Show more

Twitch have finally stopped fucking their timestamps up on low-viewership streams.
3 months of having to reset a player every 17 seconds to 20+ minutes.
I feel sorry for the person with a brain tumor who thought it would be an excellent idea to do that to stop people viewing ad-free streams with a player that can't play their ads.
And I also feel sorry for the one good person I know who unfortunately works at Twitch.

log(x) is nice and common in audio/video encoding but explodes when 0 and is mostly uselessly flat except for a narrow region.
(const1) / (x + x + x + const2) * sqrt(const3 + x * x) provides a neat curve that's better and less flat than a log() for most cases, can be easily tweaked and just has a very minor maximum near 0 (also tweakable).

"That weird Terry Gilliam movie"
Which one, all Terry Gilliam movies are weird? No exceptions.

Just finished helping someone write a Siren (G723.1) decoder for ffmpeg.
Weird codec, but weirdest part is the authors messed up their MDCT so all their imaginaries have a wrong sign. Took a while to figure out. Unsurprising, they were so desperate to use a 320 sample MDCT they based theirs on a DCT-IV rather than the standard FFT.
Unfortunately the only Siren files you can find outside of VoIP streams are VIVO files from 1997. They seem to capture the era pretty well.

Pretty much everyone who does open source multimedia gets various job or contractor offers via email.
I've never received one and its something to brag about. How much of a punk do you need to be to get ignored by companies like Google, Facebook and Netflix that need multimedia people.
(btw Google prefer to hire fresh PhD students who know 0 about multimedia and have to be spoonfed basics like PTS, DTS and end up doing disgusting hacks that don't get accepted by upstream because they're crap)

Picture one represents the full spectrum of a 37 second DSD128 file converted to PCM with a samplerate of 705600 Hz. Average bitrate for that file is 11.5 Mbps.
Picture two represents a more detailed view of the same file's spectrum in the audible domain.
What we observe is a sharp cutoff at 24Khz, suggesting a typical 48Khz master was used to create these files, with a typical bitrate of ~2.3Mbps. With the process not being lossless, some amount of resampling error is present.
DSD is very shit.

I like chocolate so you'd think I'd be happy in this discount paradise. I'm not. I had a single milka and now I don't even want to smell one.
I should have just gone to a viet place and ordered a soup. Or a burger.

Pushed integer point FFT support for libavutil. Random comparison between the 3-point transform:
libavutil: 4 multiplies, 14 additions
kiss_fft: 16 multiplies, 12 additions Btw the official translation of SAO Progressive's 6th volume is metricized! I still hate Yenpress for ever forcing a translator to localize and butcher the units just for this series, so I'm still never buying anything they do.
The cubic inches part of the first volume is so weirdly worded I wonder how it ever passed QC.

I've seen quite a few instances that mention "Content illegal in Germany not allowed".
I know they don't really have a choice if they have German users or are hosted in Germany. But it sucks how the German government is legally allowed to stifle any discussion it doesn't like, with a very heavy hand.
Read the Wiki page of "Censorship of Facebook". Makes Germany look just like China with respect to censorship.

Wow, found a pirate mp4 release of an Australian movie that was muxed with L-SMASH.
Haven't seen a GPAC release yet so that's a definite proof at least most release groups are aware GPAC is cultist commercial garbage.

VMAF is considered by many to be the best metric to measure video compression by.
I knew it was somewhat random with respect to inputs. Now I know how awful it is.
Some months ago people figured out they can cheat it by just sharpening the video before compression. This doesn't mean the video looks better though.
Of course, libaom implemented a sharpening filter as a preprocessor to increase its vmaf score by around 30%.
Just don't use VMAF alone, ever. Provide PSNR, PSNR-HVS and SSIM as well.

Being on Google, Twitter or any other website like that is kind of like real life.
You have to watch out how long you stare/put your cursor over something, and don't talk to anyone/watch a video/click a link unless you really have to.

One of the most tedious things in life is converting unrolled, unmacro'd FFT code to use butterfly and complex multiplication macros. Because suppose you ever want to integerize it and don't want to maintain 2 transform codebases.
This must be exactly what threading core rope ROMs would have been like.
Converting 16 mere lines of complex multiplies to use macros for hours only to make everything undelectably inaccurate except for one corner case you didn't test for.

全自動少女 - モダンガアル街を往く。
Made of a few members of the <echo> project circle they released one great album in 2012, then failed to get a seat at an event IIRC and released one already-made song on their soundcloud before splitting.

Now its been merged, how to crash an Intel GPU on Linux with one line:
ffmpeg -init_hw_device "vulkan=vk:0" -f lavfi -i testsrc2 -filter_hw_device vk -vf hwupload,avgblur_vulkan -f null -

There's a slim but existing chance the Olympics in Tokyo will be relocated because of the epidemic.
ffffffuck yes. Cities don't escape unscathed after olympcis.

Just pushed a 7k line FFmpeg patchset adding Vulkan filtering and mapping.
And I'm so not looking forward to it being irreparably broken by the future Vulkan video decode/encode API.

Blocking incoming IPv6 traffic: occasionally necessary if dealing with insecure devices.
Blocking incoming and outgoing IPv6 while still giving stateful DHCPv6: for what purpose?
Seen at a Mozilla office.

Show more

A Mastodon instance for people interested in multimedia, codecs, assembly, SIMD, and the occasional weeb stuff.