I kind of got some unexpected attention here with the last one.
This is mostly Pd (sounds and post-processing) except for a couple "live electronics" overdubs. Mixed in Audacity.
New composition
I kind of got some unexpected attention here with the last one.
This is mostly Pd (sounds and post-processing) except for a couple "live electronics" overdubs. Mixed in Audacity.
I enjoyed that
Thanks. Here's a screenshot of one of the reverb algorithms. I figure someone will probably be curious.
wow thats nice!
Incredible! Love the picture of the reverb algorithms. How long did it take you to make that? And what did you have to have an understanding of?
The reverb algorithm is an ongoing thing. I started working on it around August of last year. I've just been reading a lot of papers and patents and stuff, analyzing and reverse engineering various algorithms, and generally thinking about it a lot. I'm still experimenting with it and making improvements.
Hi Acreil,
Interesting composition and a serious amount of micro-structural detail in relation to your reverb algorithm. It would be helpful if you could provide some commentary on your SoundCloud timeline as to what is doing what and when, since you are using other instruments and sound sources.
R
Well, there's lots of stuff sort of coming in and out, so I don't think it does a lot of good to add comments to specific points. And I'd have to describe it at length to be useful at all. If I ever do a commercial release it will have to include a deliberately obscure 30 page paper with copious citations and strange hand-drawn diagrams. But I can write a few (!!) things here.
The first section (first half, approximately) has one "main" sound and some overdubs. It's mostly random processes triggering 16 enveloped sines at some harmonic (up to 512) of 30 Hz. These are all mixed and passed through some nonlinearities ([cos~] and [tanh~]), and that then goes through some chorus (emulation of a classic "string ensemble" type effect), reverb (based on a Lexicon algorithm) and reverse pitch shifters and feedback loops and stuff. The mix and various other parameters are randomly varied over time. It's basically a waveshaping or FM kind of sound. Because all the possible tones are harmonically related, it sounds consonant, but simultaneous tones also change the timbre.
On that section I overdubbed some nasty stuff with a half-working cassette tape delay apparatus (Sony TCM-5000ev with Alesis 1622 mixer). I had a breadboarded motor speed control circuit and everything was sort of breaking down and falling apart as I was using it. The input to that was a Dave Smith Instruments Evolver, Korg KR-55B and a guided meditation cassette. There are 4 tracks of this, two of them are reversed. There's some EQ (done in Audacity), but it's otherwise dry. I'm sort of drawing from early electronic music and musique concrete, but also early industrial and minimal synth. I thought it would be a funny contrast to the other stuff going on. It's like several different eras of music happening at once.
There's also some other Pd material in there, some sort of random number sequence FM thing. It's supposed to conjure images of 70s computer music, but it's mostly mixed in the background. I used my nice reverb on that sound.
In this section it can be hard to discern which sound source is which (partly intentional), but they all have a different sort of spatial impression.
There's a transition part at around 19 minutes. It comes from a recording of me banging on an attic door spring with a screwdriver. I processed this through some granular-like stuff and then made a few variations of that with some other algorithms, including some resonators based on 3 delay feedback delay networks. It eventually turns into a sort of background drone and finally disappears at around 28 minutes.
What also comes in around this time is a chaotic sound generator based on biquad filters. I'd been corresponding with a guy who does research on "autonomous" sound generators, i.e. chaotic systems that monitor their own output. I'd also been reading about how biquad filters get very noisy as the ratio of sample rate to cutoff frequency becomes large. Each octave increases noise by something like 10 or 12 dB (so oversampling comes with a penalty). I played around with this and found that at a sample rate of 48 kHz, a notch filter with Fc around 1 Hz and Q = 1000 is essentially a chaotic noise generator, and it takes a very long time to settle when the frequency is changed. I had several of these modulating sine tones, with some feedback loops and [fiddle~] objects to track the pitch and change stuff around. The sound was too dry on its own, so I put it through a 16 delay FDN resonator thing, frequency domain time variant decorrelation, and my nice reverb algorithm.
(incidentally I'd really like to see some double precision biquads in Pd, including double precision coefficients)
The last part that comes in starts to become audible at around 22-23 minutes, and is FFT-based additive synthesis with 1024 harmonics and again a fundamental frequency of 30 Hz. I'm randomly varying the spectrum in a variety of ways, writing the inverse FFT to a double buffer and then cross-fading between two [tabread4~] objects. So the FFT is synthesizing the waveform but it otherwise works as a standard wavetable oscillator. There are 4 parts here, with 1024, 128, 16 and 2 oscillators. I'm linearly detuning them. This was done by Jean-Claude Risset and Stockhausen for a "frequency domain grating" effect. I find it's directly analogous to a technique used by John Whitney in his computer animations. He called it "digital harmony". For this I used the version of the "nice reverb" that's in the screenshot above.
The remainder of the track is this additive and chaotic biquad stuff.
Hello acreil
Very interesting, I enjoyed a lot. Good mixing quality overall, I'd say it is very "balanced".
@acreil said:
If I ever do a commercial release it will have to include a deliberately obscure 30 page paper with copious citations and strange hand-drawn diagrams
That would be interesting. I'd go for 1 copy.
@acreil said:
It's mostly random processes triggering 16 enveloped sines at some harmonic (up to 512) of 30 Hz
That's always pleasant sounding. I did it once by harmonically (narrow) filtering white noise.
@acreil said:
... I played around with this and found that at a sample rate of 48 kHz, a notch filter with Fc around 1 Hz and Q = 1000 is essentially a chaotic noise generator, and it takes a very long time to settle when the frequency is changed
Let me try to better understand: in practise you cut a frequency band centered at 1 Hz with a very narrow BW (high Q), right? It behaves as a chaotic noise generator when it is stimulated by which input signal?
@acreil said:
(incidentally I'd really like to see some double precision biquads in Pd, including double precision coefficients)
do you think this could make some improvements? Usually I was told that 32 bit (single precision) coefficients should be a lot for most of today DSP task (at least in HW), 40 bit is kind of "luxury", 64 bit (double) ??. I saw some of the possible advances thanks to Katia's videos (especially
in indexing large arrays) but I wander what is the difference with digital filtering.
Thanks for your insights into the piece.
Alberto
@AlbertoZ said:
Let me try to better understand: in practise you cut a frequency band centered at 1 Hz with a very narrow BW (high Q), right? It behaves as a chaotic noise generator when it is stimulated by which input signal?
Here's the issue: to tune a biquad to a very low frequency (relative to sample rate), the coefficients have to be rather extreme. In any biquad filter you'll have some tiny residual oscillation (limit cycle) due to quantization inside the feedback loop. At low enough frequencies these become as big as the input signal. You can just put a constant (DC) through it and get noise out.
I figure that 2011 we have enough CPU cycles that we should be quality obsessed, but this is a disincentive to oversample if it means the filters get noisier by 10-12 dB each time you double the sample rate. I don't think there's usually any good reason to use Fc = 1 Hz and Q = 1000, but I'd like to be able to use Fc = 40 Hz at Fs = 192 kHz without having to worry about it sounding horrible.
(I'm getting this from Robin John Clark's PhD thesis, by the way).
@AlbertoZ said:
Good mixing quality overall, I'd say it is very "balanced".
Thanks, I was still worrying about that somewhat.
Oops! Looks like something went wrong!