Uh oh, maybe I built it up too much haha. It’s subdued, but I’m hearing what I think is a cool mixture of the birds and insects in Kurodama’s night recording (the impulse) and the atmosphere, wind (and bells) in my day recording. The bells strike like a high pitched drum roll partway through, and there are some interesting dynamics I’m not sure the source of…
In my case, I’m curious what exactly convolving a long form sound on another does. I understand the idea (well sort of, not really the math) behind using an impulse response to derive a convolution reverb. But with a piece of music (or field recording) that has peaks and valleys (and frequencies aren’t present at equal amplitude like they ideally are with a sweep)…I’m curious what that does
That sounds really cool. Can’t wait to hear it!!
I like it when sounds emerge unexpectedly, too. For some reason, Miles Davis’ trumpet arose out of stretched version of Jason’s farm recording. I’m guessing it’s that chicken again.
Maybe someone else can speak more intelligently to that but my experience doing this a bunch with soundhack back in the day (but using two 30-60 second sounds, nothing as long as 45 minutes – it still took a really long time on my laptop back then) is that it’s incredibly dependent on the source material. There’s a kind of smearing of the sound and a kind of filter resonance effect, but otherwise imho results feel pretty unpredictable depending on the sounds involved. If it didn’t take so damn long to do it might be easier to build up an intuition for how things will mix together, but I’ve never been very systematic about it.
Here’s 60 seconds of an excerpt from Matt Turner & Jeff Song’s Love & Fear album (with lots of unintentional line noise in the recording, I should have previewed first) convolved with 60 seconds of a tape of the Minneapolis Symphony Orchestra playing Mahler’s 1st symphony.
The Mahler excerpt:
The Matt Turner / Jeff Song excerpt (plus tons of line noise!)
And the two convolved together:
Also! Here’s the 15x speed convolution of the field recordings – I’m probably going to play a bit more with this because there’s time and why not:
Apologies if you get this already, but every sound being convolved plays out for the duration of the impulse response file. So if you have a long impulse you very quickly get many layers of sounds playing concurrently.
I’ve played with this quite a bit - at one extreme using a bunch of off-station shortwave as the impulse response, at the other the Funky Drummer drum loop. Sharp dynamics and shifts in the sound are most obvious if you only send short bursts at the convolution algorithm, otherwise the number of layers happening results in that slurred kind of effect.
When I played with using a drum loop as an impulse response, I sent short synth stabs into it at the start of each bar and set the tempo so this created a constant beat, with the result a bit like drums through a vocoder. The fun was then playing with the envelope of that stab - it created weird blurring effects that are really their own thing. I should go back to that track.
sorry to flog a dead off-topic topic but
i am one of those special dummies with a rather absurd physical copy of this thing! haaaaaa oy vey. i look at it frequently.
unsurprisingly, julius smith has nice treatments of both cyclic and acyclic convolutions, as special cases of OLA STFT. (“spectral audio processing” is another one that is arguably worth having on dead-trees.)
if you are really interested i could probably pitch in for some development of some kinda command-line tool for this. (which is to say: non-realtime cyclic or acyclic convolution of a relatively long audio file, with a relatively short file as IR, in freq domain.) not that soundhack won’t do the job, but heck, it’s 2019
My dummy implementation is in cython – I was waffling on doing the FFT b/c I couldn’t decide if I should use numpy since it’s already a dep of the project, or something like KissFFT for the FFT which I could call from the c extension more directly and maybe avoid some overhead?
Bottom line is I haven’t thought through this very much but I would love to bring this into pippi and bundling it as a standalone cli thing sounds even better. (Can be both!) Would love to work on that!
(PS I’ve been dabbling in desktop GUI stuff, made a simple piano roll thing… would love to make this a standalone tool thing, I have nothing but <3 for soundhack but yeah I also can’t use it on linux anymore so! A standalone cross-platform tool for that would be sweet. I’d love to work on that.)
Here’s the time-domain version fwiw, not that it’s anything!
It’s definitely going in an interesting direction! I really like it when the more percussive sounds come in (bells?) and you can hear the sounds in the sounds kind of. Quite eerie and alien. What’s your plan? Just do the convolution and make that the final track? or is this more to produce the raw material? Sorry if you’ve stated this before, I’ve not been watching this thread for a bit and maybe I missed something going through the posts.
Just like most others my start has been slow. Right now I’m randomly messing with the field recordings in the attempt to create raw material to work with and see where it takes me. My approach would be to mostly work with loops, extracting parts that have some sort of inherent rhythm to themselves, layer things. I think I am mostly inspired by Helm’s album Olympic Mess.
The sped-up convolution sounds kinda neat by itself but yeah now I’m thinking I’ll continue to play with it further. My idea at the moment is to treat the convolution as a wavetable (sampling from it, reducing it to a curve by envelope-following, etc) and use that as control params for further processing on the original field recordings.
Trying to keep the “balance” theme in mind, I might also treat the originals as wavetables/curves and do some more processing on one with wavetables from the other, etc.
So basically the plan is to keep screwing around.
Just realised this weekend coming is the last one before we finish up!
Whoa, that snuck up quick!! Thanks for the reminder. I will upload my two tracks today…