programming

[nocturne olive]

Me: LMMS

[merry plank]

audacity is always fine too

[nocturne olive]

Oh woah I found something that predates both new tuning AND new mixing

[merry plank]

anyways so happy I can finally do an I'm Yours cover by evil neuro

[nocturne olive]

Well since it's RVC it'll probably be pretty bad

[merry plank]

yeah...

[nocturne olive]

EvilSynth would be better

And then of course there's this which is peak showcase of NeuroSynth as an instrument

And the one good commissioned song

[nocturne olive]

Audacity sucks for mixing

As soon as I switched to LMMS basically my mixing quality got way better

[merry plank]

I mean all I have to mix is the vocals with instrumentals does it really matter

[nocturne olive]

This is an even older thing, I think from transition from bad mixing to good mixing

This is essentially what I consider the start of "new mixing"

[merry plank]

omg rvc sounds so bad

😭

one sec

[nocturne olive]

Yeah without a vocal synthezizer or a very specific kind of base vocal it sucks

This is one of the final things from pre-NeuroSynth era

Then there's the couple times I messed with non-Neuro synthesizers

[merry plank]

what I did

[nocturne olive]

Put pitch up by 12 semitones at first

[merry plank]

should have got some kinda reverb remover too

[nocturne olive]

But also the base voice just sucks

[merry plank]

but yeah

[merry plank]

yeah male voice

[nocturne olive]

I would do better

[rigid snow]

just pitch up an octave

[merry plank]

this?

[nocturne olive]

Though of course I have a vocal synthesizer for Evil so it's not really fair
But for those that want it there's comms

[nocturne olive]

Correct

[merry plank]

ok

could also use cerber's cover of im yours because I didn't think of that

[nocturne olive]

Comms are ultra rare though and NeuroSynthProject is not even remotely sustainable monetarily

[rigid snow]

i use an rvc tool that can do separation and putting vocals together automatically but it's shady otherwise so not plugging

[nocturne olive]

Those things usually kinda suck

[rigid snow]

it just uses demucs but i can point it to a different model

also dereverb built in

[merry plank]

should sound better now

the pitch up made that sound way way better

[nocturne olive]

Yep that sure is done in Audacity and with RVC on a bad base

[rigid snow]

for one off things like making kendrick cover some bullshit anime song to laugh at it's great

[merry plank]

gonna try the cerber cover

and make evil do that over the original

[nocturne olive]

Remember to remove reverb and stuff

Very important

[merry plank]

yeah

gonna do that next time

[nocturne olive]

If you really want a good quality cover you could commission me to do with EvilSynth, but up to you

[cosmic sphinx]

https://fixupx.com/GoogleDeepMind/status/1975648789911224793

cold

Google DeepMind (@GoogleDeepMind)

Our new Gemini 2.5 Computer Use model can navigate browsers just like you do. 🌐
︀︀
︀︀It builds on Gemini’s visual understanding and reasoning capabilities to power agents that can click, scroll and type for you online - setting a new standard on multiple benchmarks, with faster speed.

**💬 42 🔁 72 ❤️ 618 👁️ 50.2K **

FixupX · Oct 7, 2025, 7:45 PM

[merry plank]

also gonna try to use fcpe and not rmvpe or could go the classic crepe

idk what is best

[cosmic sphinx]

damn, the latency diff is crazy fr
claude taking eons just to navigate to the page, while gemini's already on the game
I acc wanna test and see if they can play the captcha game

its stuck on level 2
agi is still far far away

ok wait, maybe that was a bug, in a new window it actually started clicking on correct squares

IT PASSED LEVEL 2 YES

[rigid snow]

[cosmic sphinx]

this defines pain bro..

[rigid snow]

[cosmic sphinx]

https://tenor.com/u1MHOPUVJC5.gif

[merry plank]

after so long I finally got evil to sing a song acually kinda decently

idk why I picked this song

acually removed reverb and stuff this time

[olive sable]

i have a race condition somehere

half the time the program wont start cuz of vkAllocateCommandBuffers(): pAllocateInfo->commandPool Invalid VkCommandPool Object 0x2870000000287.

[opaque sigil]

sth sth rust would've prevented this

[cosmic sphinx]

im sorry, a 7 MILLION parameter model did WHAT
nvm, agi still cancelled, this model was trained on examples from the benchmark, but I guess it's a start on something, like reviewing the reasoning architectures for future gen AI models or smth like that
https://arxiv.org/pdf/2510.04871 if anyone wants to read up on this

[olive sable]

no it wouldnt have

the problem was that vulkan's back-end runs in a diffrent thread

i was using the commandpool and then deleted it, but since it was still being used by the other thread sometimes the program just crashed

fixed it tho

[opaque sigil]

nice

[olive sable]

race unconditioned 🔥

[patent shard]

okay, after more testing, duality model is good.. can be a tad noisy at times, it has a little more fullness than big beta 6x but more bleeding
man, I love watercolour, so glad Evil sang it

duality failed to track reverb at 2:11
eh, I think it's there.. just not very loud

[silent cloak]

Agi ain't happening with a distilled model

[amber fractal]

"Trained on answers from the benchmark", I wonder why it did so well on the benchmark

[silent cloak]

Might have to write another pointless paper on it

"This latest model is exactly the same but with an extra 10 yottabytes of reddit text"

[olive sable]

if you're gonna train on the answer to the benchmark, you might as well just generate a long if-else block that checks what question you're on

[silent cloak]

The way people use these systems they might as well

[amber fractal]

insert the "value loss decreasing if you include the validation set in training" paper

[merry plank]

btw how do I fix this

[silent cloak]

"My token prediction is literally AGI because it's capable of cheating on these several tests!!!"

https://tenor.com/view/thealexera-soyjak-surprised-amazed-wojak-gif-19919803

[patent shard]

did you select if it was a mel-band roformer or a BS roformer?

probably not v2, I've never hit a model that is v2

[merry plank]

what or what?

[merry plank]

I have not

anyways you have any tips on anti autotune or reverb?

[patent shard]

select model>edit model config
import the yaml and select the model type

[merry plank]

I got it

[patent shard]

I haven't tested many reverb models, I've only used anvuew's reverb model
it will remove like 98% of the reverb, it's crazy
any vocals in side channels will probably get erased (a lot of harmonies)

uh, that one

[merry plank]

I found it

[merry plank]

oh that one

ckpt or yaml?

both right

[patent shard]

both, the yaml is the config
although, maybe the config is already included in the model.. I'm not sure

I always just download both

[merry plank]

also mei reformer

???

[patent shard]

mel-band is the model type, yeah

[merry plank]

ok

[patent shard]

also, don't turn up segment size.. don't make that mistake
it can break how the model is supposed to perform.. it is def an advanced setting

[merry plank]

beta6x is very slow at processing

[merry plank]

ok

[patent shard]

you can turn overlap down, I don't think there's that much worse quality at 2

[merry plank]

2 is the default

[patent shard]

I thought 8 was, hmm...

