A terabyte of harsh noise, stored on 233 DVDs | Hacker News Reader

A terabyte of harsh noise, stored on 233 DVDs

jliat.com

52 pointsurras41 comments

Oops, the artist's source code has a bug making the 233 DVDs contain not what he thinks they contain... All his source files [1][2][3] attempt to generate the random noise with:

  // basically no wimpy numbers - only fat ones
  if (s > 0 & s < 30000) {s = (short) (s + 30000);}
  if (s < 0 & s > -30000) {s = (short) (s - 30000);}

s is a Java 16-bit signed short initialized to a random value. His intent is apparently to modify s to generate sound samples in the ranges [-32768,-30000] and [30000,32767] to make the noise "harsh" (no samples close to 0). But because shorts wrap around the boundaries -32768 and 32767, most samples will in fact be in the range [5537,32767]. See for yourself: if s=5537, he adds 30000, s wraps to -29999, he subtracts 30000, s wraps again back to 5537. s will end up in the ranges [-32768,-30000], [0] and [5537,32767], making the noise less harsh than it should have been.

Damn I would hate to have to re-burn 233 DVDs, or this set of 510(!) he produced later: http://www.jliat.com/HNW510/index.html

This type of bug, an integer overflow, is one of the many types of bugs I look for when I review source code as part of my job in info sec.

[1] http://jliat.com/HNW/HNW.java

[2] http://jliat.com/HNW/HNW90.java

[3] http://jliat.com/HNW/hnwfile.java

> ... I would hate to have to re-burn 233 DVDs...

Burning 233 DVDs is easy if you have the right equipment. At my workplace we used to use units from rimage.com and now use units from microboards.com. The units have built-in inkjet printers so that each disc can have its own label. The software that will let you queue up lots of jobs at once.

(You don't want a disk duplicator, those will typically burn you 10-at-once of the same image.)

Did you notice that there aren't logical and in the if statements' expressions, but bitwise and?

Wouldn't it result in something like `if (true & true) {...}` being evaluated, which seems weird?

I sent this to jliat and he had this to say:

"Oops indeed! but not me! :-)

they have confused THREE! works. The 1 TB set of disks was not made using the Java code! but made from samples of Harsh noise.... as MP3 files = 1 TB - Harsh Noise not HNW! Each MP3 being made of random chunks of MP3s spliced together using a VBnet program. The other piece 510 disks is HNW but again not made using the JAVA code he refers to, and this uses .WAV files.... not .AU !!! (it layers randomly HNW noise to create a .WAV)

Now as to the JAVA code - I wrote this must be 6 - 7 years ago as an exercise- but i've just run the program which creates the .AU file of HNW... max value in the file is +32729 minimum -32720... were found in a small chunk of the .AU file.... so?

here is a sample of the .AU file actually produced by the code to which he refers...

-12248 -30299 5627 -12218 6453 6684 6991 6832 6059 -30395 -30301 -30188 -30270 -12141 -30911 -32020 -32664 32208

Now i'm ready to admit i'm no JAVA expert...!!! which is why i use .Net! But this was an exercise in seeing how to make HNW using Java. It was never released as a piece. I used some code I found for making sine waves and plugged in my random algorithm... nothing smart. But it did work! Why it seems IDK.

As a rule of thumb i tend to look at the output of any programs...

Feel free to post this response as i'm not a member of the group, but would think as the code and program is there a simple test run would show its harsh. That my attempt to not use wimpy numbers doesnt seem to work as it should is besides the point. From memory i probably just tweaked these numbers until the results were harsh - that in the above data set (only a small number of samples above) gives a standard deviation of 29101.09796...

http://www.jliat.com/pics/graph.JPG

If you look at this you can see a plot for the output of the JAVA HNW.... its high! I gave the source so anyone into JAVA could play with this... as i say its not a piece but a by-product on seeing how HNW / Noise has large Standard Deviation and low SNR and as one might expect produces HNW. The harshness being a product of sudden and large changes in values...

In fact i like the fact that in spite of the (MY) mistake it still works - can you fail in noise? :-)

best

James

P.S. The JAVA HNW is way too slow to make these big terrabyte files... it would take much longer to write them than they sound when played!!!"

If you're going to master an original work in a digital format, and it's going to be a terabyte of anything, why MP3 and not FLAC?

Figure at a bitrate of 128 kbps, and a sample rate of 44.1 Khz, an MP3 1 million bytes in size usually clocks in at close to 1 minute of playback. With those statistics in mind, a ballpark figure of 2 years of noise is a reasonable estimate, and that's pretty close to the 711.5 days they cite.

FLAC compression, by comparison, is closer to 10MB per minute, so maybe you'd only get 71 days in the same space, or require 10TB of storage. You could probably fit the lossless version on roughly 400 blu-rays, and opposed to some 233 DVDs?

Even cooler though, is what seems to be some java source for the original noise?

http://www.jliat.com/HNW/

http://www.jliat.com/HNW/HNW.java

Who needs a compressed static copy, when you have the original source?

