question: finite db comparison and math best way?

[wlangfor@uoguelph.ca]

Hi all,

I've been recently working on an alogirthm to automatically subtract un-neccessary gain based upon a maximum gain limit. And I'd realized that it's a real pain, linear, db db to linear and so on.

Absmask and such and so on and so on. I have used many of these mthods; but the question is; is not making a min/max method more efficient? Like to create a min a max db and 0 to 1 being the new unit of measure?

It's an annoying thing to concieve, but of course when you take into a ccount an inverse bi-polar linear db scale; it's so frustrating to do two calculations; one for above 0 and one for below only to add them together.

Surely it's more efficient to come up with a custom mechanism. I would appreciate anyone's feedback

Thanks in advance
Robert

[wlangfor@uoguelph.ca]

I was working on a product (chieftan) lat night and came up with something that works with db's but it is not neccesarily finite math. I designed the whole project to be roughed math, roughed but seemingly perfect was the goal.

But all the same; in the 46 hours I spent trying to make a better compression algorithm with dsp; I found a lot of little things which were mathematicallly perfect which I would like to use; but the only thing holding me pack was the re-interpretation of the math.

The only thing which saved Me in the end was Sam Mungal's inuitive roughed db to lin - passive db addition asm code which is often a good solution.

Still though; I find it so frustrating that we at the forum don't have some universal solution, if we did we could make all our own gear using the same, more efficient algorithm. I want a finite 0-1 db algorithm because I know it is not impractical to convert that to either db or linear without the need more masks or anything else.

It would change the game for synths too. Anyways, I'll think about the lows and the highs and come up with something to show for this idea.

[MichaelBenjamin]

.

[wlangfor@uoguelph.ca]

for me gain is nothing you can generalise, it depends on sound input, so every mix needs to be done from scratch

Yeah, and I have felt the same way too; but what i am saying is -220 dbfs to something like 40 dbfs set to a 0-1 scale. it comes down to what the most acceptable numbers that we can agree upon. It would mean that the math to convert would only need to be done once.

rather than rely on various absolute masks, conversion and what have you; especially in the case of a multiband compressor, it would be a game changer. Or like in the instance of lfo; there would be no recalculation, everything would be finite and ultimately faster to translate. I mean, what takes more memory to calculate, 0-1 to a value, or a string of them?

It just makes sense and if people would tell me what they agreed with it'd be a go and we could collectively implement the idea.

[MichaelBenjamin]

.

[wlangfor@uoguelph.ca]

ok, so the ultimate rule is how human hearing works?
it is a complicated topic sure.

here is a frequency dependent hz scale for humans:
https://opentextbc.ca/physicstestbook2/ ... _06_02.jpg

ah perfect thankyou. That makes it a simple matter. I'll get right on that and create some schematics and data.