Tyler Nitsch

Ok I know for audio when you do an fft on a signal you specify the size of the fft(frequency resolution). Then you get a vector of complex numbers that represent the magnitude and phase of frequencies that correspond to fs/2*n*(freqres) where n is the index of the vector.... (in the looses sense here). So my question is what exactly is the format that jit.fft spits out. Since it operates on two plane matrices I'm guessing that this will determine what data is spit out. Can someone help me understand this object?

Ok I know for audio when you do an fft on a signal you specify the size of the fft(frequency resolution).  Then you get a vector of complex numbers that represent the magnitude and phase of frequencies that correspond to fs/2*n*(freqres) where n is the index of the vector.... (in the looses sense here). So my question is what exactly is the format that jit.fft spits out. Since it operates on two plane matrices I'm guessing that this will determine what data is spit out. Can someone help me understand this object?


jit-fft

On Aug 8, 2006, at 5:04 PM, Tyler Nitsch wrote:

> Ok I know for audio when you do an fft on a signal you specify the 

> size of the fft(frequency resolution). Then you get a vector of 

> complex numbers that represent the magnitude and phase of frequencies 

> that correspond to fs/2*n*(freqres) where n is the index of the 

> vector.... (in the looses sense here). So my question is what exactly 

> is the format that jit.fft spits out. Since it operates on two plane 

> matrices I'm guessing that this will determine what data is spit out. 

> Can someone help me understand this object?

I'd suggest you check out the jit.fft-threshold.pat example and other 

jit.fft examples and let us know if you have additional questions. Keep 

in mind that in an fft, you have both positive and negative frequencies 

(about the center vertically and horizontally in the 2D case), and that 

in a 2D fft the harmonics have frequencies about each axis as 

determined by the cell position. real and imaginary coordinates are in 

the matrix planes 0 and 1 respectively. More info on 2D ffts can be 

>
> Ok I know for audio when you do an fft on a signal you specify the 
> size of the fft(frequency resolution).  Then you get a vector of 
> complex numbers that represent the magnitude and phase of frequencies 
> that correspond to fs/2*n*(freqres) where n is the index of the 
> vector.... (in the looses sense here). So my question is what exactly 
> is the format that jit.fft spits out. Since it operates on two plane 
> matrices I'm guessing that this will determine what data is spit out. 
> Can someone help me understand this object?

I'd suggest you check out the jit.fft-threshold.pat example and other 
jit.fft examples and let us know if you have additional questions. Keep 
in mind that in an fft, you have both positive and negative frequencies 
(about the center vertically and horizontally in the 2D case), and that 
in a 2D fft the harmonics have frequencies about each axis as 
determined by the cell position. real and imaginary coordinates are in 
the matrix planes 0 and 1 respectively. More info on 2D ffts can be 
found online.

Thanks for the tips. Ok for future browsers and searchers of this topic after doing a little bit of digging and weeding I finally found an article that explains the relation between locations and values of the 2d fft and the actuall spatial image.

http://homepages.inf.ed.ac.uk/rbf/HIPR2/fourier.htm

Thanks for the tips.  Ok for future browsers and searchers of this topic after doing a little bit of digging and weeding I finally found an article that explains the relation between locations and values of the 2d fft and the actuall spatial image.

http://homepages.inf.ed.ac.uk/rbf/HIPR2/fourier.htm


I hope this post finds well and safe 15 years later.

I wish to learn how to properly use JIT.FFT to analyse/transform audio signals in a matrix, with JIT.BUFFER, which is basically a 1d matrix.

In the audio world we usually provide parameters like window size, window envelope, hop size, but that's not the case with JIT.FFT.

So, I wish to understand the implications when using JIT.FFT.

I looked at the jit.fft-threshold example and I read the article provided by Tyler (fortunately the url still works), but I'm still missing that eureka moment.

One sentence in the article called my attention: "The number of frequencies corresponds to the number of pixels in the spatial domain image...".

When a audio signal is stored in a 1d matrix, one pixel represents one sample, how does this translate in the audio domain? 

In the audio domain FFT the number of frequency bins depend on the sampling rate and the window size, so following the article logic, 20000 samples correspond to 20000 frequency bins with JIT.FFT? Sounds odd...

Translating into the audio domain logic, if I perform a FFT with sample rate 44100 and window size 44100 then, FFT f0 will be 1Hz (44100/44100).

https://cycling74.com/forums/spectrum-of-a-bufferaudio-file

the audio signal is sliced frame by frame (e.g. 1024 samples/pixels/cells), every hop size (e.g. 512 samples), plus they add a envelope window on top of each signal frame analysed.

Hello All,
I hope this post finds well and safe 15 years later.
I have similar questions to Tyler.
I wish to learn how to properly use JIT.FFT to analyse/transform audio signals in a matrix, with JIT.BUFFER, which is basically a 1d matrix.

In the audio world we usually provide parameters like window size, window envelope, hop size, but that's not the case with JIT.FFT.
So, I wish to understand the implications when using JIT.FFT.

I looked at the jit.fft-threshold example and I read the article provided by Tyler (fortunately the url still works), but I'm still missing that eureka moment.
One sentence in the article called my attention: "The number of frequencies corresponds to the number of pixels in the spatial domain image...".

When a audio signal is stored in a 1d matrix, one pixel represents one sample, how does this translate in the audio domain? 
In the audio domain FFT the number of frequency bins depend on the sampling rate and the window size, so following the article logic, 20000 samples correspond to 20000 frequency bins with JIT.FFT? Sounds odd...
What is the sampling rate in this case?
Translating into the audio domain logic, if I perform a FFT with sample rate 44100 and window size 44100 then, FFT f0 will be 1Hz (44100/44100).

In the examples provided in the post 
https://cycling74.com/forums/spectrum-of-a-bufferaudio-file
the audio signal is sliced frame by frame (e.g. 1024 samples/pixels/cells), every hop size (e.g. 512 samples), plus they add a envelope window on top of each signal frame analysed.
I suppose this is the right way?