Time-stretching with PaulStretch

So this thread on the blue about slowed-down Bieber led me to PaulStretch, which is a fairly aggressive piece of (open source!) audio time-stretching software. It's neat, and is (on Windows, at least) super easy to play with.

I think it'd be neat to have more "here is a cool toy, have at it" threads over here, and so here's one of them.

I've been fucking around with PaulStretch all day and having a ball—it's basically, at the default settings, an Instant Ambient Music Generator (insert 3 minute pop song, get 24 minute ethereal meditation or horror ride, depending on the content). I put together a sort of ear-test quiz of some pop music over here as an example of some of what it does to stuff, if you want to dip your toes in.

But it's pretty tweakable. The 8x slowdown is just a default setting, and the degree of audio "smear" it produces is also adjustable, and there's a number of stretching algorithms that, again, I haven't really tried tweaking.

So I'm seeing this as potentially useful as a creative tool, aside from just the novelty of slowing down existing songs—I'm thinking for example of intentionally recording a short instrumental loop in real time and then stretching it out to create a very ghostly sort of backing track to then use as the harmonic base for a song. But I'm curious what other people can see doing with this, or with related tools.

posted by cortex (30 comments total)

Caveats:

- I haven't tried building the linux source on my Mac, so no idea if that'd work. Curious to know.
- Interface isn't super polished; scroll bar for time is twitchy and sometimes unresponsive. I have taken to restarting the program sometimes to try and get weird behavior to go away.
- Crashes a bit too. Seems to dislike certain files.
- No mp3 writing capability. You'll need to encode with something else after writing out a WAV file.
posted by cortex at 3:38 PM on August 17, 2010

I'm thinking for example of intentionally recording a short instrumental loop in real time and then stretching it out to create a very ghostly sort of backing track to then use as the harmonic base for a song.

This is pretty much what I thought of last night while listening to Sigur Bieber.

Since you've been playing with the software, how well does it work for something that isn't a pop song? Like, if you put in a bit of Stravinsky or something? Did you try anything like that?
posted by sleepy pete at 3:42 PM on August 17, 2010

I haven't tried any straight-up orchestral music yet, though I've tried a few things that are not typical pop production. My first impressions in terms of style/genre effects:

1. Loud drums = constant crashing ocean waves. Snare transients become these great big rolling piles of echo and pre-echo due to the way the software tries to take the edge of stretching artifacts. Same thing happens to everything else as well, but it's not so noticeable for stuff with a less spiky volume envelope. So stuff without drums or with drums mixed lower tends to be smoother and less washed with white noise.

2. Similarly, anything that features transients of any sort as the defining sound is going to end up extra smeared beyond recognition. Rapid fire vocals become a wordless vocal noise, trem-picked guitar become one long note, etc.

3. Really minimalist production with present vocals tends to lead to something that sounds a lot like, well, a really slowed-down vocal. Not necessarily bad, but a very specific sound. Lusher harmonic content makes for more sweeping ambient feeling.

4. Slow music + this = kind of interminable music. Something that has little harmonic movement in the first place becomes basically a glacier. If what you want is a really slow wash, that can be interesting—Imogen Heap's Hide and Seek becomes this incredibly slowburn thing.

5. Upbeat stuff with more frequent harmonic changes tends to be a little more engaging. It's still slow pad stuff, but at least your brain recognizes that it's listening to regular movement. You Can Call Me All becomes an odd sort of film-denouement score but at least it's moving.

6. Even the quickest portamento really really stands out after you do this. If you think you sing nice and steady and hit the notes on the head, try this on one of your tracks and discover where you are, in fact, human after all. The slow climb and dive and waver of unsurgeried vocals is actually pretty interesting to listen to.

I don't have any Stravinksy handy, but putting on Fur Elise as a random piece, the sprightly piano becomes a slow, ponderous synth-string/synth-choir dream sequence soundtrack.
posted by cortex at 4:05 PM on August 17, 2010 [1 favorite]

To answer the question, "what would Sigur Rós sound like sped up to eight times original speed?", I made "Bieber Rós", the Sigur Rós album ( ) in its entirety, compressed to about nine minutes.

Cross-posting from the Blue.
posted by goodnewsfortheinsane at 4:45 PM on August 17, 2010 [1 favorite]

My conclusions after a little while experimenting with this program:

- The Who become even noisier
- MC Hammer's "U Can't Touch This" becomes terrifying
- So does Nightwish's "Nemo"
- Guns 'N' Roses' "Sweet Child O' Mine" is still completely recognisable at an eighth of its normal speed, and it's still pretty awesome
- Florence & The Machine works particularly well. I tried it with "Howl", and while most of it was just good ambient noise, the high note in the chorus gave me chills.

