filter would be an interesting thing, like cool edit / adobe audition used (even if it was without graphical interface, only text input).

Aha!

I really like this ambient stuff and thik i kind of understand what you are talking about. I can try to do some work tonight and setup two VST modules ready for modulation of f,q,g in LR, mid and side options for you. One will have OS and the other double precision.

I do not know Adobe Audition or Cool Edit but I can put together some stuff I think you will have fun with.

Thank you for your interest in that, but I almost have the gui completed, so I'm not too far away from the goal. Filter guts inside are working for tests, I'm just adjusting modulation routines, to have all possible options. I really would like to finish that one, because that way I will learn how it works, what to expect, what/how to modify, and so on.

Tell me something. In your filter pack, I see there are 3 copies of coeffs. But from what I could understand - coeffs are the same for all versions, yes? So in essence I can use just one copy and connect it to all 3 engines?

A question about OS versions.

I have somewhere the "oversampling_toolkit_v1.1.osm" from SM. There are "Polyphase IIR", "Triangular" and "Power Series" inside. As far I understand, the practical difference between them - is in the way how they do dithering (shaping the stream so to speak) in resampling routine? Or are there any other differences? As far I understand, I just need to switch pairs of the up/downsamplers, yes?

I also found that I have "oversampled_biquad_v01.osm". What would be (in terms of practical sound quality) the difference between 2x and 4x oversampled version? (I mean - worth to give a try?)

I'm trying to get the best quality of sound.

BTW, you wrote you use "Transpose-Form IIR Filter Structure". And the characteristics/differences (in terms of affecting sound) of this one in regards to other present on SM forum are...?

I ask, because from my experience - ways of shaping and dithering really matters if you wish to get a good results. Not everyone gives attentnion to it.

Tell me something. In your filter pack, I see there are 3 copies of coeffs. But from what I could understand - coeffs are the same for all versions, yes? So in essence I can use just one copy and connect it to all 3 engines?

The coeffs are identical except that for the OS one you nedd to downscale input frequency before the coeffs. Single and double can share.

I have somewhere the "oversampling_toolkit_v1.1.osm" from SM. There are "Polyphase IIR", "Triangular" and "Power Series" inside. As far I understand, the practical difference between them - is in the way how they do dithering (shaping the stream so to speak) in resampling routine? Or are there any other differences? As far I understand, I just need to switch pairs of the up/downsamplers, yes?

The difference is severe. "Polyphase IIR" has the best quality in all aspects but is CPU heavy. "Triangular" is a little simpler but is ok form many implementations and is less CPU hungry. "Power Series" are the cheapest and is not that good sounding but is very light on CPU. It is notting about dithering. Dithering is noise added to a signal to prevent denormals and smooth out tiny digital artifacts when reducing bit depth.

I also found that I have "oversampled_biquad_v01.osm". What would be (in terms of practical sound quality) the difference between 2x and 4x oversampled version? (I mean - worth to give a try?)

No, I dont see the point the gain with 4x is minimal and beyond what most people can percive. At 2x the gain is to move nyquist to 44100 instead of 22050 that makes an audible difference.

BTW, you wrote you use "Transpose-Form IIR Filter Structure". And the characteristics/differences (in terms of affecting sound) of this one in regards to other present on SM forum are...?

Filter response is the same as other biquad structures. My experience tell me it has a lower noisefloor (testing in VST Plugin Analyser) and it use less memory.

*Edit: When I think it over I am a little unsure about the noisefloor. It is a long time since I did test and evaluate the different structures. But I do for sure recall it was some advantage in either stability or som other aspect of audio quality besides the saved memory.

Okay, thanks for explanations. If coeffs are identical, the this simplifies a lot. Scaling frequency is then x4 I guess.

As for dithering and noise shaping in mixing/resampling, I noticed, that some apps just put these two terms into one box. I can understand that, although it produces confusion in discussions. If you resample or mix something, then shaping is often added, and it decides on how the "edges" of the wave are processed (like in image processing - smoothing vs shaping) thus - it seems that this may be considered as dithering too, because it produces different "noise floor". I have not checked yet, I probably will test all 3 options. Sometimes the "worst" or "mid" quality adds that specific taste to certain types of sound, like analog noises.

As for oversampling, I probably will try that x4 version too. I'm among these folks who may strongly notice that difference... I suspect that x4 may produce that "cloud-like nuance in subtleness" that makes a huge difference in my productions. From what I see - in order to adapt the filters into x4 oversampling - I just need to duplicate the formula inside the code box for 2 more ins/outs?

nope it is not that simple. and what exactly code do you refer to?

2x OS Biquad DF2

If the coeffs are the same for both - OS and other 2, then the only difference is in engine?

Because this oversampling architecture (attached file) would suggest so. I assume that here is somewhat different filter than yours.

A short note. The hi/low shelf filters should have limiter for "e" factor (greater than zero, like 1e-10). From what I remember in the recent past - combination of gain and resonance beyond boundaries - will produce e<0 which gives an error on sqrt part. Having the limiter - Q can be set to 100 for all filters, and shelf one will not die.

I guess you are right about the limiting of e. I never thought of it as I have not operated with high Q's. This is how I would do the 4*OS biquad in DF2:

Your solution looks as I would approach it - just the engine is doubled for two additional routes. Okay, thanks. I see where it leads me. I guess tonight or tomorrow I will do some checkups on how it sounds.

As for Qs - I use bandpass butterworth filters with Qs around 500-1000 and initially some folks said that it's crazy. But to be more precise - it is used as a band limiter, to small spikes in spectra. As far I remember - this topic is discussed in the same thread where the Dozius filters are located.