Hello everybody, this is my first post.

I'm trying to build a vanilla convolver using the frequency delay line technique which is described in these two papers:

http://pcfarina.eng.unipr.it/Public/AES-113/Garcia-PrePrint5660.pdf

http://cnmat.berkeley.edu/system/files/attachments/main.pdf

I'm currently running 8 partitions, all of with a blocksize of 4096 and an overlap factor of 4 and I've wired it together as it's described in the diagrams and in the paper. I am however not getting satisfactory results. The signal is resynthesised by the rifft~ but it's hardly what I'd call convolution.

I can think of two errors that I might have done. The first is that I neither zero-padded the input signal (how do you even zero-pad a live audio stream?) nor the impulse response. The other thing I might have got wrong is that somehow I've chosen a somewhat erroneous way of adding together the output again.

Being on a relatively modest level of DSP knowledge myself I hope someone here can shed some light on this.

You can find my patch attached to this post.

Thanks in advance.

Edit:

You will find the partitions inside the subpatch pd FDL_Convolver. There are comments in there too which explains what each part does.

http://www.pdpatchrepo.info/hurleur/DSP_convolution_reverb_project_FDL_Avoiding_spaghetti.pd