Page 2 of 2

Re: Splitting FFT spectrum into octave bands? how? worth it?

PostPosted: Thu Aug 15, 2019 10:34 pm
by guyman
Maybe I'm looking for a way to take samples of packs/ goups of monos. like let's say there are more operations I want execute on these bands before floating them. A lot of efficiency could be had by floating multiple mono channels at once. Like the signal analyzer prim allows me to build a float array out of a poly. Do I need some sort of mono/poly/mem thingy going on that basically allows me to sample x # of mono/poly channels with each sample. Like a user defined set of bins of float data.. is this all redundant?

Re: Splitting FFT spectrum into octave bands? how? worth it?

PostPosted: Sat Aug 17, 2019 11:49 pm
by martinvicanek
Here is a semitone analyzer. Not based on FFT but on Goertzel algo, which is closely related.

Re: Splitting FFT spectrum into octave bands? how? worth it?

PostPosted: Sat Aug 24, 2019 6:18 pm
by wlangfor@uoguelph.ca
Goertzel? Hmm, that's interesting.

BTW, I had done some work limiting the FFT to try and save CPU with some success furthermore. Try this example, use whatever for whatever you wish.

http://www.dsprobotics.com/support/viewtopic.php?f=2&t=17874

Re: Splitting FFT spectrum into octave bands? how? worth it?

PostPosted: Sun Aug 25, 2019 1:47 am
by martinvicanek
I tried your 34 band analyzer with a pure sine input. The readings were just random, the schematic is totaly useless. First thing is you don't use a window so your spectrum will spill all over the palce. But that's not the biggest problem. The frequency-bin mappig seems to be messed up completely. How can you seriously post such stuff and then praise it
as you usually do?

Re: Splitting FFT spectrum into octave bands? how? worth it?

PostPosted: Sun Aug 25, 2019 2:08 am
by wlangfor@uoguelph.ca
martinvicanek wrote:I tried your 34 band analyzer with a pure sine input. The readings were just random, the schematic is totaly useless. First thing is you don't use a window so your spectrum will spill all over the palce. But that's not the biggest problem. The frequency-bin mappig seems to be messed up completely. How can you seriously post such stuff and then praise it
as you usually do?


Hmm, well initially it didn't have those advancements but I'd added them. The initial version was just a prototype and it wasn't on the examples section. It seemed to work, I could try it with sine.

Sine will be with only one frequency however, so, unless a graphic EQ is very inaccurate and smooths over values it wouldn't show up anyways. I believe the command response is there so that it spreads the effect. In that sense, you're right - If you wanted such an in-accurate analyzer. Only one frequency and you wish 4 or 2 or even 1 bars to move? That doesn't make any sense whatsoever.

But, all the same; the fact that there is random responses is surprising. I'll check that out I guess. Goertzel sounds positive, of course I'd love to use a windowed function, but the reality is that 1% CPU usage is too much. Hopefully it can get down to 0.2 to 0.3%.

For instance, My latest project called the technician has 3 peaking eq bands, 2 shelves, 1 hp and 1 lp as well as 4 bands and the implementation of a four band stereo width controller and of course that required dynamic crossover. And with all of that said (including with a "windowless" audio analyzer), it's only using 0.5% in reaper, and 1% in fl studio 10,11 and 2% in 12,20. It's merely that we can't afford the cost of windows MV. Anyways, to see that project go here:
http://dsprobotics.com/support/viewtopic.php?f=2&t=23646

I really hope that the goertz is the ticket. Maybe the goertz will be with Me if that's true.