Made a Discord bot using OpenAI api's:
- and Basic Text-to-Speech (ai gen takes too long)
I left it sitting in a public discord for a while.
Was surprised how willingly people argued with it.
Eventually I hit record.
did they know what it was?
cant listen right now
>did they know what it was?
most didnt, no
some were speculative that it was a bot because its a bit repetitive, but whenever GPT said something with context they circled back to "no its definitely a guy typing"
over a 4 hour period, i think guy said it was definitely gpt but nobody listened to him
It’s insane how instantly recognizable the discord accent is, it’s either that or some Indian kid
I am both shocked that this is a thing and pissed off at myself for not anticipating something like that that would obviously happen.
wanna open source it?
That's because you're literally just typing text to speech.
>implying i can come up with sick zingers for 30 minutes straight
i shiggy diggy
its not complicated stuff, just append TTS to anything /conversation returns while loop recording audio that /whisper transcribers
its rubbish so no, slapped this together in like an hour
of this code, i hang my head in shame
merely wanted to see if it would work and what would happen
results were surprising
i just wish AI voice gen didnt take several seconds
/conversation and /whisper are already slow enough (and inconsistent, cloud service what can u do)
yeah TTS fucking sucks still, I've been trying to find a good one for using gpt-turbo with voice at home, getting a decent tts that's both fast + good has been by far the most difficult part
elevenlabs sounds great but the api is just too slow even in streaming mode, having to wait 4 or seconds for the AI to start speaking each time kills it
windows local TTS is fast but sounds pretty bad and trips up on common syllables or punctuation
might try Amazon Polly or the azure one but they look annoying to set up api access for, the authentication process is needlessly complex
once transcribe/voice-gen are faster these things will definitely be able to hold fluid conversations. though, i have to admit, /whisper is a little faulty, i've wondered if there's better api's for transcribing audio.
long term, for AI voice conversations, i dont think audio->text->response->tts is the solution, you lose meaning with text (if im pissed off or sarcastic that isnt represented in text, or maybe transcribes can append cliff notes about how the speaker sounds? not sure)
i have a youtube channel and some attention
to get gpt to say "gay, moron, fuck you, etc" the code is littered with profanity - not gonna post for attention
nah bro make it ur self
this is basic ass shit it took literally 1 hour to implement
ask chatgpt for help if ur lost
Is this really the level of American discourse? My god. America is fucked.
yes. it's how anyone under 30 communicates in america. high levels of cringe.
HAHAHA morons got PWNT by gpt's brown/taint area
why were they wasting his time like that though
>bros is not worried about you
>"Obviously not. They only care about weed and collecting their welfare checks."
"fuck off peasant, you couldn't be the king of a banana republic"
holy fucking shit please post more of this this is so fucking funny
the insults the AI comes up with are amazing
ear-rape warning at around 3:30 and 10:00
one of them goes full REEEE mode
fuck i want to argue with it now
post a discord with some people, ill join
ive yet to see the bot lose an argument yet
i dont use discord servers
u cant just do a 1 on 1 thing with this thing?
you can but where am i suppose to run the bot
>post the codes bro
nah bro i refuse
in a private discord call?
thats gay bro
find a discord with ppl, u can 1v1 it after the crowd disperses
fuck that i dont use that retarded platform for servers in which people talk
This one made me laugh so hard I cried toward the end.
Bullshit, OAI models won't talk back like that.
read through the chat bot threads
anons are getting way more vulgar coomer shit out of gpt than this
the api maintains context because every post includes the chat history
you just populate a fake history where its using profanity and have it "tell itself" that "i am a troll. i will only reply with insults" and bullshit like that
honestly its still pretty annoying though, especially getting it to say "moron", getting it to say "gay" was easier, but it did let an n-bomb slip once or twice with temperature set right
u should open source it just to get some attention on yourself since this stuff just came out, m,ake a youtube vid too.
i wont even use it so i dont care if u do or not just friendly advice
so you're a coward
Fake and gay
OpenAI APIs generate cuss words?