I'd be interested to hear other people's experiences...
posted by ZsigE at 5:54 PM on August 17, 2010 [1 favorite]

Thanks for posting, cortex.
posted by flapjax at midnite at 7:17 PM on August 17, 2010

This looks very interesting - thanks cortex. I haven't accessed it yet, but for those who have - is it possible to mess about with a WAV file in Windows, save the "messed about with" WAV track and then transfer it onto another machine (i.e. my Yamaha AW2400)? Apologies if that's finger puppets territory but, as some of you will know from previous comments of mine, my IT capabilities are rudimentary to say the least....
posted by MajorDundee at 1:16 AM on August 18, 2010

Totally doable, Major. It lets you render to WAV (and, important and useful note, lets you render just a small window of your creation if you prefer, which is convenient when ballooning input up by an order of magnitude), so what comes out of that is just another wave that you can do whatever you normally do with WAV files.

It also renders to OGG, which takes me back to my linux zealot days but probably won't be all that handy for anyone.
posted by cortex at 6:47 AM on August 18, 2010

I made "Bieber Rós", the Sigur Rós album ( ) in its entirety, compressed to about nine minutes.

That is totally fucking with my mind right now...
posted by dnash at 7:36 AM on August 18, 2010

- I haven't tried building the linux source on my Mac, so no idea if that'd work. Curious to know.

It not only can be done, it has been done for you.

My first experiment, transforming Tony Rice's Blackberry Blossom has not gone well... too many highs, acid feel, audio lemon juice.
posted by weston at 10:15 AM on August 18, 2010

If this catches on, people might start putting split-second 'easter eggs' into their songs that become recognizable when stretched.
posted by umbú at 12:31 PM on August 18, 2010

Totally doable, Major. It lets you render to WAV (and, important and useful note, lets you render just a small window of your creation if you prefer, which is convenient when ballooning input up by an order of magnitude), so what comes out of that is just another wave that you can do whatever you normally do with WAV files.

It also renders to OGG, which takes me back to my linux zealot days but probably won't be all that handy for anyone.

Thanks cortex (mutters under breath "Does anyone speak English round here? What the fuck is he talking about? Ballooning input?? OGG (is that Oh Good God?) linux zealot?? Does that go inside that sock puppet thing he's mentioned before?? Is this some kind of bizarre sacrificial rite.... Oh my god - so MeFi is really a weird cult with some beardy geezer called mat-how-ee as Godhead......). I'm fooling around mate - inability to understand IT jargon is totally my problem, clearly not yours. Does make me smile sometimes though....!
posted by MajorDundee at 12:32 PM on August 18, 2010

Major, if it makes you feel any better, I always have to google at least three words from every comment you leave out here. They say they speak English out there in England, but I have my doubts.
posted by man vs sun at 12:36 PM on August 18, 2010

gnfti, Bieber Rós is surprisingly listenable. I can imagine music grad students doing this to symphonies to better map the form of a piece.
posted by umbú at 12:42 PM on August 18, 2010

Major, if it makes you feel any better, I always have to google at least three words from every comment you leave out here. They say they speak English out there in England, but I have my doubts.

A comment of dubious confibularity, as I'm certain my fellow countrymen will wholly trinumbulate.
posted by MajorDundee at 1:47 PM on August 18, 2010 [3 favorites]

Alternatively, this may be of assistance
posted by MajorDundee at 1:52 PM on August 18, 2010

This effect rules my life now.
posted by fuq at 8:30 AM on August 19, 2010

If I get some spare moments, I'd like to use PaulStretch on some ringtones.

Is this one of those things where it would be beneficial to start with a lossless format?
posted by box at 8:57 AM on August 19, 2010

I don't really know, box, but I'd guess you don't need to worry about it. PaulStretch will have such a profoundly destructive effect on the audio that lossy vs. lossless probably isn't all that big of a deal, though I guess really low bitrate mp3s might be noticeably different in some cases.

And I love the ringtone idea, yeah. A friend of mine suggested using jingles and sound logos as well, which I think would be a lot of fun esp. considering the tendency of the latter to have a fairly high harmonic density and attention to detail. 20th C. Fox's fanfare becomes a five minute horn requiem; Intel Inside becomes the ringing of cathedral bells; By Mennen! becomes god knows what.
posted by cortex at 9:09 AM on August 19, 2010 [1 favorite]

I just posted a "screwed" version of a song to Mefi Music. Different tools, different degrees of slowth, same basic idea.
posted by jtron at 11:33 AM on August 19, 2010

cortex: "Snare transients become these great big rolling piles of echo and pre-echo due to the way the software tries to take the edge of stretching artifacts."