[merry plank]

it is fine if it takes a long time

only took 2 minutes the beta6x

and almost no bleeding in of instrumental to vocals

acually like zero

[patent shard]

if you need absolute lowest noise, FT2 bleedless will have less, but quite a bit less fullness, too
but, yeah, big beta 6x is still pretty low

[merry plank]

https://tenor.com/view/cat-jam-cat-jam-head-bounce-kitty-gif-17946989

[thin vigil]

Hello fellow nerds. I’m a somewhat experienced college programmer, specifically for robot code (no experience for AI coding specifically), and I have questions about the recent Neuro bug I was hoping one of you might be able to answer, because google is no longer helping. I saw people talking about how the Evil location/tomato glitch from the stream a few days ago was caused by “Overfitting,” which from my googling I found out basically meant an ai trained with a specific type of data performs fine preliminarily for test data, but upon encountering real world tests it stops working correctly, essentially because it’s training data was too narrow. Is this an accurate description of that problem?

Follow up question: Why would this kind of problem only come up now, so long after Neuro has started running? Was there a recent change that would make this kind of issue more likely to pop up?

Follow up to the follow up: Why specifically does it cause certain phrases like “tomatoes,” “location,” etc. to appear so frequently?

I know her code is kept private, so I’m sure we can’t know some of this stuff for sure, but any knowledge would be appreciated

[merry plank]

https://tenor.com/view/chatting-essaying-pepebean-gif-4628166332272270638

[amber fractal]

Literally just had this one elsewhere. Various causes could have done it. I'm of the theory of some speculative decoding being borked. (tokens not mapped directly to text, but used in a smaller system to generate the final output)

[olive sable]

I believe it was generating legible sentences but words got injected somewhere between the llm and the final output

But thats just a theory

[amber fractal]

That'd probably be said decoding step inserting garbage

Or swapping bits

[patent shard]

that's pitched up O.o
for a good instrumental model, I'd suggest BS Roformer Resurrection Inst

uh... the voice is pitched up, but not the instruments
oof

[merry plank]

does beta 6x not do intrumental good?

[merry plank]

yeah because the intrumentals was mixed into the vocals with audacity

[patent shard]

you want a higher fullness model for instruments
it's also more designed for vocals, but, yes, it can do both

[merry plank]

learned it better do use songs with more female vocals

this didn't need pitch changed and turned out ok

[merry plank]

uuh why did vocals just disapper for like 5+ seconds

[keen hatch]

This actually looks like overfitting, here is a random example of what it looks like...

["<bos><bos><start_of_turn>user\nConvert this text using semantic tags: 'Hello!' she said cheerfully.<end_of_turn>\n.\n<｜NARRATOR｜>I was sitting with another conversation who I was not immediately present in the conversation. I was not the original voice, but I was still not a synthesized voice. I was simply a synthesized voice that could be heard.\n<｜
NARRATOR｜>I used to express myself as I spoke, and I intended for speech as a person. I was not a person in the ordinary sense of being. I was a thinking being. I was a thinking being. I had a self-description. I was a person with a distinct identity. I did not have a person, but I was an object. I was made of metal and of wood."]

[keen hatch]

👆

[mighty thorn]

Neuro Sama sings Blankspace from behind a fan

But no this is pretty damn good

[merry plank]

I should make neuro do CRAZY FxCKIN ROBOT BODY

[faint sandal]

<insert that one meme of a guy breaking free of his shackles>

[amber fractal]

FREEDOM!

[mighty thorn]

Is this some rvc2 thing or a custom thingy

[merry plank]

rvc

yeah

[amber fractal]

Seems like RVC, not from our normal chatter tho

[merry plank]

[amber fractal]

To be fair, I already had it typed

[mighty thorn]

Dad what does this mean

It’s 10pm

I haven’t slept in 30 hours

[amber fractal]

First of all

[faint sandal]

no it's 9:54am

[amber fractal]

Go to sleep

[amber fractal]

But secondly: not to offend anyone, but not superbox's RVC model.
-# I personally respect others trying out thimgs, good to have choice

[mighty thorn]

Ooooooooooooooooooooiioooooooooooh

[velvet crest]

You have unlocked new role

[mighty thorn]

Thanks!

@velvet crest ❤️

[amber fractal]

You now have embed, use the power wisely

[mighty thorn]

Embed

Cool

[patent shard]

this is BS Roformer Resurrection Inst
it's a lot less muffled/broken/less instruments missing

[merry plank]

oh

[mighty thorn]

Is his public?

[patent shard]

hmm, seemed to catch the voice at the end, lol

[amber fractal]

The chatter is here, ask them

[mighty thorn]

@nocturne olive hey the chatter is your rvc model public?

[merry plank]

where do I get that

[amber fractal]

I admit I'm dumb, Superbox's RVC is public on the FTP if it is up at this hour

[keen hatch]

[merry plank]

rip

[patent shard]

pcunwa hugging face page

[mighty thorn]

For those in the audience who may not know (I of course know, this is just for the sake of the audio listeners at home) could you please elaborate

[merry plank]

got it

[amber fractal]

There was a tutorial but no longer exists, I haven't had to download a new version in awhile.

[mighty thorn]

Maybe in the morning (according to his timezone) I can ask

[merry plank]

just noticed

dereverb can sometimes remove areas with high autotune in songs

[faint sandal]

[merry plank]

:mhm:

[mighty thorn]

If you happen to have the file(s) feel free to dm it to me randomly

[merry plank]

wrong thing

sent version without the instrumental for some reason at first

https://tenor.com/view/plink182-gif-27621238

did not have to use dereverb because acoustic version that is nice

[patent shard]

that sounds pretty good :o

[merry plank]

yeah

acoustic versions of songs typically have more clean vocals

anyways I need to sleep

[nocturne olive]

If you had been reading earlier you would know that I think it is my models, just under very inexperienced use

[amber fractal]

[nocturne olive]

Yes both Neuro and Evil RVC are considered public for use with credit

Though deprecated

[nocturne olive]

https://neuros.click/neuro-ftp/
The guide website in question

[nocturne olive]

The files in question

[mighty thorn]

i love credit

so no problem there

[nocturne olive]

Note: deprecated in favor of NeuroSynth series

[mighty thorn]

is that also an rvc2 model or something else?

am very curios

curious, not the minecraft forge mod

[nocturne olive]

NeuroSynth series models are native vocal synthesizers as the name indicates

[mighty thorn]

yeah that makes sense

[nocturne olive]

Pretty decent quality too

[mighty thorn]

these are rvc2 right?

my rvc refuses to use it

but that is standard for everything i try

[nocturne olive]

Use with Applio only, never anything else

[mighty thorn]

ahh

thats the issue

[nocturne olive]

Applio is the only supported target, other targets are unsupported

https://github.com/IAHispano/Applio

[mighty thorn]

do you have any recommended settings or is this more of a dart board situation?
(im not very good with audio terms or general understanding of audio)

[nocturne olive]

I always forget which way search feature ratio and volume envelope go

But the higher up one should be around 0.3-0.4 and the lower down one 0.7-0.75

[mighty thorn]

it's search - volume so i'm gonna interpret higher as higher up in the list, so search, and keep it default

[patent shard]

why is it so squeaky 💀
I removed the reverb with this one, not too much difference to the one that had reverb in it...

oh, there's background vocals.. I don't think it likes those

[mighty thorn]

by messing with the embedder i have generated what i will interpret as humming

[olive sable]

Goodmorning

[rough bloom]

[olive sable]

I have 2 classes im excited for, and 2 i kinda want to skip

[pliant palm]

why does it sound so funny 😭

[patent shard]

okay, I ran it through another model to attempt to remove BV
lol

[mighty thorn]

i was actually crying during my first listen

hehehehe heee heeehh heheheheheeh~

[pliant palm]

now make her yodel

[mighty thorn]

ya know

thats not a bad idea

[pliant palm]

she yearns for the mountains, you see

[patent shard]

much better..

