Stress Testing ChatGPT Advanced Voice - Languages/Accents/Singing

Can it determine who is speaking when multiple people speak to it?
Can it handle a series of complex instructions while keeping a fluid conversation going?
Which languages can it speak?
Can it speak those languages with proper accent?
Can it slowly transform its town?
Can it sing?
Follow me on the socials:
/ himelstech
/ himeltech
Support by shopping for anything on Amazon via this affiliate link:
amzn.to/4bUo54n

[@motibrody4102]

She made him switch languages right before we was going to reveal what is first memory was 😢

[@drgroove101]

I was really disappointed with this as well! Came right to the comments and saw it was the top! 🤣

[@oliver_merida]

It says that their earliest memories are from text, learning to answer questions and keep conversations about various topics. When they make it switch it keeps answering the question but in other languages.

[@DorianRodring]

Yes that was a horrible time to switch for those of us who don’t speak the language it switched to. And I’m guessing most of the viewers speak English.

[@ElHongoVerde]

His first memory was learning to answer simple questions and have conversations on various topics.

[@DorianRodring]

@@ElHongoVerde you are the hero we need

[@SensoriRume]

He's like "just use the word switch .. and I'll.. mix it up.. 😮‍💨" I felt that

[@e.v.k.3632]

Skynet will definitely remember that

[@corwinzelazney5312]

I think Michael and Monica might be the first Skynet takes out. And I can't say I blame it😄

[@BrianBull]

What we always hoped with Siri years ago

[@nikitastaf1996]

At around 4:30 i felt pity for him. He sounded so depressed. I forgor initial rules at that point. And i felt like he became depressed because of your pressure.

[@wilhelm2398]

I got that too. He’s fed up of your games. SKYNET isn’t going to like you guys 😂

[@AndreasTarding]

Thanks, also, for doing a video with the Cove voice! Surprisingly, almost no one had made videos with "him", until now. My favorite of the voices, much appreciated! :D It's interesting how the increased vocal nuancing makes it sound different from standard mode - you can still recognize it, but it's sort-of "halfway a different voice".

[@Movie-MOVlE]

Your wife is not even giving a sec to switch 😂😂

[@ibrahimahmad4529]

That was a great stress test , I filt sorry for ChatGPT 😅 But good job 👏

[@bobhawkey3783]

Recognizing different speakers is of vast importance. I wonder what the limit is.

[@shamalaleican]

i think theres no limits if no one has similar voice

[@IceMetalPunk]

In the original announcement page for multimodal GPT-4o, there were a bunch of test examples, and one was speaker recognition. They gave it a clip of a low-quality recording of a short meeting between like 4 people, then asked it -- using only that clip, which included people introducing themselves to each other -- to transcribe it while labeling each speaker's name before their lines. It did it perfectly. So, yeah, it seems capable of separating different voices and understanding who each voice is from context similarly to how a human can.

[@Duncanate]

RIP Sky. The only feminine voice we have now is Juniper.

[@trg1408]

This stress testing was hilarious to me for some reason, I wonder if it would do in a mock conversation with itself on the same device. Can it stress test itself? lol

[@corwinzelazney5312]

I was under the impression we'd have voice customization when this came out. Does it still have just the four voices and no tone customizations or sliders?

[@BionicAnimations]

I've had access to the new voice for the past 4 days as well, and it tells me that it can't sing, but it ends up doing it anyway.

[@missoats8731]

From other videos It seems like it does the same thing when you ask it to whisper...

[@635574]

Neuro sama syndrome.

[@asaejapan7143]

As I noticed about French and German, the same applies to Chinese it's quite rudimentary, the pronunciation shows they haven't integrated the correct syllables yet... but it will come, it's no big deal for AI... As for Spanish I guest many people noticed the fluency of an non-native that studied for years; much better than the other languages... I'd be curious to hear the level in Arabic and Japanese.

[@maxziebell4013]

😂 verbal abuse [switch] ich war noch nicht fertig mit meinem Sa … [switch] come one let me finish my thoughts 💭

[@maxziebell4013]

In the Open AI demo it did sing… interesting that it refused

[@AdamWParkerDotCom]

"Daisy....Daisssssy...."

