#21

Thanks! I played it. There’s a style of playing called “crying steel guitar” in country music that exploits volume swells and expressive vibrato a lot, however I don’t think my note choices would qualify as being within that style.

3 Likes
#22

Just to quickly add to this diversion, if you wanted some more pedal steel that is not honky tonk you could look into:

  • Susan Alcorn (she does a Messiaen piece and it’s beautiful)
  • Daniel Lanois’ Goodbye To Language (instrumentals based around him playing pedal steel while jamming with the settings on a Lexicon delay unit live)
  • the two recent 12k records involving Michael Grigoni
4 Likes
#23

Some conversation about convolution came up in the course of one of the LCRP projects, including a good link to the free DSP guide about how convolution works, and some further comments from @zebra and others.

1 Like
#24

basically yes - or to put it another way: each sample of output is the weighted average of a block of 100 input samples, the weights being the 100 samples in the convolution kernel. (or, still another way, the output vector is the dot-product or projection between the input vector and the kernel.) this part (the mathematical operation) is commutative, which is a little weird and fun.

as an operation on audio signals, it isn’t of course commutative because the signal is gonna be much longer than the kernel. basically, you shift the averaging window of the input as you go along.

actually you’re using it right - we are talking about convolution in the time domain. by a mathematical trick, convolving two signals turns out to be the same as taking the DFT of each signal and multiplying them. the FFT algorithm makes this much faster than brute-force sample convolution, especially when you can pre-calculate the DFT of the impulse response. but the power spectrum per se never really comes into it.

the more complicated tricks with using “non-uniform partitioning” are just about refining this method for reduced latency. (added latency being the tradeoff of using the FFT instead of a brute-force windowed convolution.)

(and i’m sorry for derailing a second thread with this discussion, but at least it is more directly relevant here…)

on-topic, if you happen to own iZotope Trash 2 it’s also worth digging into as a nice convolution engine. comes with some funny IRs, like animal sounds, and allows importing

7 Likes
#25

Here is an impulse response I’ve had good results with:

Edit, an example:

Edit edit: the sound I convolved this with was harsh and had a pretty different harmonic content. The tones of the original and the seed got amplified and it ended up coherent. I like long convolutions.

6 Likes
#26

Thanks, that explains a lot that I had an inkling of but didn’t fully understand. I’ve tried to read about convolution online but the descriptions of the math (which I don’t really understand in the first place) and how it’s applied in audio were always disparate to me.

I don’t think you’re off-topic at all, understanding how convolution works and how devices using it work is very useful in allowing us to use it to create interesting sounds beyond the trial and error approach of trying out your entire sample library on a sound and see what doesn’t sound terrible.

I think I might’ve worded the OP poorly because the first few posts are about gear - which is cool, I think it’s a nuance worth considering - whereas what I think is the beauty of convolution is that it’s pretty agnostic as to which plugin you use, they all load audio files and apply the same hundreds of years old operation to them and the live audio to produce the same kinds of results, meaning we can share these files once we’ve created them and inspire each other.

(If anyone has suggestions how to clarify the OP, I’m open to suggestions. I don’t want to fully discourage discussion about various plugins, modules and Max patches as they do add to the conversation.)

1 Like
#27

This sounds really nice. I’m always hesitant to do things with super long IR’s like this but this one has a lot more motion to it than I would’ve expected. I reckon the different harmonic contents of the two sounds help clear up potential muddiness.

1 Like
#28

I see people talk about convolving synth drones with field recordings, and it sounds incredible (the idea of these static waveforms bubbling with rain, leaf rustles or bird calls…)

… but in practice, if I load a field recording as the IR of a convolution reverb, or even setup a cross-synthesis phase vocoder in Max… it never sounds ‘right’. What’s the secret there?

#29

I’m not tremendously experienced with convolution reverbs, but sine years ago a composer friend of mine expressed surprise that they hadn’t taken over the world. This thread inspired me to experiment a little; in particular with the convolution reverb in Reaper. I was immediately struck with how much it was capable of, in terms of Impulse creation (it has an algorithm to create a response, or can load a file, as well as make a test tone and deconvolve). Then can apply various effects to a response (trim start and length, filter, reverse, gain etc) meaning you need little preprocessing.