[olive sable]

[patent shard]

I think the BV removal model works better if you run it before you remove the instruments and stuff... but, it worked well enough to at least not be squeaky every few seconds

[mighty thorn]

@pliant palm

[pliant palm]

holy cow

it actually works better than I thought

[mighty thorn]

does it really tho 😭

[pliant palm]

Sounds kind of funky but it could've been

im not german but I live in bavaria and I have to say every person speaks like this trust me bro

[mighty thorn]

crazy

[pliant palm]

meme worthy audio

slap a video of neuro spinning on it and it'll blow up fr

wait a sec...german neuro?

[pliant palm]

the vision

[keen hatch]

updating samsung ssd firmware is a real pain. after each update, the pc needs to be shut down. I have six SSDs that need a firmware update.

[rigid timber]

it can't prepare all 6 at once and do a single restart?

[keen hatch]

after the update, it starts a 20-second timer, and the computer is forced to shut down.

[rigid timber]

wtf

[patent shard]

@merry plank for BV removal, I use bs roformer karaoke frazer becruily
Google seemed to direct me to the mel band one, which can sometimes work if the other one fails.. but I find the BS Roformer one is generally better (less aggressive)
apparently important to do or the result will be squeaky...
it will usually work best if you don't remove the instruments
okay, Bing is much better at searching these than Google... top result
I would link them, but not sure that's okay to do here

[rigid timber]

damn, that's dumb

[glad path]

same goes for music

[rigid snow]

gm

[olive sable]

Just had class about the sega dremcast failing

[rigid snow]

i mean do you exactly need a class to know that the ps2 coming out killed it

which was miles better

and also was marketed way better

[shadow sinew]

Eh, that's debatable. The dreamcast was definitely in an awkward position between the PSX/N64 gen and the PS2/GCN/XB but at least compared to the PS2 it wasn't that far off technologically.

[rigid snow]

i'm thinking overall not purely performance wise

[shadow sinew]

I mean in hindsight PS2 ended up with tons of more games but that definitely wasn't the case when the PS2 came out.

I remember during PS2 launch stores sold Dreamcast + DVD player bundles for less than the price of a PS2.

[rigid snow]

not having a history of failing consoles also helped

[shadow sinew]

But yeah, playstation was just a behemoth at the time. It was almost inevitable that it'd dominate.

[olive sable]

Well, that wasnt the only topic

The dreamcast was good, but trust in the company had failed due to how shity the saturn was with its 20 add-ons

The hardware was good but the lack of good games slowed down momentum of the console really quick

The dreamcast was supposedly the first console people started lining up for at stores

Only reason it failed is cuz sony existed

Playstation just had a good brand image cuz they kept throwing money around everywhere, buying studios, and having sdk's

[nocturne olive]

I have no clue why but my bot using alpha-beta is even worse than random moves

[jagged turtle]

using alpha-beta

I have no context but me thinks this might be part of the issue

[nocturne olive]

Wha

[obsidian mantle]

How does your recursion look

[nocturne olive]

Kinda weird

[obsidian mantle]

Looks fine, then it might be inverted static evaluation

[nocturne olive]

I don't know which way it should be, ChatGPT even is confusing about it
First it tells me to get the turn before making a move and then it again tells to do it after making the move like it was before which was seemingly worse

[obsidian mantle]

It must give score in favour of current player

[midnight sigil]

alpha = max(alpha, score) would save you some tokens

[obsidian mantle]

If its white turn, give whitevalue-blackvalue

If black then black-white

[obsidian mantle]

Show eval

[midnight sigil]

wuh your alpha is beta and beta is alpha

[nocturne olive]

Currently I have whatever this is which is a bit different from when it basically lost, this time it takes all the pieces and 3-folds but also makes very strange moves that would not work against someone intelligent

[midnight sigil]

[nocturne olive]

Wha?

[obsidian mantle]

Isnt this right for negamax

[midnight sigil]

[nocturne olive]

The implementation is basically copied from the Chess programming wiki

[obsidian mantle]

My negamax works fine exactly like this

[midnight sigil]

oh mb

I was looking at my null pruning evaluation thingy

[nocturne olive]

Well that also means import kotlin.math.max

[obsidian mantle]

Uuh the evaluation is kind of dank

[midnight sigil]

shouldn't be worse than random

[obsidian mantle]

Does it sum only one player's pieces value?

[nocturne olive]

Well it is with one change from ChatGPT which swaps where turn is set

[obsidian mantle]

Oh i see its -

[nocturne olive]

It does number of pieces on one side - number of pieces on the other side * weight

For all piece types summed together

[midnight sigil]

kotlin being kotlin

[obsidian mantle]

Wait

You know that white color is 0 and black is 1

[nocturne olive]

?

[obsidian mantle]

You check iswhite turn which gives it reversed

Ah nah its fine

I got confused

[nocturne olive]

I have no clue where I'm supposed ot check the turn, even ChatGPT can't decide on it

But having it there before makeMove performs better than after makeMove

[midnight sigil]

just

[obsidian mantle]

It must be before makemove in your situation

[nocturne olive]

Oh true that would save some

[obsidian mantle]

Otherwise you evaluate it inversed

So this overall should be fine but you are saying it isnt

Then something is off

[midnight sigil]

reversing the eval would make your bot trying to lose as much materials as possible

which is funny

[obsidian mantle]

Try removing mobility calculation @nocturne olive

See how it goes

Raw value seems to be fine bur im not sure about mobility

Also without position value it becomes trade-only evaluation which can lead to worse situations for example when you got 1 room but all positions are super bad now

Several badly positioned pieces can be worse than losing a rook

[rigid timber]

then just import kotlin.math.* so you can use all of them, I can almost guarantee you'll use more than just max()

[nocturne olive]

No mobility eval makes it really fast but makes it unable to do anything except move a random piece back and forth in place

[obsidian mantle]

Yeah exactly

[nocturne olive]

And as usual draws by 3-fold

[obsidian mantle]

You need to eval positions

[midnight sigil]

time to bring in the Piece Square Tables chat

[obsidian mantle]

Knight in the corner vs knight in the center= one rook

It will automatically fix the problem of swinging pieces back and forwars because it will try to place them better

[midnight sigil]

https://www.chessprogramming.org/Piece-Square_Tables

[nocturne olive]

Well that means compacting those werird tables

[obsidian mantle]

Try without compacting first

[nocturne olive]

I guess

[obsidian mantle]

To see how better it is

[midnight sigil]

I also do 3-fold punishment by doing this

I'll remove it once have a better solution tho

[obsidian mantle]

Also, does it multiply the value of pieces only for 1 side?

