recently i have discovered the joys of using overlapped blocks in sound patches. however, i'm having a little bit of difficulty understanding exactly WHAT gets overlapped, and what doesn't get overlapped.
it seems to me that control messages and do not get overlapped, and i can't really tell what happens to external audio signals connected to the overlapped patch.
is there some documentation somewhere on this topic? or would a kind person like to tell me basically what functions block overlap has?
cheers. matt
# If the block size is N samples (and there is no overlapping) then the patches (subpatches) calculate the audio stream after every Nth sample. At every tick the inlet~s take N samples in and the outlet~s give N samples out. # Overlapping is like you have M identical audio streams which are N/M samples out of phase. For N=1024 and M=4 the "frame size" is 1024 and the "hop size" is 256. You have four frames. First one is between samples n0 and n0+1024. Second one [n0+256, n0+256+1024], third one [n0+512,n0+512+1024] etc.
1>x1234xxxx 2>xx1234xxx 3>xxx1234xx 4>xxxx1234x o>xxxxOxxxx where O = 4>1 + 3>2 + 2>3 + 1>4
# Every frame has four parts consists of 256 samples. The output is the sum of these parts. n> denotes the streams and numbers are parts of the frames. # If you do nothing inside a patch (just connect the inlet~ to outlet~) 4th part of the first frame is the same as the third part of the second frame and they are equal to second part of the third frame and so on. (4>1 = 3>2 = 2>3...) Because they correspond to the same parts of the incoming input. (I forget whether there is an automatic normalization in PD after summing M frames) # So why we are doing this? Why do we need overlapping? An example where it is useful is DFT (FFT). You get a better resolution in your spectrum if your frame size is big. But with a long frame your time resolution decreases (when you take the FFT at every frame.) You get spectral data at every N'th sample. This is due to the uncertainty relation. # If you want to make a spectral analysis (take FFT), manipulate the spectrum (play with the table which stores the spectrum), and resynthesis the input (calculate the iFFT) then it is better if you use overlapping blocks. Sorry for my bad usage of English! # You can look at a working example/application in the audio examples folder: I03.resynthesis.pd and read the related chapter in Puckette's book. and afaik overlapping is only related to audio signals not to the control signals. -ugur-
On Wed, Jul 2, 2008 at 3:56 PM, hard off hard.off@gmail.com wrote:
recently i have discovered the joys of using overlapped blocks in sound patches. however, i'm having a little bit of difficulty understanding exactly WHAT gets overlapped, and what doesn't get overlapped.
it seems to me that control messages and do not get overlapped, and i
can't
really tell what happens to external audio signals connected to the overlapped patch.
is there some documentation somewhere on this topic? or would a kind
person
like to tell me basically what functions block overlap has?
cheers. matt
Pd-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
-
hard off -
Uğur Güney