EDIT: There's also a lossless version in 16 bit WAV audio...

http://www.jliat.com/HNW510/

And if you're making noise for noise sake, MP3 compression would make it less 'noisier'.

Yes - that's sort of what I was wondering.

What effect would the MP3 compression have on the signal's entropy? He says, "MP3 format was chosen not only for its length..." which implies that he managed to compress it at least a bit - if so, there goes his entropy! I can't see how an MP3 compressor could make any headway against a really noisy signal using psycho-acoustic masking, so I'm guessing that any compression he realized came purely from coefficient quantization, which in the case of a really high-entropy signal should actually lower the entropy.

If anyone knows more about this, please feel free to correct me.

I think the psychoacoustic masking would certainly degrade the spectrum... time for a little experiment!

Step 1) Generate 1 second of white noise in audacity, look at spectrum Step 2) export to mp3 Step 3) Import back, look at spectrum

Result: http://i.imgur.com/vZtZ6xE.png (left is original, right is mp3 roundtripped)

The mp3 (at least with this encoder) kills some of the high frequencies, so the noise is no longer white.

Observations like this are part of what turned me off of "conceptual art"; it's too easy to poke holes in any proposed interpretation, and a defender can just as easily weasel out of it once it's pointed out.

I think that good conceptual art is about the interactions among the idea, the work and its participants as much as anything else. That includes people trying to poke holes in an interpretation, and its defenders trying to weasel out of it. There's nothing wrong with that - at its most stripped down, the purpose of a conceptual work is to just be a conversation piece.

Or to put it more succinctly: Oftentimes art's biggest enemy is people making it out to be a bigger deal than it really is.

My conceptual art piece for the day is thinking you're wrong.

And I'm not going to explain, because that would deprive everyone of the opportunity to co-create the work.

Um, lossless compression (FLAC) is not going to handle random noise very well. You're probably better off with the raw WAV file.

If it's lossless, then it'll handle noise the same as the raw wav file. Am I missing something?

FLAC may take more bits than the wav. (It'll still be lossless.)

I'm guessing he meant that FLAC could actually inflate the file size (which is very probable).

Probably, but not by much -- you'd add the size of the data header. Depending on the compression block size, that could be repeated multiply to the file as a whole.

But yes, compressing random values makes little sense. Doing it losslessly increases data size. Doing it lossily decreases randomness.

If the samples don't occupy the full range, then a Huffman table or related encoding could reduce the storage requirements. Ignoring encoding overhead, at most ~12 bits per sample are needed if everything works as the author intended, or ~14.8 bits using the range from https://news.ycombinator.com/item?id=9272071

Well, if they're cutting out "wimpy" numbers, there is a little to be gained from lossless compression.

For the true audiophile looking to appreciate the noise in the exact manner that the artist intended, the distortions introduced into the harmonic balance by MP3 make for a completely different and less organic experience.

I'll sell you some magic beans to put in a bowl in your listening room for $250 that will fix that.

(No shit - someone actually sells those)

Someone actually bought those.

See also: http://wathifi.com

If you're going to author it, why not include just a few quadrillion \x00 characters every so often in a comment field (or whatever the equivalent is for mp3)?

I mean, FLAC proves that people for some reason cherish wasted storage space, especially when that storage space is used for meaningless content that they can't actually hear.

My scheme does that, only bigger.

There are audible sounds that MP3 simply cannot encode correctly. One example is in a song by Planet Funk (see https://news.ycombinator.com/item?id=3166937). However, better lossy formats shouldn't have this problem.

It's also used for archival, so that a few years down the road when a new lossy format du jour (cough m4a) comes along, you're not left with the dilemma of converting one lossy format to another. Not to mention the fact that better quality encoders tend to come along over time.

A rather John Cage-ian anti-music piece although the only thing of interest but the way it was written I wasn't sure if the 4 GB limit "discovered" FAT32's limit or if it was a MP3 format issue that players simply couldn't handle files larger than 4 GB because they were 32 bit.

Other than that, I can't really say that creating a project of 700 days of generated noise is very interesting from an art perspective, but I'll save the art critiques beyond that.

It could become even more art if it included a performance of playing this for 711 days, with someone changing the DVD to the next one every 3 days.

There's a John Cage piece called "Organ^2/As Slow As Possible" that is currently being performed on a specially-made pipe organ in Germany. Note changes happen every few years and the piece will finish in 2640.

https://en.wikipedia.org/wiki/As_Slow_as_Possible

Also Longplayer[1], played on singing bowls, which will end at midnight on the 31st of December, 2999. It's written to constantly evolve and never repeat. You can stream it!

http://longplayer.org/

  I wasn't sure if the 4 GB limit "discovered" FAT32's limit or if it was a MP3 format issue

I think it's based on commodity single-layer DVD's being the storage medium, since the process seems to produce ISO files for burning, which might imply ISO 9660 limitations.

