- ✔ Frequency points × x's are calculated repeatedly for each X, so just collapse already known data. Do not recalculate already known things.
- Overcalculation happens, but happens per-fragment.
- Each formant randomly moves by unit circle, and takes value from the initial signal by that.
- I guess that I can do soundprint by approximating initial noise by formants. By that I get spectrum approximation and