Show newer
Lynne boosted

Vote now, and maybe if I feel better one day I'll reveal the song.

Yet every single low budget director's-first-time-directing-a-movie is a cookie cutter typical horror movie that's been done and done again since 1975!

It's not like alternatives aren't popular. The analogue to the typical cartoon shooters are series like the Resident Evil films, which does its hardest to entertain or the Evil Dead series, psychological movies are popular as well with the likes of The Conjuring, and fusions exist too, I'm sure (nothing comes to mind right now though).

But scriptwriters can only come up with Friday-the-13+i movies. Even Friday the 13th movies have outgrown their slasher origins, hell, they had Freddy Kruger in space at one point battling Jason (I think)! That's innovation!

Show thread

Even video games have outgrown the "horror must be violent, with torture, and blood! yeah!"-phase, now they diversify into typical cartoony shooters with lots of gore and jump scares, like Dead Space, or more psychological stealth-orientated games, like Silent Hill, Amnesia and fusions, like Alien: Isolation.

Show thread

It's hard being a fan of horror movies, but only ones that contain no torture or gore.
It's not that hard to make a movie like that, yet it's like both of those are necessary to make a horror movie. Or at least, knowing Americans, a horror movies Americans can enjoy.

These days, there's just The Conjuring, which is a bit cartoon-y, but it works in it's favor. As for anything else, IMDB, despite having its origins from the UK, does not even list whether movies contain torture and gore in an accurate way, so I have to treat every new horror movie as definitely not good unless I hear otherwise.

Unfortunately, or fortunately, depending on how you look at it, Opus has _no_ spare bits at all. Every random collection of bits is, after all, a valid Opus packet.
So it would have to be a separate RFC. And therefore a brand new codec, with new identifiers, and new container-level specs. And I dislike dealing with the IETF, since it takes YEARS to get something from a draft to an RFC.

But the biggest improvement by far would be proper multi-channel encoding. Opus really fucked up by using Vorbis' channel layout, which makes everyone use AAC instead nowadays.

Show thread

If there's ever an Opus 2 WG, and it's not based on neural networks except for low bitrates, I think I'd very much like to be the person in charge of the Celt layer.
I have some insane but practical idea about using an N-dimensional Spherical Harmonics Transform, plus porting over Daala's PVQ predictor. I _may_ tolerate a neural network entropy encoder, but honestly, the output from the SHT should be pretty Laplacian looking. That was also a pun no one will get.

I'm not big into cosplay at all, but a YoRHa Type A No. 2 cosplay to me has such a low effort vs flashiness bar, I think it's possible.
Though you will end up with a fairly unimpressive cosplay, because impressiveness goes disproportionate to effort vs flashiness.
The Ohm law of cosplay.

Show thread

Ethernet is a 100% genuine sci-fi communication system name if it wasn't already real.

You Can Write FORTRAN in any Language ᵉˣᶜᵉᵖᵗ ʳᵘˢᵗ ᵇᵉᶜᵃᵘˢᵉ ʸᵒᵘ'ˡˡ ᵇᵉ ᵏᶦˡˡᵉᵈ.

I so want an SC-D70 (because I like digital outputs) or an SC-8850 and the space to put it.

A few years ago I was almost going to buy a cheap SC-88 but I thankfully decided against it at the last minute.

Show thread

Yuru Yuri's soundtrack is to Kerbal Space Program as Yagate Kimi ni Naru's soundtrack is to Civilization V.

It's the soundtrack you get when the default one is boring, and you want to forever annoyingly associate two completely different subjects.

Google's Lyra codec is just a bruteforce "NEURAL NETWORK GOES BRRR" solution for speech encoding. "Just stick a neural network and let it handle everything."

Neural networks don't belong in a sound generator. They belong as a predictor to make a filter reproduce speech. In other words, they belong as neural networks hooked up to fake vocal cords.

LPCNet is to me much better, and a way more elegant solution. But Google hate any research that isn't Google, and with their compute capacity, a bruteforce solution is just so easy to do. No research is even needed, programmers can just excuse themselves from work with infinite amounts of "the NN is being trained".

jmvalin.ca/demo/lpcnet_codec/

The two types of sin:
negative_sin and !negative_sin

"Is there a horror element in this game?"
"Yeah, talking to people."
- Draqu_ playing Lumberjack's Dynasty

Now that I wrote the power-of-two FFT assembly, I don't know what to do. It's like finishing a long-going show.
I don't think I'll ever write anything nearly that complicated or dynamic in assembly. The way the bottom-up FFT came together without really knowing all that well if it will.
I guess I still have to write an ARM64 version, but ARM64 is no fun. And really, dogfooding is a huge huge factor for me when writing anything that it's going to be a downer writing for an ARM machine I don't actually use.
Also ARM has a really shitty vector ISA. So much so that I'm more motivated to write a RISC-V version for an extension not even standardized yet.

Show older
Parsee

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