Although, there's MPEG/DVD mastering software in use, which would (probably) be smart enough to fragment assets, and chain them together for continuous playback (considering the availability of dual-layer 8GB recordable DVD's).

The images also note a Thinkpad is in the mix, which might imply Windows XP (or newer), and possible access to NTFS partitions, which permits files of 2^64 bytes.

The Roland MS-1, I remember wanting one of these but having to settle for a second hand Roland S-330.(which I was actually glad to have done) The MS-1 is a quirky sampler- has a simple interface and IIRC is battery operated. It did some data compression internally to bump up the sample time and stored the samples in flash memory. This is not really a "hi-fi" sampler, and depending on what settings were used, I would argue that its characteristics would smooth out harshness rather than enhance them. I think you could get around a minute of sample time if it was set to its lowest sample rate. Not bad for the 90s...

I was going to try something similar, something like a MIDIquest 'random parameter' generator running into a few analog/virtual analog synths (MS2k, SY35, mopho, whatever else i see on Craigslist). Probably the most straightforward way to generate "harsh" noise (there's blue and violet noise, emphasizing above 1kHz, according to Brian Shepherd's synthesizer textbook) is to run random parameters into a Yamaha DX7.

_______

reminds me: need to read Vertical Color of Sound: http://www.amazon.com/Brian-Eno-Music-Vertical-Color/dp/0306...

also reminds me: someday I'm going to listen to Lou Reed's Metal Machine Music

In a far distant future an interstellar scientific exploration vessel will approach the lifeless remains of the solar system and discover, stashed away in a big artificial moon, one very huge data archive consisting entirely of harsh noise. They will be so happy.

It's like a child's play: the result is utterly uninteresting, but the process is fun.
Adults just have more money and skill to put into activities like that.

Where can I get one? Will I get a guarantee that my set would be unique and impossible to recreate?

Is this an art project? What is the use of this data? Is there a sample we can listen to?

after a while: if it is music, how would it age? would it be considered old-school after 10 or 20 years? if so, would recording a new set be a remake or a new one?

Another data source for /dev/random?

/dev/random will take forever; too slow.

I believe asn0 was suggesting this as a seed / input to /dev/random, not an output.
The lossy compression (and apparent algorithmic errors) make this a fairly poor choice of random noise input.

OK - some background...
Reading the thread is interesting! My motivation for the 1 Terabyte work was nothing to do with conceptual art, I wont go into detail but originally conceptual art was the idea that art was essentially a practice of examination of the nature of art, see
http://tallervi.pbworks.com/f/Art%20After%20Philosophy.pdf
This is an incredibly important document (though I don’t subscribe to it now) as it poses the severing of art from aesthetics! (which I do subscribe to)
Unfortunately the media then took the term ‘conceptual art’ to mean any art using any ‘idea’ about anything! Which was usually linked with challenging the viewers idea of art etc. or being ‘interesting’ or smart or ‘pushing the boundaries’… And was based on the misunderstanding of the lack of symmetry in this kind of thing..
1. New art is shocking. 2. Something shocking is new art.
1. might be true in some cases, but 2 does not follow anymore than…
1. New theories in science are often thought crazy. 2. If this is thought crazy its new science.
So my motivation for 1TB– simple
“A man climbs a mountain because it's there. A man makes a work of art because it isn't there.” Carl Andre.
So not a conceptrual work but an OBJECT.
I just thought - could I make a terabyte of noise? I like the discussion about why mp3? The project began many years ago 1990s… using a lap top and VB to write the files. This was so slow that after about 2 days the program would crash… the problem seems not FAT32 but the old VB couldn’t handle files larger than 4gb. Well after several attempts the lap top gave up! Latter of course you see the header of a .WAV file allows for a max of 4gb. (“WAV files are usually stored uncompressed, which means that they can get quite large, but they cannot exceed 4 gigabytes due to the fact that the file size header field is a 32-bit unsigned integer”) So I had two problems, size, and the time taken to write! such a long file. The answer was when I found you can join mp3s and they will play… even using a DOS copy command… copy *.mp3 /B big.mp3 would copy all the mp3s in a folder to one big mp3 – which would then play! (unlike mp4!)
So I created a .vb program (called MUZILLA....) to randomly splice chunks of .mp3s to create large mp3s. The random selection giving different sound files of very large lengths. This would make very big mp3s in minutes not days…. So the project looked possible!
Moreover I found though one can count up to a billion, a trillion is impossible, I am very interested in ‘objects’ which exceed the human. http://www.jliat.com/tg/index.html
http://www.jliat.com/APCDS/index.html
This is bound up with such ideas as infinity, multiverses and a time before and after humans as being the nature of reality. (Speculative Realism) If reality is essentially inhuman – and if art is about our relationship to reality… then I think my work could be considered as being art. Or if not then I’m just messing with what I find in the world…

Reply on news.ycombinator.com