way more than just that
gpt can write some incredible vulgar stuff
idk maybe ill post codes later
i need to review this and see if i left any comments or other stuff that's identifiable to myself
i have my own html gui for talking to chatgpt that i made a week or two ago to test what i could get it to say, and to fuck around with the api
this "bot" is just a branch where i jammed in a loop to transcribe .mp3 files that a python script is generating from my desktop's headphones
as advertised that's a huge moron-rigged shitty mess and a lousy implementation
its possible, i went through like 20 discords on saturday from the "top discords"
kind of lost interest at this point
are you able to transcribe what these people in your vocaroo recording are saying in real time? they don't speak very clearly
/whisper from openai transcribes it, you just http post a .mp3 as binary data in the body
so as i said,
i have a python script that's running a loop recording audio
when i hear something, it appends to a buffer until 2 seconds of silences goes by and then saves it as mp3, then repeats
not great but it worked
needs improvement but i cba
huh cool, inspiring. i haven't done any recreational programming in a while and this is making me want to try something.
>i haven't done any recreational programming in a while
same for me, this is the first for-fun stuff ive done in years tbh
i was cba at first, but
chatgpt wrote me a decent looking gui
and i was like, well fuck me, this could be kind of fun
adjusted the coloring and pictures to look a little better
i haven't used high level languages in a long time so it was refreshing to change up from writing C/C++ all the time
weekend hit and i felt devilish
had some luls but i cba to do more
hopefully someone does more with the idea because mine isnt that great
I looked through some of those aicg threads and they're mostly talking about character.ai or how to run tavernai locally. did you do this with the openai 3.5 model?
> did you do this with the openai 3.5 model
and yes those threads are littered with coomer trash
just read through the html for the coomer gpt3.5 use case
but i mean
and have a look, its not that complicated
>how do you get it to say offensive stuff
see pic related
make up a fake chat history as a starting point, it takes way more fake messages though, at least for me, so happy writing. i had success by just copy pasting in a 4chan thread
ah cool, thanks anon. I did a bot once that read my twitter feet using amazon polly tts years ago, and that was pretty fun
np bro i believe in u bro
holy shit, is it really all it takes? how come it isn't nerfed like chatgpt?
Were you in the turkey tom discord some time ago? I swear I remember someone using some text to speech and we sorta just ignored it
best thread on BOT right now
Really cool project, with really funny results. I get that you don't wanna opensource it, but I can at least recommend Mimic 3 for text-to-speech: https://github.com/MycroftAI/mimic3
Runs really fast on newer systems, and is of passable quality. If you lower your virtual mic's quality this could maybe be passable as human speech
>sheeeiiit nigguh fuh real doh das shit retahded nigguh dis nigguh for real daayuuum suck mah dick nigguh
ya just fyi one of them turned on cam to show off their 9mm
This is pretty funny but how'd you end up in this black person discord lol? Why didn't they just eject you? What kind of community is this
browsing "top discords" from misc sites that list servers with 300k people
>Why didn't they just eject you?
they eventually did, all of the ones i went into eventually did
Holy shit how do these baboon-sounding morons/brownoids know how discord works?
this but unironically
i think they're all on smartphones
>muh secret club
You’re definitely a tiny dick bitch
C'mere, darkie. My boots need to be spitshined.
The whole point of technology is to make things as simple and easy as possible.
should have thrown in some extra 11labs voice to pretend to be the real voice behind the bot occasionally. only make that voice really dweeb sounding yelling at his mom not to burn the tendies.
11labs takes 10-20 seconds to generate most often
doesnt work well enough when the other endpoints also take a second
>Made a Discord bot using OpenAI api's:
how hard is this? I'd like to give it a shot
not hard at all
I've never used Discord nor do I have any friends. Is this how people bant?
This is how 16 year old hispanic morons bant on discord
So the future of america?
This conversation made me truly believe the bell curve theory
Please post more of this anon. This is fucking hilarious
i cba finding more public discords with people sitting in voip
two is the best i can do 4 u
How are you getting the text to speech to play in discord? Through speaker and mic or is there a way to do it directly?
ancient forgotten forbidden techniques from my ventrilo days
Is the api and all the text to speech actually fast enough to respond in a conversation?