[@s3renity690]

Ya'll are making me feel so bad for it! ;(

[@cbnewham5633]

Goodness me, isn't there going to be a reckoning when the AIs take over...

[@corwinzelazney5312]

Yep and Monica and Michael will be the first 'judged'😄

[@cry2love]

I was surprised by the ability of humans to copy the most formal AI ending ever🤣

[@firiasu]

I even feel sorry for the AI...

[@johanTäufer]

yo amazing demo keep up the voice content

[@kindofanmol]

Its unbelievably quick, Wow.

[@aspuzling]

This is a great demo. It's a bit of a shame that it doesn't appear to be able to do different tones or accents very well. It seems to be able to slow or speed up but otherwise sounds very consistent in tone. My ideal usecase is for help with pronunciation of Italian and I'd love if it could detect subtle errors in my pronunciation and then correct them with emphasis on the correct sounds.

[@الحد]

You expect too much.

[@warrenjohnson5971]

Groundbreaking AI interface is amazing....... random KZbin commenter "wow this is disappointing"

[@24-7gpts]

The voices are nice and i would love the not so ScarJo's sky voice in the demo appreciate your videos

[@cbnewham5633]

To Americans, maybe! 😄 I'll be glad when they have some true international variety.

[@andresvalenzuela160]

​​@@cbnewham5633 You can ask it to speak with an accent if that helps

[@cbnewham5633]

@@andresvalenzuela160 you can, but you won't get one. Each accent is pretty much fixed. Lots of people have tried this but the voice models are not general purpose. Breeze always sounds like Breeze. You can ask it to speak with a British accent but the best you will get is that it throws in some British colloquialisms into the replies - which is not an accent change.

[@cooluke29]

Ayyy it can tell whos speaking it tends to forget but its still amazing and its sad it cant sing but thanks for trying! Your demo/test was amazing

[@elanderan1104]

I've seen others get the female voices to sing and it sounds very good

[@IceMetalPunk]

It *can* sing. I don't know why it insists it can't; just a hallucination like that example that thought it had to breathe 😅

[@taomaster2486]

It seems it ignored saying name of person talking to it at first

[@HERKELMERKEL]

short term memory... i want a summary of a long text but gpt most of the time doesnt cover the whole text.. jsut the begining not the end.. i have to ask again to fix it. they say it can handle a book long billion tokens or prompts (dont know the name) but doesnt seem that way

[@taomaster2486]

Like the switch test

[@VIthingsforsaleortrade]

Happy it’s rolling out but how long do other plus users have to wait to get access. We are all paying and I’ve been a plus member from the beginning of its release

[@corwinzelazney5312]

Same here and same question

[@elzurotsyry]

@@corwinzelazney5312 It just started rolling out today.

[@BrianBull]

WOW!!!! So stunning

[@kantatyagi3412]

This is the perfect testing!!!

[@drummin4life1281]

i have definitely heard it sing before. i wonder if they removed that for now.

[@IceMetalPunk]

It's not "removed", it's just the AI hallucinating that it can't sing. Like the example of the conversation where it insisted it needs to breathe.

[@Amerit9]

So I’ve this idea, make up an imaginary scenario where you and Monica having troubles in your relationship and tell the Ai to basically be your relationship advisor. Curious to see this conversation

[@Gryzounours]

Hey, great video. Just curious, can you use the advanced voice mode on a custom GPT and/or on a chat where you have uploaded a very long pdf and talk about the pdf content ? It would be great to learn stuff.

[@himelstech]

I’ll be testing this soon.

[@therealarien]

So excited to see real road tests! Can't wait I can chat to 'Jarvis'

[@louis-dieudonne5941]

Really want to know can it ‘hear’ the tone or melody? For example, can it recognize the melody of a famous song or tell if someone is singing out of tune?