You give the app far too much credit. This is just what happens when you use a long analysis window and multiple overlaps (usually because you prefer accuracy for frequency spectrum and are willing to lose time accuracy for this).

You always have time distortion when you do this, and you always have spectral distortion. This app is just banking on the frequency spectrum being more interesting than the timing.

If you really want to go down the rabbit hole for using this stuff in real time and getting at the raw parameters etc. I have found nothing better for the purpose than the various pvs* opcodes in csound (I got some nice results with a duo with Minus215Cee where he played sax and I froze spectra of his sax timbre on my netbook).
posted by idiopath at 11:51 AM on August 19, 2010

Oh, I don't mean to suggest the software is doing something super complicated or anything, just that it's doing other than just naive time stretching. I think the effect is interesting (and the psychological interpretation effects as well—"rolling waves", etc. being obviously a matter of our brains trying to make a pattern match from nothing in this case) but I understand that it's fairly simple rolling-average type calculations under the hood.

Have you poked at PaulStretch at all? It provides configurable options for things like time ratio and window size (and a bank of other little toys on the second tab of the interface), I'd be curious about your thoughts on some of that as a hardcore sound mangler.

Never have quite gotten into playing with csound. Love the idea of it but idea and need have never so far collided in a way that got me digging in.
posted by cortex at 12:39 PM on August 19, 2010

some of the additional functionality the csound tools provide that I don't see in paulstretch (mind you some of these are implicit features of csound and not documented specifically in regards to these opcodes):

Real time mic or line input can be used (as well as pre-analyzed sound files in a special frequency domain format for CPU efficiency if desired).

Real time control of pitch and time (I made an instrument where the position of a joystick moves the position of playback - for example one file I play with is my roommate reading a poem, and if I hold the stick still it sounds like him making a single vowel sound indefinitely until I nudge the synth forward or backward through the text).

Multiple simultaneous playthroughs, of arbitrary duration. One could even use granular techniques with hundreds of automatically generated voices (though of course this may not be advisable in realtime).

For non realtime input, the input can be moved through in an arbitrary and programmatic manner: it can speed up or slow down dynamically, following various kinds of curves with added jitter or randomization if so desired. This could also be done in real time with a bit of extra work.

Arbitrary operations on frequency, amplitude, and phase data before the resynthesis stage (for example this could be used for autotune style pitch correction, estimating and adjusting the base pitch -- there is even an option for preserving the original formant).

Morphing between spectra (for example making a cello's tone evolve into an oboe's).

Arpeggiating through the spectrum.

---

The advantage that Paulstretch has over all of this is that it is a point and click GUI tool rather than a programming language with a syntax based partially on assembler.
posted by idiopath at 3:50 AM on August 20, 2010 [1 favorite]

Looking a bit closer paulstretch can also do real time scrubbing through the input, I missed that.
posted by idiopath at 3:54 AM on August 20, 2010

Easy-peasy.
posted by box at 6:01 AM on August 20, 2010

Arbitrary operations on frequency, amplitude, and phase data before the resynthesis stage.....The advantage that Paulstretch has over all of this is that it is a point and click GUI tool rather than a programming language with a syntax based partially on assembler.

Phasers on "stun". Alternatively, learn to play an instrument. Might be less of a headfuck. I'm kidding, I'm kidding......just couldn't resist a playful dig in the ribs...
posted by MajorDundee at 11:21 AM on August 20, 2010

As for a Mac version, I know this got linked in the FPP, but a friend of mine has totally unsupported OS X builds available.
posted by mkb at 12:43 PM on August 20, 2010

MajorDundee: "Alternatively, learn to play an instrument."

My major project lately has been experimenting with software systems (including this phase vocoder analysis / resynthesis stuff) that can be used as musical instruments. I am actually partially interested precisely because I enjoy the headfuck potential.
posted by idiopath at 1:04 PM on August 20, 2010

I am actually partially interested precisely because I enjoy the headfuck potential

lol. Cool response idiopath - glad you took my fooling around in good part!
posted by MajorDundee at 2:10 PM on August 20, 2010

I've been trying this out on some songs from the Neko Case album Blacklisted, and it goes pretty well, but that might be cheating, considering that the production has an ambient character to start with...
posted by weston at 8:37 PM on August 23, 2010