< Back to IRCAM Forum

Supervp for max/ supervp.scrub~ vs. supervp.play~ timestretch quality

Forum Max Apps
#1

dear axel, norbert, manuel and the whole supervp gemeinde.

i spend some time now to work with the max supervp objects. there is one thing i became aware of while using supervp.play~ and supervp.scrub~ for timestretching:

following the forumnet maxapps tutorials, these objects are very similar, with the difference, that the latter one its playposition is controlled by a signal. however, for me the result of both objects sounds very different – especially when playing a sound/scrubbing through a sound slowly. i compared the objects using the same sound file and corresponding audio/playback settings. supervp.play~ plays nice and smooth while supervp.scrub~ creates artefacts. i tried to change the max audiosettings as well as the objects arguments (like windowsize and oversampling) but to no avail.

is there something that i miss or misunderstood here? something like: supervp.scrub~ was not meant for very slow playback speed… or even to sound like the scrubbing in audiosculpt.

or is there a special argument-setting needed to get the optimal supervp.scrub~ performance?

 

for demonstration i append a mp3 files with the supervp.scrub~ results.

without touching any of the arguments of the supervp.scrub~ helpfile i was scrubing slowly by hand through the cherokee.aif.

btw, i got the latest version of supervp vor max - objects, using max 6.0.8 on osx 10.7.5.

1000 dank,

johannes

#2

i now append soundfiles of a direct a/b comparison of both objects(supervp.play~ & supervp.scrub~) using the

same stretch factor. this time i used the corresponding “basic examples” patches from the forum max apps.

in the play example patch i set the stretchfactor to 10.

in the scrub example patch i set the stretch-frequency of 0.0351, which is the result of 1000/ (samplelength(ms) * 10)

and i choose a accending sawtooth as waveform for the lfo.

didn´t touch any of the other parameters.

in the recordings that i did, the difference in soundquality is more than obvious.

is that behaviour/soundqualtity of the supervp objects reproducable for you (by using the patches with the same settings)?

 

btw. i like the forumnet max apps very much. they are very good introductions as well as brilliant sounding examples.

especially for the supervp objects it offers a lot of informations, that are missing in the helpfiles. good job!

thanks.

johannes

cherookee_supervp_final.zip (392 KB)

#3

i append the supervp.play~ and supervp.scrub~ results as 2 mp3 files.

 

thanks to the support for solving the upload issue.

ciao, johannes

cherookee_supervp_final.zip (392 KB)

#4

Hi Johannes,

I am not very apt with Max/MSP, but still I can answer a number of questions (note that I use Max 5).

  1. the scrub module uses exactly the same technology that is used in AudioSculpt

  2. playing the cherokee sound through the supervp.scrub help patch works perfectly for me

I suggest you try to use the scrub help patch and see whether you can get it going there.

Best

Axel

#5

hello axel, thanks for your responds.

2) playing the cherokee sound through the supervp.scrub help patch works perfectly for me

did you change the pvoc configurations or the i/o vector size in the help patch?

i already tried the scrub help patch. using this help patch (in max 5 and max 6) with the default settings results in the same artefacts.

for testing (and comparison) i change one of the message boxes in the upper left part of the patch to

(0, 10000 1000000) so scrub´s playback-rate is very low.

somehow reducing the hopsize from 4 to 3 improves the sound qualtity (reduces the artefacts a bit) for me.

hmm, i thought i understood the pvoc parameters but it seems that i miss something important.

 

btw. did you compared my uploaded scrubing results to the results you get with scrub? is there a difference?

 

thanks for help.

johannes

#6

Hi Johannes,

did you change the pvoc configurations or the i/o vector size in the help patch?

no I did not change anything in the scrub or play help patches.

I tried now the (0, 10000 1000000) tweak that you mentioned with the cherokee sound and a play speed of 0.01 in the play help patch setting shapeinvriant mode to on in both.

I selected the same io vectorsize for play and help patches (1024). This obviusly does not sound any more like natural speech, but seems rather similar to me for the play and scrub results. I have Max OS X 10.6.8 though. maybe I should try Mac OS X 10.7? I’ll see whether I find that.

best

Axel

 

#7

hey axel,

you are right, setting shapeinv on makes supervp.play~ and supervp.scrub~ sound similar, especially when working with speech samples.

but as soon as working with non-speech samples (where shapeinv couldn´t be used) the difference is more obvious. for me scrub sound more uncontinuously – with more abrupt changes in the amplitudes and frequencies (of the bins) – than play. i also get frequency-modulation-like-artefacts when using a oversampling factor bigger than 4 and a windowsize of 2250 samples while shapeinv is off.

regarding 10.6.8:

i made the object comparison on mac osx 10.6.8 as well and it makes no difference to me … still get the same results.

sorry for the whole error hocus pocus …

i just try to understand how to get the best results out of the supervp kernel in max…

thanks for taking the time.

greetings,

johannes

#8

Hello Johannes,

it seems you are right, there is a difference that appears especially for very low playing speed and that is due to the difference in the interface between scrub and play. We will investigate whether we can improve, and let you know what we find.

Best

Axel

 

 

#9

Hi Johannes,

Thank you very much for your report.

We have just fixed this behavior and the next version of supervp.scrub~ should perform straight playing with any speed as good as as supervp.play~.

We will release a beta version of SuperVP within the next days.

Cheers

Norbert

 

#10

hi axel and norbert,

nice that you guys could track and solve the problem.

iam looking forward to hear the supervp.scrub object contained in the next beta release…

btw. supervp.max objects: is there a specific reason/useage for the the “get parameters” messages, that one can send to the objects?

 

thanks.

ciao, Johannes

 

#11
btw. supervp.max objects: is there a specific reason/usage for the the “get parameters » messages, that one can send to the objects?

Well, honestly I don’t use them a lot…

Let’s say there are there for completeness.

N.