# Delaying the analysis will not work, because analysis data is an one block audio signal. If you delay it not a integer multiple of a block, the synchronisation of different delayed bin analysis, get lost. # I made two different versions of spectral delay, one using delay before inverse FFT, other after it. Both doesn't work :-) but the second one gives good results for some certain delay times. I don't understand why this result happened. -ugur-
On 6/4/07, ugur guney ugurguney@gmail.com wrote:
http://crca.ucsd.edu/%7Emsp/techniques/latest/book-html/node179.html# I think something like this could work: If the number of bins is 2*N, prepare N tables with size N samples. 1st table has one in 1st sample and zero everywhere. Similarly n'th table has 1 in n'th sample and zero in other samples. This tables will be used to pick up n'th bin from the mixed analysis. # Then prepare 2*N [*~] objects and multiply real and imag output of [rfft~] with [tabreceive~ preparedTableN]. This way every bin will be seperated, every pair has the amplitude and phase information of only one bin. # Then apply delay's to every pair and sum all delayed real and imaginary parts seperately. Finally put these two signals to [rifft~] -ugur guney-