Hi, quick OT question for any of the granular synthesis brains out there... I want to reconstruct a signal from multiple overlapped segments. To use an overlapp-add method that perfectly reconstructs the original waveform I would need to window the segments... now thats fine as long as the overlap amount is less than 50 percent. I can use a Hanning window on the overlapping parts and sum them..... Hanning windows satisfy the sum to unity criterion. however, when the overlap is more than 50%, the half-Hanning windows applied to the start and end of the segments windows begin to overlap.... see this link: http://www.loopit.org/images/overlap.jpg
Now multiplying the overlapped half-Hanning windows together creates a new window function for the segments that suddenly does not satisfy the 'sum-to-unity' criterion.... The question is (if anybody can is at all interested):- WHAT window function should be used given ANY arbitrary overlap amount???
Any answers would really help me, though I'd be surprised if this gets a response.... ;) sorry 2 b so off topic, though for me it's pd related anyway(!). cheers
matt
-=-=-=-=-=-=-=-=-=-=-=-=- http://www.loopit.org/ -=-=-=-=-=-=-=-=-=-=-=-=-
assuming that you are treating the sum-to-unity criteria as sacred, my first smart-ass suggestion would be:
an envelope that changes size to ensure the overlap is always 50% (or some known value, depending on the actual shape of the envelope)
... ;)
i'm not sure if there is a non-smart-ass answer.
pix.
On Thu, 3 Jul 2003 17:24:31 +0100 "matthew jones" M.Jones@signal.qinetiq.com wrote:
Hi, quick OT question for any of the granular synthesis brains out there... I want to reconstruct a signal from multiple overlapped segments. To use an overlapp-add method that perfectly reconstructs the original waveform I would need to window the segments... now thats fine as long as the overlap amount is less than 50 percent. I can use a Hanning window on the overlapping parts and sum them..... Hanning windows satisfy the sum to unity criterion. however, when the overlap is more than 50%, the half-Hanning windows applied to the start and end of the segments windows begin to overlap.... see this link: http://www.loopit.org/images/overlap.jpg
Now multiplying the overlapped half-Hanning windows together creates a new window function for the segments that suddenly does not satisfy the 'sum-to-unity' criterion.... The question is (if anybody can is at all interested):- WHAT window function should be used given ANY arbitrary overlap amount???
Any answers would really help me, though I'd be surprised if this gets a response.... ;) sorry 2 b so off topic, though for me it's pd related anyway(!). cheers
matt
-=-=-=-=-=-=-=-=-=-=-=-=- http://www.loopit.org/ -=-=-=-=-=-=-=-=-=-=-=-=-
PD-list mailing list PD-list@iem.at http://iem.at/cgi-bin/mailman/listinfo/pd-list
i've done a TINY bit of granular synthesis in csound, and i don't remember doing any FFTs or windowing. i don't remember very well how it all worked though. i'm pretty familiar with overlap'n'add because i've had to use that for convolution externals. i do it in 'externals' though and not in pd. if i saw a patch that does all the shit that i do in my convolution external it would make my head spin.
On Thu, 3 Jul 2003, matthew jones wrote:
Hi, quick OT question for any of the granular synthesis brains out there... I want to reconstruct a signal from multiple overlapped segments. To use an overlapp-add method that perfectly reconstructs the original waveform I would need to window the segments... now thats fine as long as the overlap amount is less than 50 percent. I can use a Hanning window on the overlapping parts and sum them..... Hanning windows satisfy the sum to unity criterion. however, when the overlap is more than 50%, the half-Hanning windows applied to the start and end of the segments windows begin to overlap.... see this link: http://www.loopit.org/images/overlap.jpg
Now multiplying the overlapped half-Hanning windows together creates a new window function for the segments that suddenly does not satisfy the 'sum-to-unity' criterion.... The question is (if anybody can is at all interested):- WHAT window function should be used given ANY arbitrary overlap amount???
Any answers would really help me, though I'd be surprised if this gets a response.... ;) sorry 2 b so off topic, though for me it's pd related anyway(!). cheers
matt
-=-=-=-=-=-=-=-=-=-=-=-=- http://www.loopit.org/ -=-=-=-=-=-=-=-=-=-=-=-=-
PD-list mailing list PD-list@iem.at http://iem.at/cgi-bin/mailman/listinfo/pd-list
"640K ought to be enough for anybody." -- Bill Gates, 1981
-
J. Scott Hildebrand -
matthew jones -
pix