I wanted to check my intuition for how the spectral content of the response translates into the reverb. Assuming the impulse has equal energy at all frequency bands, in a purely dissipative reverb each frequency band will monotonically decrease in amplitude over time. For a dissipative reverb with resonance, certain bands will decrease much slower than their peers. For a modulated reverb, certain bands will increase and decrease (with a downward trend). For a “feature reverb” all bands vary arbitrarily. For a discussion rhythmic reverb the bands drop significantly and have repetitive pulses of recurrence.

This understanding leads me to some thoughts. To use an arbitrary sound as a believable modulated reverb, you need the sound to have broad spectral content (e.g city ambiance). To turn this sound into a usable modulated reverb, I think you could take said sample and just have a shallow LPF pitch sweep down over time alongside having it fade out (shrink amplitude wise). In fact, enveloping it with an LPG might be useful here. I’d think you ideal want the release/decay to be in the form of an inverse exponential.

Another question I have, is: what happens during deconvolution if the test tone omits certain frequencies? Does behaviour in such absent frequencies get interpolated or is it absent from the resulting response? My asking is to say, if you had an impulse response (or indeed any sample) what if you made an impulse (or short pitch sweep) with the same spectral content (energy per frequency) as your entire sample. Would deconvolving you sample against that “test tone” make a “more meaningful” impulse response?

In any case, I plan to experiment some more. I made a “clap” impulse response in my garage and got a reasonable result with a 2 seconds tail (after various trimming, fading and filtering as there was background noise), and a test tone based/deconvolution response in my stair case, also with about 2 seconds of trail. I also tried Paul stretching both to interesting effect (though it added, or at least significantly accentuated resonant frequencies). Hopefully I’ll be able to put some experiments to more artful effect: thanks for an interesting thread!

3 Likes
#30

I find this topic relevant to my interests because I recently ordered a Tasty Chips ECR-1 Eurorack Convolver and look forward to trying out some of these IRs once it arrives.

I sort of had an idea of how convolution reverb worked, but I found this SOS article on creative convolution helpful in gaining a deeper understanding.

There are a lot of IR resources in this Reason blog post on finding impulse responses.

Thanks for starting this thread!

1 Like
#31

So here comes the promised drum experiment:

I don’t have a real drum so I created this drum sound by hitting the body of my sazbüş while muting the strings and pitching it down quite a bit:

Then I hooked up a camping table to a contact microphone.
I tried using a dish-brush on the table to get something similar to a brush pattern on a snaredrum.
Dry brushing:

Through the IR:

It doesn’t fully work, partly because the IR doesn’t have any snares but I think primarily because when drummers use brushes on a snare, they press down and mute the drum while brushing, which creates these stark dynamics where the tail is completely cut off.

One could perhaps crossfade the IR with another that has a much shorter tail based on the dynamics of the input but I didn’t want to take the experiment that far today.

As unrealistic as it is, I’m pretty happy with this sound, probably because of how punchy and compressed it is.

I also ran this percussion patch I made on the modular through it:

Which sounded like this:

This one has some frequency shifting controlled by the envelope of the incoming sound. It doesn’t seem to add much more than a bit of phasing in this case. Overall I was a bit disappointed with how the frequency shifting turned out. I expected it to more realistically model deformation of the skin and it did work for extremely hard hits on a very loose skin (which I didn’t record, sorry) but otherwise it just added undesirable phasing.

There’s also a bit of ground hum that gets amplified by the convolution. Pretty easy to edit out if anyone were to reuse this for something.

And here’s me freeform tapping the camping table:

The frequency shifting here sounds a lot more natural (although it does go into phaser territory). Excuse my terrible drumming, I’m not a drummer or percussionist in any way.

