Compressing a music video into a floppy, is it possible?

[Chainmax]

I imagine that XviD won't do at such tremendously low bitrates, that's why I'm posting this here. What audio and video codecs would you guys choose for this task?

[Tommy Carrot]

The best audio codec for ultra low bitrate would be Speex ACM imo (decent quality at 20 kbit /stereo!/). For video? Maybe RV9. But this is only theoretically, because you cannot use both codec with each other. :D

[bilu]

1440 KB = 11520 Kb

500 kbps = 23 s
400 kbps = 28.8 s
300 kbps = 38.4 s
200 kbps = 57.6 s

Even if tempted by this thread
http://forum.doom9.org/showthread.php?threadid=64789

to point RV9, I'm not sure if you would call 30 seconds or less a music clip... :rolleyes:


Bilu

[BoNz1]

Hmm, speex might have a tough time with music. I remember a while ago garf at HA.org was working on some a modified ogg vorbis encoder which sounded rather impressive at 4kbps I think it was called floggy. The goal I think was to put an entire cd onto a floppy :). I think the floggy stuff is in GT2 and GT3 encoders. For the video, xvid might be hard pressed at that bitrate but you might try RV9 and put it all into matroska I guess.
EDIT: You would have to of course drop the res a ton like postage stamp sized, :)

[Tommy Carrot]

Lol, i was able to squeeze a 4 min. long clip into 1.4M. :D

Audio settings: 32khz, stereo, speex q5 (19.4 kbit)

Video settings: res=192*144, WMV9 codec at quality level 30, resulting 23 kbit!!!!

It's horrible ofcourse. :D

BTW, speex is not so bad, considering the low bitrate. The larger problem is the video part. Perhaps RV9 does better, but 23 kbit/s is awfully low.

I remember Garf's experiments too, but the encoder was not released afaik.

EDIT: Is it in GT3? I didn't know that.

[Ramirez]

I don’t think so, not entirely anyway, the lowest possible bitrate I've manage to squeeze out of GT3 was (3:03mins stereo) 12Kbps: resampled: 8 kHz. ()

[Atamido]

Originally posted by Tommy Carrot
Audio settings: 32khz, stereo, speex q5 (19.4 kbit)

Video settings: res=192*144, WMV9 codec at quality level 30, resulting 23 kbit!!!! You don't see many encodes where the audio is almost the same size as the video. :p

[BoNz1]

The floggy hacks were included in GT3b1. Commandline is:
oggencgt3b1 --resample 6000 -q -2 file.wav
EDIT: I just encoded some tunes on my hd and got about 5.4kbps and it is ok all things considered, ;)

[RadicalEd]

Zarxrax (http://forum.doom9.org/member.php?s=&action=getinfo&userid=3007) compressed his Doujin Addiction AMV (http://www.animemusicvideos.org/members/members_videoinfo.php?v=14832) to a beautiful 323 KB using 10 kbps 64x48 15fps RV9 and 6 kbps mono 8khz RA8 (cook-8).
Avatar-sized video (http://digitalinsanity.ath.cx/~tab/doujin.rm).

[Blueseb]

Give a try to vp6

[slavickas]

Originally posted by Blueseb
Give a try to vp6

take alook at http://vejas.pit.ktu.lt/~slavvita/ogmsamples/minima/

but actually everything depends on content, live concert (with lights etc) would be impossible

[Atamido]

Fun with videos. (http://www.jkfanclub.com:81/3_videos.mkv)

[Blueseb]

Originally posted by Pamel
Fun with videos. (http://www.jkfanclub.com:81/3_videos.mkv)

slogan proposal:
welcome into "Matroska Power" era :D :D :D

[Ramirez]

Thanks for the hint bonz1, I've tested it and it does works as you suggested (I assumed that GT3B2 also includes this feature, apparently not)

[r6d2]

Originally posted by Chainmax
Compressing a music video into a floppy, is it possible?
Do you mean a 3 1/2" floppy or a 130-KB Atari 5 1/4" floppy?

:D:D:D

P.S.: Sorry, could not resist ;)

[General Lee D. Mented]

Originally posted by r6d2
Do you mean a 3 1/2" floppy or a 130-KB Atari 5 1/4" floppy?

:D:D:D

P.S.: Sorry, could not resist ;)

Nah he means a 256K 8" floppy. :P

[r6d2]

Originally posted by General Lee D. Mented
Nah he means a 256K 8" floppy. :P
Oh, I get it. The problem with those is that they don't fit in some standalone players. :P

[The Collector]

Well, first I would format the floppy in DMF2 (1.68MB) and can be read by every windows version and even DOS. Writing is possible with most OS's as well.
http://www.alkonost.com/maxformat/

Then I'd try HE-AAC @48kbit/s (44.1khz,stereo) or maybe @32kbit/s (32khz, stereo) VBR offcourse. Altough I didn't compare the Speex-codec, this one sounded very good when I tested it a while back.

For the video, well, probebly not XVID/DIVX, maybe RV9 (altough I never tested it), but RV supports variable FPS as far as I remember. WM9 is also a good candidate.
When a good one would be available, a H.264 video-codec would be the best choise offcourse!

