chthonic

My post is regarding the article posted by dudas, discussing a real-time phase vocoder using complex math and the Cartesian Coordinate System. I have been comparing it with the phase vocoder presented in MSP Tutorial 26 which converts the Cartesian coordinates to polar ones, thus eliminating a lot of complicated math from the patch. I have two questions: why does the patch using complex math with Cartesian Coordinates sound more metallic (worse) than the one presented in tutorial 26? Also what imposes the limit of the fft bin size in the patch using complex math? If you look at the patches (provided in the tutorials section of the website), it limits you to using a bin size of maximum 4096 with an overlap of 4. How would one raise this limit?

Any answers would be much appreciated. I hope this is clear enough.

comparing-of-two-phase-vocoders

chthonic wrote on Wed, 13 February 2008 01:16

My post is regarding the article posted by dudas, discussing a real-time phase vocoder using complex math and the Cartesian Coordinate System. I have been comparing it with the phase vocoder presented in MSP Tutorial 26 which converts the Cartesian coordinates to polar ones, thus eliminating a lot of complicated math from the patch. I have two questions: why does the patch using complex math with Cartesian Coordinates sound more metallic (worse) than the one presented in tutorial 26?

Sometime between your initial post and now I thought I had experienced a similar problem with the dudas' cartesian example (article link: 

https://cycling74.com/story/2006/11/2/113327/823

). After many late night attempts to "fix" complex math that as far as I could tell should have been correct, I gave up on trying to figure out why the cartesian example sounded much less accurate than the polar version. I think I was getting a similar metallic sound as you were experiencing (sounded like the wrong phase information was getting through). I was about to post here to let you know I was experiencing something similar when I decided I would try it out one more time to doublecheck the problem. I'm on another computer at the moment so I had to redownload the patches and this time as far as I could tell, the phase voders sounded the same when using either polar or cartesian coordinates. I'm guessing that some how with the way I was cycling DSP or maybe other open patches I had screwed with with [send~]/[receive~] pairs. Anyway, I'd recommend you download them again, freshly open MaxMSP, and try again and see if you are still having problems.

Also what imposes the limit of the fft bin size in the patch using complex math? If you look at the patches (provided in the tutorials section of the website), it limits you to using a bin size of maximum 4096 with an overlap of 4. How would one raise this limit?

EDIT: As far as I can tell, when using [fft~] to do your FFT, you are limited to a maximum FFT size of 4096. So that is what imposes the limit, I'm not sure what (if anything) can be done to raise the limit.

I would be curious to know if you are still experiencing the metallic sound in the complex math version of the patch.

chthonic wrote on Wed, 13 February 2008 01:16My post is regarding the article posted by dudas, discussing a real-time phase vocoder using complex math and the Cartesian Coordinate System. I have been comparing it with the phase vocoder presented in MSP Tutorial 26 which converts the Cartesian coordinates to polar ones, thus eliminating a lot of complicated math from the patch. I have two questions: why does the patch using complex math with Cartesian Coordinates sound more metallic (worse) than the one presented in tutorial 26?

Sometime between your initial post and now I thought I had experienced a similar problem with the dudas' cartesian example (article link: https://cycling74.com/story/2006/11/2/113327/823).  After many late night attempts to "fix" complex math that as far as I could tell should have been correct, I gave up on trying to figure out why the cartesian example sounded much less accurate than the polar version.  I think I was getting a similar metallic sound as you were experiencing (sounded like the wrong phase information was getting through).  I was about to post here to let you know I was experiencing something similar when I decided I would try it out one more time to doublecheck the problem.  I'm on another computer at the moment so I had to redownload the patches and this time as far as I could tell, the phase voders sounded the same when using either polar or cartesian coordinates.  I'm guessing that some how with the way I was cycling DSP or maybe other open patches I had screwed with with [send~]/[receive~] pairs.  Anyway, I'd recommend you download them again, freshly open MaxMSP, and try again and see if you are still having problems.

chthonic wrote on Wed, 13 February 2008 01:16Also what imposes the limit of the fft bin size in the patch using complex math? If you look at the patches (provided in the tutorials section of the website), it limits you to using a bin size of maximum 4096 with an overlap of 4. How would one raise this limit?

EDIT: As far as I can tell, when using [fft~] to do your FFT, you are limited to a maximum FFT size of 4096.  So that is what imposes the limit, I'm not sure what (if anything) can be done to raise the limit.

I would be curious to know if you are still experiencing the metallic sound in the complex math version of the patch.


Comparing of two phase vocoders