Nicholas Solem

Hi all. I've been trying for a while to do some morphing in the cepstral domain with pfft~. As you may know (probably a prerequisite to help with the issue at hand), there are main 2 kinds of cepstrum: real cepstrum and complex cepstrum. From Udo Zolzer, my first goal was to apply the spectral envelope of one sound to the source of another sound, by low-time liftering the spectral envelope sample and high-time liftering the source sample.

So, with the real cepstrum, I got it to work, but it sounds lossy (softened transients, compressed decay curve, other loss of detail). I thought that by implementing the same principle with a complex cepstrum, the issue would be resolved.

Unfortunately, the complex cepstral morphing fails. I must be doing something very wrong somewhere. Take a look at the attached project, and follow the steps in the order listed to see what I mean (watch your volume before turning on the complex one).

The complex cepstral morph is based on this paper: 

That explains the bit about switching out cartopol~ and poltocar~ with the phaseunwrap_itoh~ and other stuff.

Any help on what I may be doing wrong for the complex cepstral morphing, or another way to get the real cepstrum to sound better, would be greatly appreciated.

Hi all. I've been trying for a while to do some morphing in the cepstral domain with pfft~.  As you may know (probably a prerequisite to help with the issue at hand), there are main 2 kinds of cepstrum: real cepstrum and complex cepstrum.  From Udo Zolzer, my first goal was to apply the spectral envelope of one sound to the source of another sound, by low-time liftering the spectral envelope sample and high-time liftering the source sample. 

So, with the real cepstrum, I got it to work, but it sounds lossy (softened transients, compressed decay curve, other loss of detail). I thought that by implementing the same principle with a complex cepstrum, the issue would be resolved. 

Unfortunately, the complex cepstral morphing fails. I must be doing something very wrong somewhere.  Take a look at the attached project, and follow the steps in the order listed to see what I mean (watch your volume before turning on the complex one).
The complex cepstral morph is based on this paper: http://vud.org/papers/ApelCepstral.pdf
That explains the bit about switching out cartopol~ and poltocar~ with the phaseunwrap_itoh~ and other stuff. 

Any help on what I may be doing wrong for the complex cepstral morphing, or another way to get the real cepstrum to sound better, would be greatly appreciated.


cepstrum-morph-issue

Anyone? Here's a screenshot of the whole project:

Anything I could clear up to help someone help me? Really trying to use this for a bigger project

Alright, I figured out the main problem: within the [phasewrap_itoh~], there was the use of a [frameaccum]. It needed to have its phasewrap flag set to 1.

Alright, I figured out the main problem:  within the [phasewrap_itoh~], there was the use of a [frameaccum]. It needed to have its phasewrap flag set to 1.


i think that apel paper looks very dodgy to me - a whole bunch of stuff not explained / left out.

this seems a very difficult problem in max. even with your [frameaccum 1] i still get noise on the outputs of the complex.

i would say that if {full spectrum-flag} is on (it should be in your case), internal ifft/fft pair need to be same fft size as parent [pfft~] object. but if {full spectrum-flag} is off, internal ifft/fft pair need to be half fft size of parent [pfft~] object. also there should be a relationship between the parent overlap and any manual overlap internally on the ifft/fft.

my instinct tells me those should be windowed with same window as parent, too.

did you get any further? it is very interesting.

Thank you for your response. Yes, the project has gotten further. Maybe I will attach an update soon. I'll experiment with the changes in window size as you mention. However, what you say about windowing doesn't seem right to me, because windowing the spectrum prior to getting its cepstrum would correspond with applying a bandpass filter over all of the signal... Maybe I'll experiment with it anyway though.

Cepstrum Morph Issue