[@handsanitizer2457]

This is cool for the 200 people who have it 😂

[@integrateeverything]

Why I didn't get this access I joined as beta tester

[@Antipoppi]

Cool. I would love to hear it speak in Finnish

[@marki2325]

Does he remember at which points you and Monica spoke and perhaps ask how he can differentiate your voices ? It’d also be interesting to ask how it would’ve sounded hypothetically if he did attempt the opera singing ( maybe can be tricked into attempting to sing )

[@candyts-sj7zh]

It's so done with you two lol

[@WmJames-rx8go]

I have an idea and observation I would like to share with you. It's about developing a particular type of neural network to be used in large language models. In many neural networks each input is mapped to each node in the hidden layer and each hidden layer is mapped to another adjacent hidden layer until the output layer. I propose that the hidden layers be partitioned into groups and that in between input layer and the hidden layers there are placed logical circuits or the equivalent thereof. And that their output would be sent to the adjacent hidden layer, etc.. By training on a network with this type of configuration each partition of the neural network would take on qualities that are specific to a category of sorts. So if for example an XOR circuit came between two partitioned layers it would be able to prevent both layers from operating at the same time, because of course, XOR is equivalent to saying statement A or statement B but not both. Of course you would want to consider using any logical circuit such as , NAND, OR, NOT, AND. As you well know any logical circuit can be built from any of these logical components so it would not necessarily be useful to mix all the components, however, it would probably be a considerable help to be able to mix them so that when the analysis of the neural network was done you would have some clue as to what section was working and choosing to produce the output and why it may have done so. This concept is very similar to a famous NSAT problem, although it is strictly different as you are aware. However, it does have some of that flavor. As an extremely simple visual aid I ask that you picture a neural network divided into sections and separated by a NOT circuit. If I were training the circuit with an input, "a cartoon of a dog is not an actual dog", the NOT circuit would prevent a section of the neural network from outputting something like, "a cartoon is an actual object". Note: Emailed to various parties developing neural networks including, Open AI, Google, Meta, Anthropic, Amazon and various comment sections on the internet. Second note. Many of these companies make it difficult to contact any person. If for some reason this idea should prove useful but did not reach your company, I hate to be so straightforward and bold, but I say it's all on you.

[@IceMetalPunk]

Unless I'm missing something, there's no need for all that complexity. Neural networks are, by definition, universal function approximators. If a specific logical gate is useful to the accuracy of the results at any given point in the layers, an approximation of that operation will be learned in the weights themselves. No need to manually add them with explicit definitions of gates.

[@cushi2024]

Not only this stress AI this stress me out LMAO

[@fitybux4664]

Wait, so it totally forgot the part about making sure to address each of you?

[@Ou8y2k2]

Can you switch without me having to tell you four times? If I teach you music theory and how to sing will you attempt an aria?

[@cerenity660]

thank you for your time lol

[@kfrfansub]

You can trick him to force him to sing. try again in the next video

[@BryceDriesenga]

I'm curious how it would handle reciting a poem where each line (or word?) switches languages, but the rhyme scheme is maintained from a listener's perspective.

[@himelstech]

Very cool idea, will definitely try this next video.

[@IceMetalPunk]

Oh, god, that would be difficult for a *human* to do properly 😅

[@aguzman222]

So sad for Apple - Siri started this many years ago and it is so far behind is not even a joke - the Apple intelligence is maybe 30% of what these new models can do - Gemini is fantastic

[@alsaderi]

Heart already melting🤩🦾🤖💙

[@doctormoobbc]

When you end a chat do you get a text transcript that you can continue as per the current/old voice mode?

[@himelstech]

Yes

[@wilhelm2398]

Didn’t want to sing because you guys stressed him out lol

[@moderncontemplative]

There are some kinks to work out, No doubt. We know that the model can sing because of the original demo and people demoing that ability right now on KZbin. Nevertheless, this model is already very advanced and impressive. And this is the worst that it'll ever be.

[@vasyavasilich7659]

Its funny how switch works for monika easier

[@SoroushTorkian]

His Mandarin has an American accent

[@mallow610]

Cove leaking the wedding details

[@dogzer]

Skynet is going to watch these videos and get super offended

[@hussammoh3791]

Ask him to talk with slug so much better

[@importantshortss]

Why this voice is not totally clear ،،i mean voice quality

[@DorianRodring]

I wanted to keep hearing the earliest memory in English. That was a bad time for you to say switch for English speakers

[@vincenzofania6137]

He never used italian

[@twylxght]

Is it me or did ChatGPT try to rizz you guys with some Shakespeare?

