Cheng Chien-Wen

I am studying FFT right now and find that vectral~ sometimes appear in these FFT related patches. I read the manual and help file but could not understand when and why we need to use it.

when-and-why-to-use-vectral-2

if I dont remember bad (but too less sleep hours i've got last night 

vectral~ lets you do some operation between succesive buffers and 

create very useful ramps for producing reverbs and delays.

Technically vectral can act like a smoothing filter that doesn't work 

but bin by bin: that means that it works with samples, but it doesnt 

act through consecutive samples, but escaping samples.

Because of this feature you can actually do better things easily in fft.

i remember once, in Basel (ch) , a great teacher from Cycling, mr 

R.D., tell about vectral... " it was there? was it always there???"

Il giorno 24/ago/06, alle ore 07:33, Cheng Chien-Wen ha scritto:

> I am studying FFT right now and find that vectral~ sometimes appear 

> in these FFT related patches. I read the manual and help file but 

> could not understand when and why we need to use it.

if I dont remember bad (but too less sleep hours i've got last night  
to tell the truth :-p)
vectral~  lets you do some operation between succesive buffers and  
create very useful ramps for producing reverbs and delays.
Technically vectral can act like a smoothing filter that doesn't work  
sample by sample
but bin by bin: that means that it works with samples, but it doesnt  
act through consecutive samples, but escaping samples.
Because of this feature you can actually do better things easily in fft.

i remember once, in Basel (ch) , a great teacher from Cycling, mr  
R.D., tell about vectral... " it was there? was it always there???"

hope this helps,
mabe some other can be more specific

>
> Hi,
>
> I am studying FFT right now and find that vectral~ sometimes appear  
> in these FFT related patches. I read the manual and help file but  
> could not understand when and why we need to use it.
>
> Could anybody explain more clearly ?
>
> Thanks.
>
> Chien-Wen
>
>
>


It's just a bin based envelope follower, so it's useful for anything 

which typical envelope followers are good for, perhaps most common of 

which being dynamics processing like noise gating, compression, 

expansion or visualization purposes like an eq meter. However, 

there's a number of more interesting things which can be done like 

tying them to oscillator bank frequency or vocoder filter maps, 

etc...I think my initial reason for making it was to assist a patch 

that performed FFT based noise reduction/spectral thinning in the 

It's just a bin based envelope follower, so it's useful for anything  
which typical envelope followers are good for, perhaps most common of  
which being dynamics processing like noise gating, compression,  
expansion or visualization purposes like an eq meter. However,  
there's a number of more interesting things which can be done like  
tying them to oscillator bank frequency or vocoder filter maps,  
etc...I think my initial reason for making it was to assist a patch  
that performed FFT based noise reduction/spectral thinning in the  
vein of the old favorite ionizer.

Are there any example patches or tutorials that can examplify the use of vectral~ ?

It seems that the output of vectral~ is just like some amplitude info, doesn't it ?

I observed that the data from vectral~ output is frequently connected to the fft~ real and imagery ouput as if it is used to scale the amplitude. But I am still confused about how vectral~ works. Since FFT deals with frequency domain, I feel confused about this bin-based envelop follower which seems to be something dealing with time-domain stuff. What exactly is the output of vectral~ ? The overall amplitude data for each consecutive FFT frame in time domain, the amplitude data for each frequency bin in a FFT frame, or something else ?

Normally if you used something like rampsmooth in pfft, rampsmooth would interpolate between the values of successive bin indexes. 

i.e. the interpolation all happens within one frame.

With vectral, the interpolation occurs ACROSS separate frames. So for example if you were running a pfft with a FFT size of 8, you get 8 samples per frame. Each of these samples represents one eigth of the frequency spectrum, and are indexed from 0 to 7.

You dont want to be interpolating from index 1 to index 2 to index 3 etc. 

You want to interpolate between index 1 of the first frame and index 1 of the second frame.

When vectral is wired to receive the index values from an fftin object, vectral knows which frequency band is which. This way, when it receives a value from index 1, it holds that value and waits until the next index 1 comes around and then performs its calculations.

I hope this helps, or at least didn't hurt. :]

Normally if you used something like rampsmooth in pfft, rampsmooth would interpolate between the values of successive bin indexes. 
i.e. the interpolation all happens within one frame. 