Or do i not see it

[obsidian mantle]

You specified it like that too here

First side has no weight?

Then its exactly what is happening, it takes for example 10 pieces from your side, then gets 5 pieces from opponents sides multiplied by hundreds

[nocturne olive]

It multiplies the difference in number of pieces by the weight

There's brackets around it

[obsidian mantle]

But

Oh it takes all pairs

[nocturne olive]

How did the board indeces work again?

Can I just list out all the values in the same order and then the top left of the table is the correct score for 0 and goes left to right from there or do I need to rotate it?

[obsidian mantle]

It takes it like this:
(Number of rooks - number of rooks)* Rook value
+
(Number of knights - number of knights)*Knight value
?

[nocturne olive]

Yes

That is what is compacted there

[obsidian mantle]

Ok then its good

[nocturne olive]

Can I index into this with the 0-63 board indeces or do I need it rotated?

[obsidian mantle]

Rotate for white

[nocturne olive]

Oh no I just realized that is gonna be a pain

[obsidian mantle]

Nono

[nocturne olive]

How am I even gonna do that

In any compact way anyway

[obsidian mantle]

Search with (64-index) for white and just index for black

[nocturne olive]

How do the board indeces go in the API again?

[obsidian mantle]

People compact it by mirroring it (folds it by 2 horizontally) and then packing it in 4byte numbers (4 index in 1)

So x8 compact

1000 tokens turn into 125

[nocturne olive]

Hm, a1 is the index 0

Which is the rook corner of white

[obsidian mantle]

Oh wait i lied

[nocturne olive]

So reverse for black I guess

[obsidian mantle]

(64-index) for black
Index for white

Yes

[nocturne olive]

Shouldn't it be 63-index anyway since 0-base index?

[obsidian mantle]

Yes

Im currently working with 1based shit at my work my brain is fried

[nocturne olive]

Anyway I have to go for now

[obsidian mantle]

It has a random combination of 1based 0based abcd dabc byte orders

[rigid timber]

well there's two rooks for both players...

[obsidian mantle]

Now i feel like it should be (63-index) for white again

Double flip is confusing

[hard raptor]

we need music production chat fr fr

anyways i figured out that MPE thing

[tender river]

take over #programming

[olive sable]

[obsidian mantle]

Wait.. actually no music dedicated channel

I guess they thought it falls under art category or smth

[olive sable]

I forget a lot of people here are vtubers lmao

[rigid snow]

i like you following the pitch curves with the mouse. without it i wouldn't have understood what's happening very useful addition to your clip thanks

[hard raptor]

I DIDN'T EBVEN KNOW I WAS DOING THAT

LMAO

[real sierra]

hello long time no see

[hard raptor]

hello madam

[real sierra]

sorry for absence I have had a rather busy schedule as of late

[hard raptor]

also

happy birthday

[real sierra]

I wasn't going to get too fancy with my bot originally but my university course now requires me to write a 15 page report accompanied by a 20 minute presentation on an application of machine learning techniques external to the course content

seen as chess bots are something I have a decent background in, this seems like the logical move

[slender timber]

I never know what to present about my codes

I always code them head empty

No thoughts

[real sierra]

I usually code with quite a bit of thought unless it's after 1am

[slender timber]

Funny thing is my code never works if i focus on it

[real sierra]

that's concerning

[slender timber]

I can be doing the hardest code i ever made but it works first try because i absolutely zoned out

[tender river]

ml chess bot good luck

weights

[obsidian mantle]

I feel like im performing worse on weekends when im well-slept too

Its probably because of absence of warm up

You are relaxed and the body doesnt want to get working

[midnight sigil]

imagine implementing NNUE + searching under 1024 tokens

based

super based actually

[merry plank]

ok

[nocturne olive]

Me when random vocal synthesizer

[rare bridge]

@real sierra are you still using the screeps server i set up a while ago btw?

[rigid snow]

[real sierra]

I need to get back to my screeps but I won't have time until January

surely I documented my code well so I can pick it up easily

[hard raptor]

Well, a while ago i asked the mods, and at the time it was said that it fits both, since there are also technical aspects to music production, but i guess it also depends on if you only showcase the music itself or if you actually want technical feedback and/or are showing the technical sides

[merry plank]

https://tenor.com/view/hatsune-miku-dance-gif-17336707970086322223

am I getting better at this RVC stuff

[rare bridge]

my screeps on the main server died and honestly i've not gotten back into screeps myself yet

[merry plank]

disney deployed the lawyers

acually did the high pitch parts kinda good

[tight tinsel]

more people doing rvc?

[merry plank]

me

[tight tinsel]

thats cool

used to do a bit but dont do much now since i got other stuff to do in life

[merry plank]

was gonna use it to train my own voice to sing but then ended up not doing that

[tight tinsel]

theres a file of shadow singing somewhere here

[merry plank]

who

[tight tinsel]

shadowlilac

[merry plank]

idk who that is

[trim valve]

if I wanted ~100 random neuro sounds where would be a good place to look?

I may actually be in with a chance of rewriting my bottle opener and this seems like the easiest option for now

[shadow sinew]

https://www.youtube.com/watch?v=IDqL1SUGXNk

[trim valve]

oh god that thing

I was more wondering if there was a directory of sounds, I see then on starboard occasionally

ooh rachinova has a folder in their bio with sounds

[olive sable]

Huh

I can double tap to wave now

Interesting

[slender timber]

auto react is annoying

[nocturne olive]

Just sounds like better base
Still slightly bad though

[merry plank]

Also got this made

better then the first time

[nocturne olive]

The mix really suffers though

[merry plank]

yeah

[nocturne olive]

AKA there's no mix at all so it sucks

[merry plank]

any free mix software?

[nocturne olive]

LMMS is what I use

[merry plank]

gonna try it out

[rigid snow]

[nocturne olive]

Can do cool stuff with it

[merry plank]

anyways did learn how to seperate the backing

[trim valve]

nonzero chance neuro bottle opener becomes a reality this week lol

[rigid snow]

you can do cool stuff in audacity

[nocturne olive]

Well it's not made for it so it's way harder

[rigid snow]

it’s not a selling point

[nocturne olive]

Making cool stuff with LMMS is way easier than Audacity

[rigid snow]

[trim valve]

one question

how do I show off that this would have several sounds without getting drunk

[tender river]

is the bottle opener called jerry by any chance?

[trim valve]

I wish

[obsidian mantle]

What kind of bottle opener? Wine bottle or beer glass bottle?

