slowing down sound - a lot.... suggestions needed!

Question

hello, i am working on a project where i crash together 2 foot copper and steel boxes in mid air (using a pneumatic colliding machine i made - yes, silly...), and i use old 16mm film to record slow motion of the collisions, and record sound using piezo film sensors and standard mikes. what i would love to know is possible ways to slow down the audio 40 - 100 times (yes, a lot) using digital means. i am NOT a sound designer, but i am trying as much as i can. i have tried phase vocoder stuff, which smears the transients and all detail, i have tried paul's extreme sound stretch (which does the same, obviously, though is more suited to extreme stretching), i have tried reel to reel, re-recording and building the stretch over several passes (which becomes quite noisy), and digital resampling (which from 192kHz ends up being quite glitchy....) i am wondering if recording 1bit 2.8MHz DSD is worth considering - giving me enough sampling speed to really play with resampling and slow time (i realise this will lower the pitch also....) or are there other options to consider??? what interests me most right now is getting the sound accurately slowed, so that i can understand what is happening and link it to the visuals of the film. i may perhaps end up editing heavily, or even foley-ing the whole thing, but at the moment i really want to understand what is happening sonically at very slow speeds - to be able to hear the smallest times and what is going on there....
in the past i have just recorded at 192kHz, slowed the sound down in Soundtrak Pro's stretcher, and then added transint peaks on top to preserve some of the punch that i imagine would be happening (!) this is a very old version of the project (two more itterations now), but you get the idea - 2 foot tall copper and steel boxes in mid air, wanting to get an accurate sonification of the collision....

[vimeo] 28642463 [/vimeo]

i hope someone has been playing with slow-down in this sort of range of 40 - 100x and has some possibilities worth exploring that they are willing to share. thanks, sean.

Answer 1

Try XForm. I've had decent results, granted it takes a VERY long time to do it's processing That's only only other option I can think of. However, it's a processor which is going to add interpolation points (e.g. new sound generation). The only other option to retain transparent quality, that I can think of, is do a pitch change with no time compression, meaning the pitch is cut 50% for example, and the sound is elongated by 2x it's original. But it's only viable, in the case of how much you want to slow it down by, to have the original recording be 192kHz.

The deal is, slowing down by that much is going to create digital artifacts unless you use the latter means (and accept working with the pitch drop). Period. There's no way around that, just like how Moore's law breaks down at the atomic level. There's a limit to the forgivability factor of time stretching. There are only so many 1s and 0s, and for each time you cut the speed in half, interpolated voltage points have to be created by averaging two voltages. When averaging, it has no way of creating all of the modulated harmonics which come along with those wave forms. Hence, it has a washy and stretched-sound. Especially because in that much of a slowdown, you're creating three or more "averaged", interpolated voltage points - which in of themselves are averages of over voltages. The fidelity and quality of these interpolations exponentially decay when compounded.

Those ideas you tried sound solid (DSD may be an option, I'm just not familiar with it), and the above are some other recommendations I can provide. But unfortunately within the constructs and limitations of 'standard' digital audio formats we all work with, there is a very finite limit to forgivable time-stretching within the digital domain. And there always will be artifacting, sometimes not easily audible. The art is learning where to walk the fine line of doing the stretching you need but not beyond the point in which it reveals the problems inherent with digital stretching. But in some cases, like the speed adjustments you are trying to get, are virtually unobtainable without receiving audible artificting in return.

Unfortunately that's the best answer I can provide, both from my own experience and what I know about how digital time stretching functions. Maybe there is a 'secret sauce' I'm not familiar with, and maybe somebody knows what it is. Although from where I'm sitting, I unfortunately don't see a perfect no-strings-attached solution to the problem you are presenting.

Answer 2

Try:

paulstretch

or

photosounder

Answer 3

Sorry to sort of hijack this thread by asking a question within question! But is it possible to record at a high sample rate, say 192kHz or even up to 1bit 2.8MHz DSD (granted I dont know much about DSD!), and then play it back at a 44.1kHz or 48kHz for example to slow it down? In a similar way to high speed cameras that record at 1000's of FPS and then being play the footage back at around 24-30 FPS?

Answer 4

Photosounder probably comes the closet to doing what you want to do. Mostly if you fiddle with the parameters for precision/resolution in the configuration file (see manual). Also see this http://www.youtube.com/watch?v=zTi_HwdlJ20

PaulStretch is nice but not as precise in time, even when slowing things just 8 times it sounds like a smear, and most other time-stretching algorithms aren't suited to such ratios so they sound too unnatural.

Answer 5

This does a good job of slowing things down.

http://www.csounds.com/cecilia/