 With vectral, the interpolation occurs ACROSS separate frames. So for example if you were running a pfft with a FFT size of 8, you get 8 samples per frame. Each of these samples represents one eigth of the frequency spectrum, and are indexed from 0 to 7. 

You dont want to be interpolating from index 1 to index 2 to index 3 etc. 
You want to interpolate between index 1 of the first frame and index 1 of the second frame. 

When vectral is wired to receive the index values from an fftin object, vectral knows which frequency band is which. This way, when it receives a value from index 1, it holds that value and waits until the next index 1 comes around and then performs its calculations. 

I hope this helps, or at least didn't hurt. :]


Thank you very much. I understand it a little better now.

Tht patch posted above is also very interesting.

But the patch posted above uses vectral~ differently from what I saw in the following patch. The following patch is part of someone's piece which I am studying. When I take part of it out of the original patch to test it, it does not output any sound from pfft subpatch.

I am thinking that the programmer uses "bin shift" (mentioned in MSP reference) to transform the sound. (I have no idea what bin shift is, but I know it does transform the sound.) But I don't know if the lookup table connected to the rightmost output works or not. Perhaps the lookup table is used to scale the FFT output. However, the patch does not output any sound from FFT. Any idea ?

Thank you very much. I understand it a little better now.
Tht patch posted above is also very interesting.

Could any one see what's wrong with it ?
I am thinking that the programmer uses "bin shift" (mentioned in MSP reference) to transform the sound. (I have no idea what bin shift is, but I know it does transform the sound.) But I don't know if the lookup table connected to the rightmost output works or not. Perhaps the lookup table is used to scale the FFT output. However, the patch does not output any sound from FFT. Any idea ?

Il giorno 30/ago/06, alle ore 02:48, Cheng Chien-Wen ha scritto:

> I observed that the data from vectral~ output is frequently 

> connected to the fft~ real and imagery ouput as if it is used to 

> scale the amplitude. But I am still confused about how vectral~ 

> works. Since FFT deals with frequency domain, I feel confused about 

> this bin-based envelop follower which seems to be something dealing 

> with time-domain stuff. What exactly is the output of vectral~ ? 

> The overall amplitude data for each consecutive FFT frame in time 

> domain, the amplitude data for each frequency bin in a FFT frame, 

>
> I observed that the data from vectral~ output is frequently  
> connected to the fft~ real and imagery ouput as if it is used to  
> scale the amplitude. But I am still confused about how vectral~  
> works. Since FFT deals with frequency domain, I feel confused about  
> this bin-based envelop follower which seems to be something dealing  
> with time-domain stuff. What exactly is the output of vectral~ ?  
> The overall amplitude data for each consecutive FFT frame in time  
> domain, the amplitude data for each frequency bin in a FFT frame,  
> or something else ?
>
> Could someone help explain more ?
>
> Thanks.
>
>
>
>


On Aug 30, 2006, at 1:27 AM, tommaso perego wrote:

vectral~ is just a temporal data smoother for framebased data that is 

serialized as an audio signal. What that data is and how or why to 

Perhaps the confusion is over how a frame of information is 

serialized as an audio signal, or what precisely a bin within a frame 

is, but I'll leave it to someone else to take the time to walk 

through those points if there's still confusion.

> it should be amplitude for each bin
> maybe joshua can tell us :-)

vectral~ is just a temporal data smoother for framebased data that is  
serialized as an audio signal. What that data is and how or why to  
use it is completely general.

Perhaps the confusion is over how a frame of information is  
serialized as an audio signal, or what precisely a bin within a frame  
is, but I'll leave it to someone else to take the time to walk  
through those points if there's still confusion.

> I observed that the data from vectral~ output is frequently connected

> to the fft~ real and imagery ouput as if it is used to scale the

> amplitude. But I am still confused about how vectral~ works. Since

> FFT deals with frequency domain, I feel confused about this bin-based

> envelop follower which seems to be something dealing with time-domain

In mathematical land the frequency domain seem disconnected from the 

time domain but then you'd need the whole duration of your piece as 

In the real fft world there remains a time aspect, you slice the signal 

into short slices (the fft window size) within one slize you have no 

time structure (but you can get that information back with ifft). 