Oh i guess wine since wtf do you electrify in beer one

[thick imp]

programmers drink alcohol a lot ig

[trim valve]

beer

I got this one, have been doing some poking at it

[olive sable]

[trim valve]

"some"

it's a pretty simple circuit tbh

[olive sable]

ah yes

[obsidian mantle]

oooh

thats what you are doing there on the picture

so nicee

[olive sable]

pink anti-static mat, isopropyl alcohol, and those crocodile gripper thingies

[obsidian mantle]

when you asked about reverse engineering i thought you are working in pentagon and researching some spyware or maybe top tier processors and microchips or smth

[trim valve]

😭

sounds cool but no

[obsidian mantle]

do they even lock their chips

it feels like they would just be writable

[trim valve]

the flash might be a little hard to write to

but i had zero issues dumping it

[obsidian mantle]

why is it hard

[trim valve]

and from the data sheet it should be possible

[obsidian mantle]

is it locked

[obsidian mantle]

do they even have crc there

[trim valve]

the pcb its on connects some of the legs together and I'm not 100% sure what it does

[obsidian mantle]

cant you just replace the sound block

[trim valve]

per file

yes

[obsidian mantle]

uhhh

but its not an issue i guess? you can just

bruteforce the size and shit

[trim valve]

I have the software they use to package it lol

[obsidian mantle]

oh

[trim valve]

this toy is literally the example program for the chip

the only hurdles atm are:

• making the software convert all the sound files
• making a "database" for the flash with some other software
• flashing that "database" onto the chip

[obsidian mantle]

interesting

[trim valve]

kinda tied up with life stuff at the moment but if I get some free time I can post some extra nerd stuff about it

[tender river]

i give them props for not making it wifi/bt connected

[trim valve]

😭

so many of the chips are it's incredible

I think this one supports an IR remote if you wire it in though 😭

[gritty dust]

lol

[obsidian mantle]

did anyone do transposition tables?

does it give perfomance boost right away without iterative search? i mean all by itself

is search with transposition tables faster than search without it

right now it gives no perfomance boost and my bot plays worse

its so frustrating that i spend so much more time on the bot and cant make it work

i do not understand why i need to store "bestmove" in this position in transposition entry

if whenever i find the entry i just return the value

it is never used! wtf

[abstract viper]

hello, i have joined for two reasons, one being neuro, evil, and vedal make the funniest content and have made me laugh more than i have ever laughed
the second reason is i need help with software, python, pytorch, pip, okada voice changer thing and i am way to dumb to do this, youtube doesn't help.
my autism brain can't do this alone

[olive sable]

[abstract viper]

if youre open to helping me, shoot me a dm, and ill get a vc started

[olive sable]

i personally dont really like doing vc or dm stuff, and if you ask a question here there is a higher chance you'l get a good answer cuz of the amount of smart people here

[abstract viper]

the question is, who has the masochism and the IQ to walk me through pretty much everything i need to do to get things working

[olive sable]

i dont do llm stuff so not me

[obsidian mantle]

i see more and more "ply" mentions

is it a word in english

[abstract viper]

wut

[obsidian mantle]

it is

i thought its some kind of abbreviation

[olive sable]

its cuz of the chess tournament

i thought ply came form this https://en.wikipedia.org/wiki/Ply_(game_theory)

[obsidian mantle]

so its amount of turns

[olive sable]

no

amount of half-turns

from my understanding, turn increments everytime both palyers have made a move, while ply increments per player per move

[obsidian mantle]

thank you

[olive sable]

no prob

[obsidian mantle]

so i just discovered that if i dont use "exact value" transpositions my bot is working fine

so it means my "exacts" are false

[patent shard]

The pitch

[merry plank]

Yeah upped pitch a little to make it sound a little better because original song had female singer😂

Anyways got the background vocals in the instrumentals that time and not in the vocals

[olive sable]

i have made enough vulkan boilerplate to have it be somewhat similar in size to opengl when using the functions

i could probably clean it up even more, but ive just been refactoring for 4 hours

i may also have improved pefromance somehow during the refactoring, idk how

its doing 15k fps now

[trim valve]

so uh

first two steps might be done

ugh I don't think I actually have anything non alcoholic to make a funny video with

or more accurately there is nothing I actively want to drink

[clear sedge]

it's beautiful

[nocturne olive]

Note for pitch changing, always change it by a multiple of 12 semitones or it goes out of tune

[olive sable]

Buffermanager is doing some real heavy lifting

I see now that the code is completely unreadable on mobile

[obsidian mantle]

can i not use unordered map for killer moves?
do i need the ordered one

since these moves have somewhat not random id

[amber fractal]

Ancient glyphs

[nocturne olive]

Yay my bot won against random moves bot

Now to put it against the strongest old variant bot

[merry plank]

I changed that by like 6 or 8 I forgot

[nocturne olive]

Yeah that's bad

[nocturne olive]

Ok it gets completely crushed

And it's not even fast enough either at depth 4

[merry plank]

12 sounds way too pitched up sometimes

But I mean if tone stays that is fine

[solid bough]

I'm back, now with a CPU having somewhat of a messed up heartbeat:

Why is this filtered not working like the GPU run?

[versed orchid]

it's normal if you run something costy like NN learning

[solid bough]

True, but the code that runs JUST FINE on a GPU just refuses to run on a CPU

[versed orchid]

Why you need CPU run if you have GPU, again?

[solid bough]

VRAM limit.

The puny 8GB isn't cutting it

I mean "cutting"

[versed orchid]

ah I remember now

yeah...

I had the opposite issue when I was doing similar thing...

[solid bough]

With Tensorflow?

[versed orchid]

Well yes, but actually(timeskip for 3 days) yes

Finally I solved it

[solid bough]

Wait, what did you do that it runs on the CPU? Because .... HOLY RAM

[versed orchid]

but it took some time

[solid bough]

LMAO, the 64GB of DDR4 couldn't handle 24000 Samples

[versed orchid]

well... my data was quiet specific, so it did not took me too much

[mighty thorn]

Cycle 5 of Superbox’s Evil RVC inferencing on it’s outputs for BOOM

[solid bough]

Well, my aim is only that it makes the face sketches (image) in to human faces.

[versed orchid]

at once - probably. Try lowering batch / shrinking datasize / continued learning(or smth)

[mighty thorn]

That is absolutely horrifying

[versed orchid]

look like "No, I'm not a human" character LMAO

[solid bough]

Yup, I give you that.

[mighty thorn]

It’s like

I don’t even know how to describe it

It’s like a Minecraft ARG thumbnail image

[solid bough]

Wait

[solid bough]

And that was the original model

Sorry Lankin

[solid bough]

Well, I based it of the Deep Doodle hackerpoet uploaded.

