Hi

I'm currently trying to use pd as a sort of sequencer, to arrange and playback audio loops and samples mostly via arrays.

I've tried out a few timestretch options, but none of them seem to maintain decent audio quality for, say, a well-recorded 4-bar drum loop at 120bpm. I don't want to stretch the audio far (say, 10-15 bpm either way), but even routing it through a timestretch abstraction of any kind at regular playback speed seems to deteriorate the quality more than I'd like.

Can anyone recommend a good timestretch patch or abstraction? Someone recommended a patch named [synchgrain~] as being useful, but I've been unable to track it down. I know PD isn't usually used for sequencing recorded audio while trying to maintain audio quality, but it's the best approach I can see for what I'm doing!

You might have already tried it, but the included patch I07.phase.vocoder.pd (filed under pure data/03.audio examples in the help browser) is a fairly decent and relatively artifact-free timestretcher, in my view. Having said that, I can't vouch for how well it passes audio that isn't being stretched.

Thanks Stutter - yes, that I07 patch does seem to be the best solution I've hit on so far! I'm doing my best to reduce it into a simple timestretch abstraction that I can use in other patches, but haven't quite made it due to a lack of understanding of how it works :/ Working in a successful loop is still evading me, but I'm sure I'll get there... more tips or advice are welcome in the meantime?

Unrelated to pd, i know, but since we're on the topic of timestretching:

http://hypermammut.sourceforge.net/paulstretch/

Quite an amazing piece of software.

i've never heard a good proper all-purpose timestretch done in PD. From what i've read, it's never really as simple as a single process. You need to analyze the sound for noisy transients, use smaller/faster window sizes for those..and then use longer windows for your more tonal parts. Then you need to design a good process to playback the stretched audio accordingly.

a few people made attempts at porting 'aubio' (i think that's correct?) to pd, which was a good looking library of audio stretch tools. not sure how far that got, but i never saw a completed product.

I did a few attempts at timestretchable sample players in my DIY library, using a few of the simple early 90's techniques, so you might find something in there that sort of works, but there are much better techniques these days...just they don't seem to have found their way to pd too easily.

http://puredata.hurleur.com/sujet-1982-1.html

actually, looks like there might be some aubio tools available...

let's see what we can find (by making a thread here, and asking other people ;0)

or...i could just use the search function..

http://puredata.hurleur.com/sujet-6869-1.html

doesn't look easy...and also is maybe more just for the analysis, rather than for the actual playback?

there have been some nice analysis patches posted here lately...i wonder how far we are from hooking those up to some good playback engines???

any ideas, anyone?

Thanks mod, your DIY library of patches is an awful lot of fun to poke through! Your timestretch patches are great, but still not quite what I need.

I've chopped and changed the I07 tutorial into an abstraction which takes inputs of: bang to play the loop, timestretch %, and filename to load.

There's no inherent looping function, that's something I'll have to work up outside of the timestretcher, and could maybe include into it later on... but figuring out how to drive a consistent loop from a sample of given length being stretched by a given value shouldn't be too difficult. I assume. ...I hope.

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

I don't really know what you've tried, but a couple of years ago there was a thread where I posted this granuler time-stretcher:

http://puredata.hurleur.com/viewtopic.php?pid=14588#p14588

It's not perfect, of course, but it might be worth a look if the phase vocoder is a bit much to handle (I think the granular approach is easier to understand, though it may not always sound better). It might be better if it was more asynchronous, but I've never bothered to try. I believe mod is right, though, about stretching audio with sharp transients such as drum loops. It requires some pre-analysis to make sure the transients are just played back without stretching in order to preserve them and avoid the smearing.

bsaylors [pvoc~] can be usd for time stretching a soundfile. sound quality and cpu usage depend on the window size you're using.
also csounds partikkel does an amazing job, but it's cpu intense and complicated to handle in pd. have a look at my Nigredo patch in the patch section for an implementation of that.

i use [pvoc~] on a new project and ran into a strange bug (?). i record into an array and then use pvoc~ to play the array. pd crashed every time i tried to record to the array. i found out that if you assign the pvoc to play a certain array only _after_ you have recorded the sample (via "setarray xxxx"-message) it works just fine. strangely enough, it works just fine after that (the first recording). i can resize the array, record into it. no problem. weird.

osx 10.6.8 / pd-ext 0.42.5