Outside you do have a time structure, the change from slice to slice. 

But each slice is divided and ordered by frequency bins.

If you want to change the timestructure of a single bin (like for 

example have an envelope follower) then you need tools like vectral,

If you would split each bin into a seperate signal processing chain, you 

could downsample by the window size and process each bin as if it would 

be a normal signal. But you would have to deal with two parts seperately 

the real and imaginary parts or if you converted it, the amplitude and 

phase. Yes, this is getting complicated, but consistent...

Cheng Chien-Wen wrote:
> I observed that the data from vectral~ output is frequently connected
> to the fft~ real and imagery ouput as if it is used to scale the
> amplitude. But I am still confused about how vectral~ works. Since
> FFT deals with frequency domain, I feel confused about this bin-based
> envelop follower which seems to be something dealing with time-domain
> stuff. 

In mathematical land the frequency domain seem disconnected from the 
time domain but then you'd need the whole duration of your piece as 
latency to start processing it...
In the real fft world there remains a time aspect, you slice the signal 
into short slices (the fft window size) within one slize you have no 
time structure (but you can get that information back with ifft). 
Outside you do have a time structure, the change from slice to slice. 
But each slice is divided and ordered by frequency bins.
If you want to change the timestructure of a single bin (like for 
example have an envelope follower) then you need tools like vectral,
frameaccum~, framedelta~.
If you would split each bin into a seperate signal processing chain, you 
could downsample by the window size and process each bin as if it would 
be a normal signal. But you would have to deal with two parts seperately 
the real and imaginary parts or if you converted it, the amplitude and 
phase. Yes, this is getting complicated, but consistent...

-- 
Stefan Tiedje------------x-------
--_____-----------|--------------
--(_|_ ----|-----|-----()-------
-- _|_)----|-----()--------------
----------()--------www.ccmix.com


i still have a question on how to use vectral~:

if i would like to manipulate the way, a sound develops in time, 

across multiple frames, i have apply envelope following on both 

real and imaginary data, or in polar value, magnitude and phase, 

since both carry time information, is this true?

but as soon as i manipulate time, i seem to affect also frequency or phase? 

i must admit i have some difficulties how to treat those 2 parameters seperately.

hi,
i still have a question on how to use vectral~:

if i would like to manipulate the way, a sound develops in time, 
across multiple frames, i have apply envelope following on both 
real and imaginary data, or in polar value, magnitude and phase, 
since both carry time information, is this true?

but as soon as i manipulate time, i seem to affect also frequency or phase? 
i must admit i have some difficulties how to treat those 2 parameters seperately.

> if i would like to manipulate the way, a sound develops in time, 

> across multiple frames, i have apply envelope following on both 

> real and imaginary data, or in polar value, magnitude and phase, 

> since both carry time information, is this true?

Best is to try both (and more) ways and listen...

You might only want to change the amplitude part. Go experiment and 

listen. I'd try vectral~ only on the magnitude and leave the phase 

untouched as a starting point. Try to understand why it sounds like it 

sounds... If this leads to too much artefacts, try to feed the phase 

into framedelta~/phasewrap~ and after eventual processing into 

Sometimes even randomisation of the phase does miracles... ;-)

You will also learn from listening to the difference in results for 

varying frame sizes. Do the extremes (very small/very big...) to find 

out what will be appropriate... (watch out the CPU usage though...)

kassian schrieb:
> if i would like to manipulate the way, a sound develops in time, 
> across multiple frames, i have apply envelope following on both 
> real and imaginary data, or in polar value, magnitude and phase, 
> since both carry time information, is this true?

You might only want to change the amplitude part. Go experiment and 
listen. I'd try vectral~ only on the magnitude and leave the phase 
untouched as a starting point. Try to understand why it sounds like it 
sounds... If this leads to too much artefacts, try to feed the phase 
into framedelta~/phasewrap~ and after eventual processing into 
frameaccum~...

You will also learn from listening to the difference in results for 
varying frame sizes. Do the extremes (very small/very big...) to find 
out what will be appropriate... (watch out the CPU usage though...)

-- 
Stefan Tiedje------------x-------
--_____-----------|--------------
--(_|_ ----|-----|-----()-------
-- _|_)----|-----()--------------
----------()--------www.ccmix.com

when and why to use vectral~