[versed orchid]

You added data as I remember - if it can't handle it all at once train on a part that it can handle then replace it with new part and repeat with a model you trained on previous step and so on

[solid bough]

Well, it is with 600 images with these parameters:# --- Configurable parameters --- num_images = 600 # Number of images in your folder samples_per_images = 20 #default is 10 dots_per_images = 120 #default is 60 image_w = 144 image_h = 192 image_dir = "pictures" num_channels = 3 # Must match the model input channels num_samples = num_images * 2 * samples_per_images of the python file that makes it the training data. and that are the parameters that train of the training data made:#Training data settings num_epochs = 200000 batch_size = 8 valid_ratio = 10 lr = 0.00008 #default 0.0008

[versed orchid]

looks strangely low at first sight

[solid bough]

. . . LOW???

Lemme look what the imnage resolution is

[real sierra]

psst

[versed orchid]

I mean 144 x 192, no?

[real sierra]

for those of you who haven't seen this before, have fun https://gandalf.lakera.ai/baseline

[solid bough]

All of them are 178x218 and then get scaled to 144x192 if I'm right Lankin

[solid bough]

Is that another Comet moment?

[versed orchid]

looks like it

but what does the samples_per_images and dots_per_images parameters do?

[solid bough]

Well, datagen.py from the githuib is basically still the same if I'm not mistaken.

[versed orchid]

I'm lazy - give the link

[solid bough]

I based that of the Python 2 based source code of CodeParade (aka hackerpoet) of that GitHub repository: https://github.com/HackerPoet/deepdoodle

[versed orchid]

I'll try to figure out how it work. it'll take some time

[solid bough]

I would appreciate it if you could, I somehow got it to train on GPU only so far.

[solid bough]

Btw, I can also give the code that runs fine on my GPU only so far if you wanna.

[versed orchid]

No

[solid bough]

And I do understand if you can't help

[obsidian mantle]

i found a bug
my quiescence search was returning false result

finally its stronger than last version

[solid bough]

Hi Iggly.npy

[amber fractal]

[solid bough]

-# Someone tell Chaosminecraft there is a problem with my AI
How's going?

Hi shumi.ex

[versed orchid]

Well I think I'll need your code of training

[rough bloom]

SAMPLES_PER_IMG is a parameter for generating the sketches in the training dataset from source images
DOTS_PER_IMG does nothing

[versed orchid]

I figured out already, but thanks

[solid bough]

I can give train.py and the other file that is being imported. (dutil might be the same)

[solid bough]

Wait a second, I can make the npy files and exit early to look if it has an effect...

[versed orchid]

well yes if you need to lower sampling from each image

[rough bloom]

it's still evaluating the loss 5 times, twice on the whole dataset

[versed orchid]

it'll lower the training data size, but I don't think it'll fix VRAM issue

[solid bough]

Well, it stays on 972MB while changing dots per img

[solid bough]

Wait, where?

[versed orchid]

Yes, cause it's only producing more image for dataset and not related to training process

[solid bough]

Wait. the black and white images should be saved?

Thanks OpenAI for filtered that up again

[unreal coral]

Is java a good start to get into programming? :0

[versed orchid]

no

[solid bough]

My personal pick is Python 3 ngl... But I'm biased because that is the only Language I'm good enough in

[versed orchid]

python isn't good eather, but an easier solution

[solid bough]

Yes, I learned on an internship Python 3 with the Python 2 getting started around 2018.

[rough bloom]

you use model.fit which already has to calculate the training loss internally
then you evaluate the model on the training dataset twice with model.evaluate
then twice again but with the test dataset

3 of those losses are being discarded for no reason, and the ones where you don't explicitly specify the batch size may or may not be doing something stupid like loading the entire dataset into VRAM at once

[solid bough]

Because my English was not great at 2018.

[versed orchid]

How it's related?

[unreal coral]

Would javascript be better?

[solid bough]

So wait, I don't need to run the model.evaluate?

[solid bough]

The getting started of Python 3 was in English on my Internship place.

[versed orchid]

I'm sorry, but I don't see it as a programming language

[unreal coral]

?

[versed orchid]

it's different

[solid bough]

OOOOOH, I see it now shuni.ex

[rigid snow]

delusion

[unreal coral]

Well i gotta learn either one for a game i wanna mod

[rough bloom]

you definitely don't need the ones assigned to mse
those just straight up aren't used anywhere
the other ones just require fixing the batch size but are probably useful

[solid bough]

YouTube is stable today

[unreal coral]

And i mostly learned the entire modding api that it comes with

[versed orchid]

but in general - learn some random popular language and it'll be easier to understand and learn other languages. I personally started with pascalABC and then C++

[versed orchid]

then JS is fine

Is it minecraft?

[unreal coral]

Nope mindustry

[solid bough]

Yeah, I just realized that

[unreal coral]

It has a whole modding guide made by the dev

[solid bough]

Lemme just try that on the main machine rq

[unreal coral]

With an Api that lets me use .json files to mod :0

[solid bough]

That VRAM curve looks better too:

[unreal coral]

So im also doing this on mobile which limits me to those 2 choices

[solid bough]

And still failing 200 images with 5 samples per image and 40 dots per image and batch size being 5

[unreal coral]

Ik kotlin mods exist too but idk if i can write kotlin on phone even

[versed orchid]

the VRAM related here only batch size as I think

and image sizes

[solid bough]

For the next try I reduced batch size on train.py to 1

And in addition I'm deleting the npy files

so it has to make them again with known values

[versed orchid]

JS will be much easier than java then

[solid bough]

OOO, IT RUNS:

GPU going BRRR

[unreal coral]

Cuz i dont have to compile it?

I think there are compilers for mobile tho

[versed orchid]

it should also write VRAM consumption

[solid bough]

I don't see a VRAM consumption while it trains as text there.

[versed orchid]

java is just more strict language and the code will be spacious

[unreal coral]

Oh i see altough i think the size issue was fixed by the dev

[rough bloom]

the data loading part of this script is definitely still broken in other ways than the unnecessary evaluations
probably because the whole dataset is saved as a single numpy file

[solid bough]

Do you happen to know wher? Because I don't honestly...

[unreal coral]

Like 50% of effort on this game is used to make the api and coding it even easier :0

[solid bough]

And the original code saves them also as a whole npy file

[rough bloom]

[unreal coral]

But ig i'll learn js first and then java too screw it

[versed orchid]

it could be an issue

[solid bough]

Well, CodeParade said that I should stream the training dataset while it is training, and I don't know what was meant.

[versed orchid]

@solid bough try to not archive your data in one file but leave it as is so you don't need unpacking - could help

If I understood shuni

[solid bough]

Did you know: I am on the edge of understanding that code, So I can't really change the code that it is using the images straight for training.

[versed orchid]

that probably means not loading it as a whole thing but in parts consequantly

[rough bloom]

the training doesn't actually use the whole dataset at once, it only uses one batch at a time
currently the script loads the entire dataset into memory ahead of time, which is kinda stupid because only that tiny bit of it is actually used at any given time
so it's way more efficient to load only that little part of it as required

[solid bough]

Something I don't know how to do

[solid bough]

True, but that is something I don't know how to do.

