Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[ChairmanPoo]

I've learned new things about potatoes. Apparently if you cut up an uncooked potato, stick it in the fridge for a couple days, and then leave it out for a while it will blacken. Like, we've literally never wasted a potato like this my life so I've never seen it happen before. It's mildly terrifying, hahaha!

If you stick it in the freezer and then shove it up it also prevents coronavirus infection, it is known.


But seriously, the potato thing is normal oxidation. You dont even have to put it in the fridge, and, indeed, it can happen while in the fridge.

If you want to preserve a potato (for coronavirus prophylaxis or otherwise) put it in water before putting it in the fridge. That way it wont blacken.

[Frumple]

Yeah, we were testing a cutting thing on a potato we didn't care enough about. Normal preservation I'm aware of, heh. Family's ate plenty of potatoes over the years, it's just vanishingly rare they sit uncooked for long enough for that to happen once they've been chopped up or whatev'. Cause, y'know, normally we just eat them.

[Bumber]

If you want to preserve a potato (for coronavirus prophylaxis or otherwise) put it in water before putting it in the fridge. That way it wont blacken.

How about putting it in delicious butter, instead?

[methylatedspirit]

I think I've discovered some kind of glitch in Audacity, where audio gets shifted ahead by a couple of samples each time the "Low-Pass Filter" effect is applied. Look:

Spoiler (click to show/hide)

Each marker represents the start of the beat after the effect is applied so-and-so times.

Granted, if you're using Audacity the way it was intended, this glitch doesn't matter. At a sample rate of 44.1 KHz, 1 sample is ~0.00002 seconds of audio. But if you're like me and use it to databend images, it can get slightly annoying, as the shift essentially causes the image to be shifted to the right, which causes the right side of the image to wrap around to the left.

Spoiler: Demonstration (click to show/hide)

(original images are planar RAW files. However, since PNG is lossless, these images are identical to the originals)
This is a color gradient.
After low-pass filtering the image in Audacity to 1000 Hz, this is what I get:

Notice how the image is quite literally shifted to the right, and what was on the right edge ends up on the left edge, hence the sudden 'burst' of brightness and color at the left edge.

(I checked, there is actually a databending/glitch art thread, but I'm just too WTFed by this thing to care)

[Reelya]

That's not shifting the image, it's blur. It's treating the data as a single row of data. So it's bleeding over data from one row to the next.

It's just extra noticeable because the dark area contrasts so much with the light area. Notice that the banding on individual segments hasn't actually moved very much at all, much less than the apparent bleed-over from one row to the next. Grab a ruler, and line it up with one of the inner color segments and halfway down the bottom image. You'll notice that where the line is strongest is in the exactly same place, and not shifted to the right as the left-hand color bleed over would imply.

Same thing with the sound I suspect. What low-pass filters do is:

Low-pass filters provide a smoother form of a signal, removing the short-term fluctuations and leaving the longer-term trend.

The effect could be possibly because beats tend to have a fast attack and a slower decay. If you shift that to a lower frequency, then the peak may shift to the right.

[methylatedspirit]

I'm somewhat doubtful about that. I put the green band of the original and the filtered one side-by-side, and the banding is slightly misaligned.

Spoiler (click to show/hide)

I should probably try a harsher transition, like from #000000 to #00FF00 or something.

Spoiler (click to show/hide)

Same thing? But given the level of blurring, I can't really confirm it.

Edit: I just thought about it, and I've decided to a) halve the resolution and b) make the final segment of each row black, #000000. This doubles the shift, and makes it impossible that it was blur. I filtered it, as usual, then I resized it back to 850x600.

Spoiler (click to show/hide)

Spoiler: Green band only (click to show/hide)

It's really hard to tell if it's shifted forwards, but there is still a tiny amount. How much of that is from actual shifting and how much is just blurring, I can't tell.

[ArchAIngel]

My daily WTF is the French response to Corona.


Have 3,500 people dress up as Smurfs and go gather up to set the world record.


No. Really. This happened.


https://slate.com/news-and-politics/2020/03/french-coronavirus-smurf-world-record.html

[Bumber]

Maybe they're planning to surrender to the coronavirus.

(Not sorry.)

[Reelya]

...

Those are still both blur. Put a ruler down some of of the green bands more towards the left in the top image and the area of greyness in the bottom part is right in the middle of the ruler. It's just a perception issue since you perceive the band as starting where the color is solid. But that's not actually the center of the blur region.
 
Also, notice that in the bottom "Green band only" image, the gradient in the right-most panel starts exactly where the line is. It's just more clear when going from light to dark than dark to light.

What I think is happening here is that an audio filter always works from left to right. i.e. signal to the right hasn't happened yet, so there's right-blur but not left-blur. So you're seeing signal bleed over from the previous samples to later samples, but not vice versa. A simulation of a real-world audio filter would be expected to work like that.

[wierd]

The issue is with the arbitrary boarder assignment.

The audio processing codec has no concept of "image width".  It is all one continuous data stream. As such, the data at the edge of the "image" wraps around to the beginning of the next line in the image. This creates the appearance of the offset.

To prevent this, your processor would need to be aware of where the image's constraining boundary is per each line, so that it can treat the data in an appropriately special condition when processing near that boundary.

[methylatedspirit]

What I think is happening here is that an audio filter always works from left to right. i.e. signal to the right hasn't happened yet, so there's right-blur but not left-blur. So you're seeing signal bleed over from the previous samples to later samples, but not vice versa. A simulation of a real-world audio filter would be expected to work like that.

The audio processing codec has no concept of "image width".  It is all one continuous data stream. As such, the data at the edge of the "image" wraps around to the beginning of the next line in the image. This creates the appearance of the offset.

To prevent this, your processor would need to be aware of where the image's constraining boundary is per each line, so that it can treat the data in an appropriately special condition when processing near that boundary.

So if I get this right, the filter right-blurs images so hard, that it makes images appear to shift to the right? And, because Audacity only sees a continuous data stream, rather than a series of rows of pixels making up an image, it doesn't compensate for this?

[wierd]

Yes. Pretty much that.  The data at the rightmost edge is linearly adjacent to the data on the next row, at the far left.  The blur effect takes that data, and blurs it "left", via that lack of cutoff.

[methylatedspirit]

So if there were some uncompressed image format (compressed formats, lossless or not, generally don't like being messed with) that stored its images as a series of columns of pixels (top-to-bottom, then left-to-right), would applying the filter cause it to be shifted down?

[wierd]

No, it would cause a strange vertical banding pattern.


You can calculate the exact "timestamp" range for the implicated row though, and batch process.  I think audacity supports scripted invocation? If so, that is how I would accomodate the missing understanding of line-ends.  (Select the audio data one "line" worth of time, at a time-- run the lowpass for each selection. Rinse, repeat, until the whole file is processed.)

Ok Yes-- totally--

Abuse the scripting engine to incrementally process selections.
https://manual.audacityteam.org/man/scripting_reference.html

This will give the hard cut-off you want at the end of image lines, and prevent "next line" bleed.

[methylatedspirit]

No, it would cause a strange vertical banding pattern.

Woah, I gotta try this myself. How do I pull this off?

Abuse the scripting engine to incrementally process selections.
https://manual.audacityteam.org/man/scripting_reference.html

This will give the hard cut-off you want at the end of image lines, and prevent "next line" bleed.

See, that's just using Audacity as a poor man's image editor. I'm pretty sure you'd get roughly the same effect by going into any decent image editor, selecting the radial/motion blur effect and applying that. Still, I find the idea oddly funny. What next, using Photoshop to edit audio?