Create a filter (quickly) with list of frequencies and dB
I'm working on a patch that does onset and descriptor analysis and produces a filter shape in frequencies and decibels.
I'm then creating an impulse response for it (using the HISSTools) and applying that to a sample.
Here's a little video demo-ing the idea.
https://www.youtube.com/watch?v=qvMxNJLiopE
This is working at the moment, and sounds really good. BUT the problem is that creating an IR 'on the fly' like this is a bit slow. By "slow" I mean it can take anywhere between 4-20ms, which isn't very slow, but in the context of sharp attacks and sample playback, it makes the difference between being audible and being masked by the initial attack.
So I want to see if it's possible to create the required filter without having to create an IR.
The part of the patch that creates the analysis generates a list of 20 frequencies and corresponding dBs. Here are a couple examples:
221.833218 -33.6 362.220601 -36.893817 523.987791 -40.530337 710.390753 -43.361263 925.181305 -44.610198 1172.682637 -45.796093 1457.876324 -46.420653 1786.50259 -49.265432 2165.17585 -45.506487 2601.517837 -44.333595 3104.311008 -46.474693 3683.675316 -47.684246 4351.271895 -49.680367 5120.537773 -49.981456 6006.95632 -50.487615 7028.368896 -47.975821 8205.333946 -49.788011 9561.540794 -51.071733 11124.286444 -49.155568 12925.025005 -49.777985
221.833218 -47.567153 362.220601 -45.287216 523.987791 -39.264205 710.390753 -25.401317 925.181305 -18.077634 1172.682637 -21.708286 1457.876324 -31.175341 1786.50259 -32.382667 2165.17585 -32.84 2601.517837 -36.17214 3104.311008 -35.166051 3683.675316 -37.49658 4351.271895 -37.9 5120.537773 -38.04486 6006.95632 -37.585687 7028.368896 -36.502354 8205.333946 -37.013693 9561.540794 -40.354406 11124.286444 -39.163458 12925.025005 -35.990496
Is there a way to quickly turn these into filter coefficients which I can feed into a cascade~ or something similar?
Or is there a faster way to create something along what I need here?
I have some ideas but not sure if they would prove suitable. You could use the fffb~ object and give it your list of freqs and dB (would have to convert to linear amp). You could use mc.reson~ in a similar way but the control would be different. mc.filtercoeff~ to mc.biquads~ could work as well.
Interesting! I haven't ever used (or really seen) fffb~. That might be ideal. Will mock something up and test.
Ok, this seems promising. I'm running into some issues where the filter being produced is waaay to quiet, and then I realized that part of the process I'm doing with the HISSTools is compensating for the gain of the final IR, so I need to find a way to do that in Max-land.
Trying some crude normalization of the linear amplitude version of the filters, but I'm not sure that's the correct approach.