HN Theater @HNTheaterMonth

The experiment has also been brought to the 21st century and its video and audio codecs: https://www.youtube.com/watch?v=icruGcSsPp0 - uploading a video 1000 times to YouTube.

⬐ iamacyborg

There’s also a fun version of this but it’s a song run through google translate.

https://m.youtube.com/watch?v=LMkJuDVJdTw

⬐ dylan604

reminds me of the old days of 4 track recording where you'd bounce 3 channels down to one, rerecord, bounce again. lather rinse repeat. by the time you got to the 3 or so round (depending on the quality of the tape) it would be pretty gnarly.

back when i worked in a VHS dubbing facility, we'd get bored and do stupid stuff like this taking a master to VHS, then using the VHS as a master to make dub of dub of dub. after the 8th dub, it was pretty trash. Similarly, we'd take a video in -> video out to another VCR -> video out cascading down until the original signal to the video in of the last VCR was trash. (the place was wired correctly with proper video DAs. we did this because it was 3rd shift and no supervision)

⬐ mrloop

This brings back memories of art school, a friend recorded themselves running up and down a hill and made many copies of copies of the video on VHS. They displayed different generations of the degrading video side by side on a bank of monitors.

If you don't have 15 minutes to waste on exposition, this is the first video that did this, and it's only 3 minutes total: https://www.youtube.com/watch?v=icruGcSsPp0

I remember back in like 2010, someone uploaded and ripped a YouTube video 1000 times and then edited the natural degradation of the compression. Went from perfectly watchable video to horrible static-y mess.

Here it is! https://www.youtube.com/watch?v=icruGcSsPp0

⬐ citrusui

Thanks for the link!

This may sound crazy, but I wonder if there's a way to "automate" this deconstruction with ffmpeg...

⬐ citrusui

Update: here's a one-liner that will do exactly that

ffmpeg -i input.mp4 -strict -2 -crf 51 output.mp4 && rm input.mp4 && mv output.mp4 input.mp4

For better results, run the above command ten times:

for ((n=0;n<10;n++)); do ffmpeg -i input.mp4 -strict -2 -crf 51 output.mp4 && rm input.mp4 && mv output.mp4 input.mp4; done

⬐ rothron

Theoretically, it should be possible to recompress the video back into the compressed file with no loss, provided the same codec is used, but it would be far more expensive computationally.

⬐ pbhjpbhj

Aren't most image/video compression algorithms lossy? There would surely be more than one possible "decompression".

⬐ bringtheaction

Here’s another one up the same alley that is interesting to watch: VHS generation loss.

> What happens if you make a copy of a copy of a copy (and so on) of a VHS tape? This experiment shows how the quality degrades with every generation.

> The copying was done using two PAL VCRs in SP mode.

> (The video is Fading like a flower by Roxette)

https://youtu.be/mES3CHEnVyI

⬐ Trisell

Note to self. Upload my ransom request videos 75 times to YouTube before sending it. That way I will have an awesome sinister bad guy voice.

Edit: Engrish is hard

⬐ justsaysmthng

Isn't kind of a similar thing happening to DNA from generation to generation ? Codec artifacts being similar to gene mutations in DNA ?

⬐ jakobegger

Nope. DNA copying is surprisingly reliable, and there are several mechanisms that prevent this kind of degradation.

Also, as soon as a serious bad (deleterious) mutation happens, the resulting individual won't be able to reproduce further -- this selection process ultimately prevents this from happening.

⬐ foolrush

Also see David Rimmer. He pioneered several experimental film works, including optically printing film repeatedly in a very similar fashion to this.

http://mikehoolboom.com/?p=78

⬐ chris_wot

If you can bare to listen to Gangnam Style, this is pretty interesting:

https://www.youtube.com/watch?v=ywWJH_BvBDs

⬐ DanBC

(With thanks to Detaro for pointing out my rookie error https://news.ycombinator.com/item?id=11230051 )

⬐ None

None

⬐ dang

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

⬐ ChrisGranger

It would be great if there were an app that could simplify the process of recording your own versions. (Is there such an app?) It would be interesting to try this with songs and other sounds instead of just a speaking voice as well.

⬐ imaginenore

You don't have to do it through youtube. Just use FFMpeg. At 10 min per encoding you can be done in 7 days, completely automatically.

⬐ Intermernet

If you want to emulate this purely in software there are convolution reverbs[1] available for both Pure Data[2] and Csound[3] that you could probably re-purpose / script to do tens or hundreds of iterations on any input material you wanted.

Note: the linked examples are both based on "partitioned convolution" algorithms[4], which allow lower latency than standard convolution algorithms, and would probably be a better choice for running many iterations in a reasonable amount of time.

I may play around with this when I next have some spare time.

[1]: https://en.wikipedia.org/wiki/Convolution_reverb

[2]: http://puredata.info/Members/bensaylor/partconv~-0.1.tar.gz/...

[3]: http://www.csounds.com/manual/html/pconvolve.html

[4]: http://cnmat.berkeley.edu/system/files/attachments/main.pdf

⬐ don_loemax

Also, if you are using OS X, there's a spectral processing program called SoundHack that let's you convolve (along with several other processes) any two sound files. I am not sure where to grab the latest version, I believe development has been abandoned but it was a stand-alone freeware that one of my '" Intro to DSP" courses used to introduce certain functions.

⬐ DanBC

That's amazing! Thanks dang.

⬐ joshschreuder

A performance of the original: http://youtu.be/fAxHlLK3Oyk