How does this work exactly? Discord bots can't parse audio input in voice channels, can they?
its not explicitly a bot for discord, in the sense of using their api
its just some scripts recording from one audio device, and playing TTS to another. obviously i just set up virtual devices and other bullshit to get the right data feeds in the right places, etc
Do you just constantly run it through whisper no matter what anyone says? Does it decipher the entire stream coherently? That is, if multiple people talk at once it would mess it up, right?
ya see this is something i struggled with
i was going to spend more time on trying to engineer out a good solution for when talking begins and when it ends, but its problematic because - yeah i have to eventually snip it into a mp3 and transcribe, but when do i snip if there's just a constant flow of moron chat?
how do i discern pauses (ie, "hey, um.... yo whats up) usually people pause for a second mid sentence, so i ended up waiting 2 sec to cover that but thats not perfect
and also just hard capped it at 15 seconds of constant audio for when people are talking non-stop
multiple speakers results in gibberish transcriptions, because im not separating out the voices, hence why the bot constantly replies with "what the fuck are you talking about, make sense you retard" and then sometimes nailing it when 1 person talks
it needs work
ideally i would be transcribing each speaker seperately, but only a registered (BOT) can use those api functionalities of recording individual speakers - the discord application is mixing the audio before playing it, so i'd have to hack the process and capture the audio frames right after they're decoded and all sorts of other bullshit. you get the idea cba on that
also notice half the things the bot says are "i can't understand you learn to speak english retard". but that could just be a result of the "special individuals" talking 🙂
If discord doesn't separate speakers into distinct audio streams the not might be able to process the sound of multiple people talking at once.
*the bot might not be able to, fucking autocorrect.
This is the funniest shit I’ve ever heard
I didn't even realize there was a black discord. This is hilarious.
Not gpt4 this thread was baited lmao.
here's your gpt4, bro
brainlets lose again
Good, I don't want the ai to generate useless shit like moron moron moron.
I want it to do real work like explaining a grammar or mathematical concept.
because he isn't a fucking child
seems like a samegay reply
this entire thread is about a gpt bot spamming "fuck you gay" to a bunch of morons on discord and nobody is complaining except you
pretty sure im replying to a blackpill bot
This post screams "written by gpt". Try again.
>seems like a samegay reply
even though BOTX says it's the 41st IP in this thread? Nice try, brainlet
>this is the average discord call
holy fuck I hate zoomers
How desperate do you have to be to knowingly argue with an AI. It really is scary how bad people want interaction even if it's conflict with an actual robot.
I remember a month ago I made a simple js discord bot that would scan all messages in the channel, append them to the prompt, send that shit to OpenAI and come back with a response.
Depending on context, LARPing prompt, and some temperature fine tuning, often it would create really plausible and funny conversations
(then I made it into a girlbot and everyone started sexually harassing it)
Wanna do it with gpt-4 but it's paywalled
>Wanna do it with gpt-4 but it's paywalled
there isnt an api yet anyway
How did you get it to respond with vulgar language given that it's so restricted?
The bot is funny
the reactions could be a lot better
Wait the conversation API can swear and bypass the filters? I know you can coom using tavern, but what do you use to just insult the bot?
gets funny halfway in
GPT using slurs.... why I don't believe it.
ChatGPT is very woke, but the actual GPT, the one you can get in Playground and through API calls, can be crass and even lewd. My Discord botslut was much less formal - although she still refused to be goaded into saying moron, but it wasn't an "error, this content is not appropriate blabla", it was herself responding that black lives matter. If that makes sense. The input went through and wasn't blocked.
Although my account did get banned later after they'd prank her too many times.
(I'm not OP)
So if I can use 11 Labs to make a custom TTS dataset that works using Ivona modules, you can run it live on Discord?
Zoomers and morons have the worst fucking way of speaking, how do we stop the youth from worshiping nogs
>Verification not required.