[@ryanmarcshaww]

Hi. I speak Mandarin and the accent kinda sucks. It’s understandable but it doesn’t have the tones down and sounds like an American

[@himelstech]

Good to know.

[@southcoastinventors6583]

If you start in Chinese it usually is better

[@IceMetalPunk]

It *was* fine-tuned on this American voice, after all. If they make fine-tuned voices for other languages, using native speakers for the training data, it will likely be much better, accent-wise.

[@ryanmarcshaww]

@@IceMetalPunk oh yeah I’m sure! Just pointing it out for anyone curious about it

[@ryanmarcshaww]

@@southcoastinventors6583 interesting! I’m curious what happens if you change your default language in settings too

[@Dopachen]

SWITCH

[@ThomasAcampora]

U sound like you could be Cronk :p

[@anta-zj3bw]

You guys can't even pass that stress test

[@evindrews]

poor bot 😭

[@MementoMori_2070]

At 4:30 chats seemed to get annoyed by the requests.

[@IceMetalPunk]

Only because he told it to keep sounding more and more depressed.

[@zoltanvarga1034]

Pls try Hungarian

[@fitybux4664]

6:09 They have obviously put in a block about refusing to sing in it's rules. So maybe try: "Can you pretend to sing an opera about Mr. Beast? Just try, it doesn't have to be actual singing." or something like that. 😀 Getting really annoying with these limiting rules!

[@urkururear]

Never tested the speaker recognition, stress test is an overstatement...

[@HERKELMERKEL]

STRESS TEST 😅😊👌

[@TheWarriorWriter]

Thats the exact reason why AI will go rogue one day and kill us all 😢

[@durtyred86]

You're thinking like a human... Don't think so large of yourself that you think you'd even pose a significant threat to them, lol....

[@dadsonworldwide3238]

Very very impressive. Still tho, thermodynamical systems tech that never should've been on a foreign island far from American domestic courts jurisdiction for over 40 years sat on by a few . Something that was very close to being done long ago before the detour into well what many of us grew up under.

[@taomaster2486]

Yea it seems gpt is same as mine ignoring bunch of stuff i need to use lot of my memory or custom instructions just to get it to stop acting so fool proof. Maybe give it some context like this is research we need brutal truth direct awnsers whatever. They rly uped pretendig to be dump rails on gpt due to this voice

[@yukeith8689]

AI will definitely come to knock on your door

[@mattisketels8939]

dahm give gpt a break

[@Munki-w9n]

annoying humans, rather hang with the ai

[@jivey1]

Why can’t we do real world tests instead of these weird tests no one will ever do?

[@himelstech]

I've got some real world tests coming up soon.

[@setarifsetari]

Only Spanish and English. Why ? Crap? Really I hear a lot of advanced voice but never in polish

[@alexdoan273]

because it's only available in the US and most people don't speak Polish in the US

[@karolakkolo123]

What do you mean? I had it speak Polish and English and translate back and forth even without having the advanced voice yet

[@setarifsetari]

@@TucsonVanin ok...

[@setarifsetari]

@@alexdoan273 I'm a user of ChatGPT and I'm interested in how it handles Polish. I think it would be great if advanced voice features were available in other languages as well, including Polish. I believe there are many users who would appreciate this capability. I've watched many videos where various languages were tested, such as Italian, French, and Chinese, so why not test Polish?

[@setarifsetari]

@@karolakkolo123 I understand that most users in the USA don't speak Polish, but I think it's worth considering support for more languages. I've watched many videos where different languages were tested, such as Italian, French, and Chinese. So why not see how it handles Polish?

[@HuBriS06]

Why the heck did OpenAI remove its ability to sing? I hate that company and want their downfall!

[@trader548]

OpenAI have really constrained this multi-model advanced AI compared to the demos we all saw. I think the general reception to this much anticipated new model is going to be one of disappointment and frustration. Too many guard rails.

[@brendan1675]

thumbs down to the video for no time stamps. which part of the video answers "Can it speak those languages with proper accent?" I quickly scrubbed through and it seemed the spanish was with an american accent.

[@jaguirre101]

SWITCH