For all of these I re-pitched the IR to whatever suited the sound. One thing that was a bit bothersome was that I made a very open-sounding resonance, which made it unsuitable for bass drums for example.
I would probably use pitch-shifting over frequency shifting if I were to do this again. I read somewhere that freqeuncy shifting was supposed to sound better for repitching drums and there may well be merit to that when you’re changing an entire recording and leaving it at pitch but in this case it sounded worse.

I did try some other sounds through it to see if there were any notable results but they mostly just sounded like sounds going through a very tiny resonant chamber, which I kind of expected.

Overall I’m not as disappointed as I may sound through most of this post. I did enjoy the sounds I got and I will likely use them at some point.
Hope some of you find this useful.

6 Likes
#32

Neat idea, and I like the results! There’s a whole class of interesting things that might be possible here… Shepard tones, overlapped ascending and descending sine sweeps, or playing with convolution engines that let you modulate the “pitch” of the IR.

2 Likes
#33

Something I like to do for ambient tracks is take a synth patch I like and either put a sample together of all notes across several octaves playing for 5 - 10 seconds and fading out, or just all notes within whatever scale the song is in. Make the level of the sample quite low and load it up into a convolver and anything you play into it will be embedded against a gorgeous wash of sound. Sometimes I mix a noise burst or drum sample into it at the beginning too to get an initial reverby response.

I also use convolution for modelling physical bodies. Attach a contact mic to a guitar body, or a baking sheet, or a bottle or anything really, and clap as loud as you can near the object. The mic will pic up the response of the body. Slapping any of these on top of a synth sound can make it sound a lot more organic and interesting, even if you’re just playing raw oscillators through it.

10 Likes
#34

One idea I’m interested in (but haven’t actually done much of yet :-/ ) is ‘found’ impulse responses. Here’s one example, sampled from a facebook video (Brutalism Appreciation society post, sorry if you are not a member or left because of too much “IS THIS BRUTAL!?!?!” + thus can’t see) of someone exporing a massive abandoned silo. At one point they clap their hands to demonstrate the extremely long and weird echo. Very lo-fi and noisy as an impulse of course, but interesting to play with :slight_smile:

7 Likes
#35

I’m going try to make a couple more convolutions similar to this soon. I tried it on a couple different sounds and it made vocals completely unrecognizable in a very interesting way, like additive synthesis but working exclusively with the temporal offset of each partial.

I think, while your suggestions are interesting, convolution is very easy to overdo and end up with hard to define mush. I think part of what makes that particular is that it is a single sweep that represents every single frequency, but only for one sample of the tail, meaning nothing gets overrepresented and resonates too much.

I’m thinking an image editor would be the best way to work with this style of impulse response. I’ll try to get that to work and see if there are any interesting results.

1 Like
#36

Background noise doesn’t seem to ruin IR’s, especially ones that extreme. Since it’s mostly like white noise, it will mostly just add a bit of background smear to the sound.

Sounds nice on the guitar, despite being mono:

#37

Got inspired, and tried convolving the ice sound found in this video:

Here’s the extracted bit:

Applying it to @oot’s slide guitar sample had a surprisingly conventional reverb sound to it (truncated it to 30 seconds).

So, I figured I may want to try using some more percussive samples. I went with some Kalimba and Hang Drum one-shot samples I found on freesound:


Below is the convolved output. I imagine with some additional processing variation, one could build up an interesting multi-sampled instrument.


7 Likes
#38

The mention of a multi-sampled instrument makes me think it would be really cool to have a multisample convolver plugin, where each note on a keyboard could convolve the input with a different sample. I bet you could get some really cool stuff playing noisy sounds into it and using the convolver itself to create something tonal out of it. I wonder how much processing power something like that would take though, especially if you wanted it to be polyphonic…

6 Likes
#39

oh man… i have those somewhere… i bought them when i was using MOTU’s built-in convolution in DP… and dropping in spirit canyon’s weird audio files… it was very cool. i probably have them on a backup drive somewhere…

4 Likes
#40

Thanks for posting that, really cool to hear, sounds good indeed. Yeah shame that particular sample is mono, thats a disadvantage of this found-sound approach, too many phone-camera mono videos

1 Like