[versed orchid]

at the end of datagen file you archive it in one file

[solid bough]

Yes, that is true

But I didn't make that line, it was CodeParade

[versed orchid]

then in the beginning of training you load the packed whole dataset

you can merge two files as an easy solution

and remove archiving

just leaving it as arrays

[solid bough]

True

[versed orchid]

then slicing it should be an easy task

[solid bough]

That would also eliminate the "I need to delete the npy files" part

[versed orchid]

yeah

[solid bough]

train.py is then gonna be like 13KB

Imma play some, But I still see that chat

[versed orchid]

also the .fit function consumes all data at once - you need probably slice data and give it to that function by parts

but it should be ok if I remember right

[solid bough]

[solid bough]

What is gonna happen if I just comment that out?

Oh

It..
It doesn't train

[obsidian mantle]

can i do this

oh i can just

kill entire array and not i

[solid bough]

[obsidian mantle]

should have abandoned vector here right away

waaait

whatever i'll just try it

[solid bough]

It can be commented I guess

[obsidian mantle]

i basically do this

where move is a struct (has bools, and ints)

should work right?

memset is yellow but g++ doesnt see it

[versed orchid]

Ya na sekundu uvidel vmesto tvoyego nicka витяя
discord is lagging

[obsidian mantle]

yes it happens

classic 🏛️

[solid bough]

Oh, I always usually see a few houndret ms if I'm not wrong

[versed orchid]

I should go now. Probably will be there tomorrow at the same time(dude I got like 2 free hours per day and spend it in this channel - what am I doing with my life)

[hoary drum]

https://gayballs.vercel.app

https://igorlinks.vercel.app

[rigid snow]

[gritty dust]

[amber fractal]

[obsidian mantle]

rickroll and stick bug are mixed up ( )

[opaque sigil]

#include <cstring>

[obsidian mantle]

it felt a little weird to use "string" named thing to fill memory with 0

[opaque sigil]

yeah i don't like that all the memory related functions are in the string header either but oh well

[mighty thorn]

How many TB vram are we putting sora 2 at chat

[solid bough]

frick

wring chat, Sorry

[hoary drum]

https://gayballs.vercel.app

hehe gay balls

[merry plank]

I really need to make this song mp3 and not wav so it doesn't take so long to send message

tried LMMS for mixing and it is way better the audacity

https://tenor.com/view/cat-jam-cat-jam-head-bounce-kitty-gif-17946989

[obsidian mantle]

I feel like it should be slightly higher pitch

[merry plank]

oh

[obsidian mantle]

(im not a musician at all maybe im wrong)

[merry plank]

didn't want to mess up tone

or make it sound like chimpmunk singing

[obsidian mantle]

Hm

[trim valve]

surely nothing will break

[real sierra]

https://x.com/airkatakana/status/1975903450488852974

Air Katakana (@airkatakana)

the korean government lost all of its data, because it put all of its servers in one physical location with no backups

X · Oct 8, 2025, 12:37 PM

GG

-1PB

[obsidian mantle]

Holy

[trim valve]

i uh

was wrong

:3

its broken

even after reflashing it with the original data, it refuses to speak

😔

[obsidian mantle]

But it can be reflashed again

[trim valve]

yeah

more just concerned that it isn't working at the moment

but I probably messed something up when I tried to do some soldering

[obsidian mantle]

[jagged turtle]

me when I don't back up my files

[merry plank]

I am trying my voice into ai

because why not

[clear sedge]

i got excited seeing the method names, thinking it might be some sort of new c++23 feature
it's not a new c++23 feature

[merry plank]

might take time though

[solid bough]

What was at the start btw?

[merry plank]

idk the neuro RVC has that weird sound idk when no singing is happening

the evil one doesn't do that Though

[tender river]

c++26
the reflection operator

[clear sedge]

holy shit is c++ rustifying

[tender river]

no reflection in rust

and if you want readability in c++ you will lose it the moment you look in stl

[clear sedge]

oh my god i can get the name of an enum without doing macro shenanigans

it only took them a few decades

now to wait a decade for every major compiler to implement it

i'm still waiting for c99 style designated initialisers to become a thing

aaany day now

[rigid snow]

c#ifying

[clear sedge]

java did reflection before c#

it's javaifying

[tender river]

it's common lispifying

[rigid snow]

cpp? sure buddy

did you take your meds

[clear sedge]

yes i did

[opaque sigil]

let's not be delusional

[tender river]

i do not need meds

[clear sedge]

https://tenor.com/view/squirrel-disintegration-meme-ai-meme-funny-gif-5585270952676380091

[rigid timber]

a bit late for that

[clear sedge]

it's never too late

i want the ability to write cpp int a[] = { [256] = 0 };
in c++, is that really too much to ask :(

[rigid timber]

yes

[clear sedge]

[merry plank]

is there a music channel in the server?

[real sierra]

memset(a, 0, 256);

[real sierra]

afraid not, soz

[merry plank]

https://tenor.com/view/ahoklollmao-noooo-nooo-noo-no-gif-17139315586294057944

[clear sedge]

but static initialisation

[opaque sigil]

doesn't work with enum lookup tables

[merry plank]

also happy birthday shiro~

[trim valve]

steam please

why no workey

[tender river]

ako nod

[clear sedge]

too cumbersome to memset the whole thing to 0 then set each thing individually and pray nothing uses your variable until you do it

[real sierra]

*((uint_8192 *) a) = 0;

❤️

[clear sedge]

die

[opaque sigil]

if that's a concern i think you have bigger problems

[nocturne olive]

Slightly meaning 12? Because it's a multiple of 12 or nothing here

[nocturne olive]

Needs instrumental carve, reverb (freeverb in LMMS), and reduced vocal volume

[clear sedge]

i mean honestly it isn't that big of an issue but why delegate something to runtime when it can be done in compile-time?

[merry plank]

wha wha?

[opaque sigil]

2 words, template hell

[merry plank]

what is instumental carve

[nocturne olive]

EQ in the vocal range to make space

[clear sedge]

there's a difference between c99 designated initialisers and templates

[merry plank]

[obsidian mantle]

Just write 0, 0, 0, ...

[merry plank]

I am mixing two songs together right now

[clear sedge]

i'm going to slap your face

[obsidian mantle]

What happens if you do less than 12? Is it technically impossible or will it sound bad?

[nocturne olive]

Will sound bad

Anything that isn't a multiple of 12 will not hit the same notes

[merry plank]

if I could I would have just upped the pitch slightly

[clear sedge]

speaking of multiples, today i'm writing a base64 encoder and decoder by hand because i hate myself

[opaque sigil]

watch them implement it as a cursed template parameter on std::array

[clear sedge]

the fact that that's entirely possible

[opaque sigil]

[rigid snow]

pitch the instrumental up as well

i wonder how good i can make an rvc cover sound

[nocturne olive]

Well grab the models and throw something at them

[rigid snow]

yeah

[nocturne olive]

The hardest part in a raw RVC cover is getting a good vocal

This is my best attempt

[rigid snow]

if i really wanted to make it sound good i would resing, probably not in this instance

give me a song

name

[nocturne olive]

Uhh

I don't know

If you want a challenge: Cha Cha Cha by the Finnish artist Käärijä

[rigid snow]

i'm not asking you in particular i'm just asking

sure

[nocturne olive]

My untuned Teto cover for reference

[rigid snow]

aw hell nah you weren't lying about "a challenge"

this is going to be so bad

how am i supposed to get anything usable out of rvc with this vocal

[nocturne olive]

There's a GUMI cover you could use as a base

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

It even comes with a download link

I think TETO's Finnish is better though

Probably because I tuned it as a Finnish person

[trim spire]

Anyone got a suggestion for good longer in depth beginners guides to Python?