Then we have the container, AVI is out of the question, it has too much overhead, I guess Matroska is a good one for the job.

[Tommy Carrot]

Originally posted by The Collector
Well, first I would format the floppy in DMF2 (1.68MB) and can be read by every windows version and even DOS. Writing is possible with most OS's as well.
http://www.alkonost.com/maxformat/

Then I'd try HE-AAC @48kbit/s (44.1khz,stereo) or maybe @32kbit/s (32khz, stereo) VBR offcourse. Altough I didn't compare the Speex-codec, this one sounded very good when I tested it a while back.

For the video, well, probebly not XVID/DIVX, maybe RV9 (altough I never tested it), but RV supports variable FPS as far as I remember. WM9 is also a good candidate.
When a good one would be available, a H.264 video-codec would be the best choise offcourse!

Then we have the container, AVI is out of the question, it has too much overhead, I guess Matroska is a good one for the job.

That's right, h.264 is far the best. Hdot264 codec with bframes and RDO enabled is far superior to wmv9 or VP6, and better than RV9 too. I was able to get quite decent quality at 23 kbps, while with the other codecs the image simply fell apart. Too bad it's slow like hell. 3100 frames with 192*144 res took about 45 min on my athlon1700.:eek: VSS_h264 codec is faster, but gives far worse results than Hdot264. It would be so great if a decent implementation would be available from h.264.

[Sirber]

H264 better? Couls you do some PNSR comp with hdot264?

[Tommy Carrot]

Originally posted by Sirber
H264 better? Couls you do some PNSR comp with hdot264?

Well, i don't like to use psnr to evaluate codecs, but ok, here are the average psnr numbers:


WMV9: 34.4516 /slowest setting/
VP6 : 34.9629 /slowest setting/
Hdot: 35.2540 /bframes and RDO on/


I couldn't get a psnr result for RV9, it gave back strange numbers. Quality-wise it's close to h.264, but a little bit blurrier.

[Sirber]

what was the bitrate again? 23kbps?

[Tommy Carrot]

Originally posted by Sirber
what was the bitrate again? 23kbps?

Around that.

WMV9=21
H264=23
VP6=26

The filesize of RV9 is about the same as h264, but i don't know the correct bitrate.

BTW, here is a new contender with surprisingly good result.

Xvid dev-api4: 34.4805 /at 23 kbps/

Used the following settings: h.263 matrix, chroma motion, vhq mode 4.

[Sirber]

Can you post all links for us to see? :D

[Tommy Carrot]

Originally posted by Sirber
Can you post all links for us to see? :D

Be patient, i'll upload them somewhere tomorrow.

[Tommy Carrot]

Allright, here (http://removed, sorry) they are, if you're still interested. :)

[Chainmax]

AFAIK, floggy wasn't implemented in OggGT3b1, but you could somewhat emulate it. At least that's what I remember from reading Garf's page (http://users.pandora.be/sjeng/floggy.html) and some HA threads.

[Prettz]

The short answer to your question is no. Given the length of even the shortest music video (and even ignoring the fact that music videos have the fastest and most frequent cuts of any kind of video you're going to find), there's simply too much "unique" video information.

Maybe you could if you find and extremely short, simple, video with very few cuts, heavily subsample the video (even more than YV12), reduce the resolution down to maybe a dozen or two pixels wide, and compress the audio down to a cell phone ring tone...

[RadicalEd]

Originally posted by Prettz
The short answer to your question is no.

It's been done.
See this thread (http://forum.doom9.org/showthread.php?s=&threadid=64883). ':|

[Prettz]

Originally posted by RadicalEd
It's been done.
See this thread (http://forum.doom9.org/showthread.php?s=&threadid=64883). ':|
Unfortunately I can't view the videos at the moment so I can't comment on anyting about them. Basically in my other post I was getting at the idea that you're not going to be able compress a typical music video to that size unless you really really skimp on the resolution. I'm thinking that in a real music video there's going to be just too many scenechanges.

On the other hand, the link to those low-bitrate Ogg 1.0 songs is nothing short of mindblowing. I'm still having trouble believing that the Sultans of Swat was at 4kbps and 6KHz. So I clearly stand corrected on the question of audio quality.

[Chainmax]

Sirber, could you try to do it?

[vinouz]

I just looked at Tommy Carrot's encodes (but hdot264 one) without re-reading the psnr results (and I had forgotten it). To my eyes, I prefer the rv9 encode, followed by the wmv9 one, which is 103kb less (which means approximately 13% less in filesize. at this bitrate, it counts much I think) and the xvid-devapi4 one. The VP6 one could be nice, but its quality jumps make it much harm...
What I prefer in the rv9 encode is that it gives the impression to be fluider than others, and makes me less remember it's digital encoded video (not much macroblocking, although not as blurry as wmv9). I think that's a strong point when targeting the average spectator's eye.

but the comparison to wmv9 is nonvalid given this size difference.

[Tommy Carrot]

Vinouz, this is true, wmv9 is undersized, but i've already made the psnr-tests when i realised it. Anyway, here (removed) is the fixed version. A little bit oversized, but close enough imo.