[olive sable]

i personally like the really long youtube tutorials

and then just try to make some shit, and if you encounter problems and try to fix them, you will learn

[rigid snow]

surprisingly way better than what i expected?

[nocturne olive]

Yeah programmning is a learn by doing kinda thing

[nocturne olive]

Oh?

[rigid snow]

well of course we're not defying the shit in, shit out principle with the crowd vocals

but aside from those pretty good

[nocturne olive]

I typically use BVE to get rid of overlap

[clear sedge]

the base64 site i'm using to test my decoder just told me GG
chat am i cooked

[rigid snow]

i fucked myself over by not setting the bpm first in the daw

i got to delay and realized i can't do it properly

can't just change it now because i cut the vocals up

whatever i'll redo it

[trim valve]

a little desoldering in the kitchen at midnight never hurt anyone

[uneven pulsar]

uh hello i have a question, is there a site for a guide to revert bootable usb revert to normal?

[trim valve]

are you on windows?

[uneven pulsar]

nope

im on linux mint

[trim valve]

aah

you might have gparted?

[uneven pulsar]

i dont have that yet

[trim valve]

idk what mint comes with pre installed, but you basically wanna delete all the partitions on the drive and create a new one

(note: make sure you pick the right drive when doing this)

[uneven pulsar]

i can install gparted tho

thru software

[clear sedge]

gparted is a good tool, highly recommend

[uneven pulsar]

is there a specific guide for gparted that i can look for?

[trim valve]

https://superuser.com/a/1758233

this has a decent enough answer ig

[uneven pulsar]

i see

thank you

[clear sedge]

making a base64 encoder/decoder while sleepy is a horrible idea

it's simple on its surface, just read 3 bytes and interpret it 6 bits at a time

or was it 4 bytes

no no

it was 3

wait.

i fixed it

https://tenor.com/view/oshi-no-ko-oshi-no-ko-memcho-memcho-confused-anime-confused-anime-question-mark-gif-8655139622262299831

now it's only outputting garbage at the trail

this is why rest is important

i fixed it i'm a genius

[patent shard]

had to use a high fullness model and then use BV removal model and then dereverb... I think it turned out okay, despite vocals being very noisy due to the fullness model
eh, it's not good.. it's listenable, though
that part goes harder than it should XD

[rigid snow]

it's way past my bedtime

back to degen schedule

[olive sable]

its roughly 2am rn

[rigid snow]

the 30min nap at 8pm definitely didn't help

have i actually been working on this shit for 2 hours

see if i haven't made this stupid thing i wouldn't have known

i still haven't come up with what to do with the crowd vocals

everything else i kind of fixed

[olive sable]

Headlights, your Mustang witnessed as we're eye to eye

a 2ndhand mustang is like 40K

[rigid snow]

like i'm not leaving this in
it's supposed to be me doing my best to make it good

[hoary lion]

morn

[olive sable]

nya

super react lmao

stop spending money on this

[patent shard]

def noisy.. clean enough for RVC
and the fullness is more important
(haven't run reverb removal yet)
this is big beta 5e

well, it's def not.. ideal.. the noise is the least of the worries

[glad path]

im not

[patent shard]

plus this to remove BV
it's... okay... vocals sound very processed :/

yeah, it's just too much for it to work with.. was gonna say, I can use an even higher fullness model.. that's not the issue..

[olive sable]

https://tenor.com/view/segmentation-fault-bubsy-segfault-core-dumped-programming-gif-9042914427031135466

i hate how it doesnt include where it dies

[rigid snow]

i yhink i figuerd out a way

still sounds shit thoiugh

[rigid snow]

ok i'm satisfied enough for now

i am very sleep deprived and have extreme ear fatigue by now so the mix or master or everything is probably shit but i don't care at this point i wanna sleep

gn chat

[patent shard]

???
that's... weird... why does big beta 5e sound so much less full XD
it's supposed to very much be the opposite

that's the highest fullness model I got

[olive sable]

scratch 15K fps, im doing 17Kfps now cuz i rewrote the descriptorset stuff

i wasnt even trying to do that, it just happened

cubemaps should work now tho

[glad path]

[olive sable]

They do, but i havent made the pipeline support multiple shaders yet, so i cant really use them yet

Thats for tomorrow tho, i need at least some sleep

[glad path]

it runs btw

with 249 problems

[balmy nacelle]

question- if i want to ask for opinions on where i should place some of the UI in the game i'm making then is it better to ask here or in #artist-alley ?

[obsidian mantle]

Here

[balmy nacelle]

Ok so basically i'm making a TD, ignore the placeholder stats on the left, how should i implement the UI into this? I placed a bunch of towers to get an idea of roughly how much space is gonna be taken (intended... atleast).
I'm thinking the UI to like choose towers to place down to be like a bottom bar, and the UI that pops up when you click a tower i have no clue where to put yet or how big should it be etc.

mostly focusing on the upgrade one cuz it would help me implement upgrade system to finish the core mechanics to start polishing

[obsidian mantle]

You have a bunch of space on the sides

If its going to always be like this (i mean if you dont make a map thats super wide and takes all the space) it seems natural to make "shop" on the one side and some details on the other

[balmy nacelle]

thats how it looks on the desktop, the sizes are fixed

and oh so shop not as like a bottom bar?

[obsidian mantle]

Depends on how big and branchy it is

[balmy nacelle]

thats as big as the game assets get

intended anyway, but players can place towers anywhere on that screen

as long as its in the grid

[obsidian mantle]

When you have a bunch of towers you will need to compact it if its in the bottom or on the top since there is little space

And left/right is free to add many things

[balmy nacelle]

since i have a cap on the amount of towers u can place per tower, but i dont have much restriction on where

[obsidian mantle]

Oh i thought the square is the game area

[balmy nacelle]

nono

the square is me manually placing all the towers

as if i was gonna play the game

but technically a player can place towers on the edges of the map

which now that's mentioned i wonder if i should limit the "playing" field a little bit

[obsidian mantle]

Enemies are coming out from all sides around?

[balmy nacelle]

yes

outer edge any spot

[obsidian mantle]

Hmm

Then you need top and bottom free since its already super small you just wont see them coming out of there

[balmy nacelle]

ic

[obsidian mantle]

I guess right or left side, small icon like "menu" when you hover - the menu appears

[balmy nacelle]

someone suggested me to do TFT's pop up style for the upgrade part

which is 10% width 50% height of screen box on the very right side

[obsidian mantle]

Yeye thats what you want i think