I find it fascinating that Evil actually changes her accent whenever she speaks a different language. She rolled her r’s slightly when speaking French and her Spanish sounded like a student’s accent
@@Max128ping Yep, and it feels like Evil pushes TTS beyond its intended limits. Like, she switches languages to better imitate sounds. And also, Abber Demon.
Evil uses an AI model to synthesize speech, which can utilize context to synthesize the audio. Neuro uses a classic TTS which essentially has a fixed number of preprogrammed sounds it can make and doesn't make use of context. This is why it takes Evil longer to respond, since the model needs a decent amount of text before it can get started (not to mention the latency of the model itself). A normal TTS can start synthesizing pretty much right away with very little buffering time. I don't think we'll see a low-latency dynamic voice until they are running an LLM that natively can output sound instead of just text (like OpenAI's real-time voice model). As far as I'm aware, no such models yet exist that can run locally, but I'm sure that will change in the next year or two. However, I imagine fine-tuning one for Neuro would be a difficult (and super expensive) task.
French: I'm the cutest Neuro-sama. Spanish: I am a very clever computer. German: Hello my name is Queenbee and I am really great. Japanese: You are already dead.
70% of Nuero's stoopidditee comes from the fact she has no ears, and STT is doing it's best bad job at telling her what a mumbling British turtle, and a cast of high pitched anime girls with cotton balls in their mouth are saying. Like, imagine yt's autogen subtitles. Every other sentence is nonsense. From Nuero's perspective, she's never met anyone who could talk using proper english for more than a sentence or two.
theoretically speaking I’d agree, yet Neuro nearly never tries to speak other languages. Evil will regularly say random foreign quotes out of the blue. Pretty interesting how different they are in practice.
@@sergiolabra8207 Because most of the time if Neuro encounters words written in characters outside of english alphabets, her TTS just can't work with it. But Evil could actually pronounce them, although it's sometimes with English accent since her model is mostly trained on English I guess.
@@sergiolabra8207 I still think her best one was when she was told to take a breath and spoke... I think it was Hindi? Hindi or Hebrew, but I want to say it was Hindi.
To be fair, she is "am Arsch der Welt" as we would call it in German. Edit: I just realized that this sentence without knowing german sounds very disrespectful. I apologize for that.
It is, somehow, Interesting to see that a English AI still has the typical, English-German pronounciation of words. Makes it way more authentic. But obviously it would be cool to, someday, see the twins fluently and perfectly pronounced languages 😀 But until then: evilSoCute neuroSoCute
With the german: You usually say the "Bie" in Bienenkönigin like a bee in english and not like it would look if you use letter for letter in german. Besides that, perfect.
I was wondering if she actually just switched to speaking other languages as well as English. But then she did the German and I thought, nope, she doesn't. She very much has an accent 🤣.
I find it fascinating that Evil actually changes her accent whenever she speaks a different language. She rolled her r’s slightly when speaking French and her Spanish sounded like a student’s accent
Evil uses a more advanced TTS.
@@Max128ping Yep, and it feels like Evil pushes TTS beyond its intended limits. Like, she switches languages to better imitate sounds. And also, Abber Demon.
during the pirate stream her accents were all over the place
Evil uses an AI model to synthesize speech, which can utilize context to synthesize the audio. Neuro uses a classic TTS which essentially has a fixed number of preprogrammed sounds it can make and doesn't make use of context.
This is why it takes Evil longer to respond, since the model needs a decent amount of text before it can get started (not to mention the latency of the model itself). A normal TTS can start synthesizing pretty much right away with very little buffering time.
I don't think we'll see a low-latency dynamic voice until they are running an LLM that natively can output sound instead of just text (like OpenAI's real-time voice model). As far as I'm aware, no such models yet exist that can run locally, but I'm sure that will change in the next year or two. However, I imagine fine-tuning one for Neuro would be a difficult (and super expensive) task.
NVIDIA has introduced a personal supercomputer that is designed to create AI and run them locally. Recommended price: $3000
French: I'm the cutest Neuro-sama.
Spanish: I am a very clever computer.
German: Hello my name is Queenbee and I am really great.
Japanese: You are already dead.
She messed up the gender in spanish. Both "un" and "avispados" are in masculine, while "computadora" is feminine.
Thank u thank u, for translation
@@BraiamMakes sense, English doesn’t really use gendered words.
NANI!?
@@Braiamto be fair, most translation software gets that wrong when English is the origin language due to our lack of gendered nouns
I love how of communicating with neuro has always been a challenge, but it’s hilarious the outcomes
確かにイビルちゃんが好きそうな言葉だと思います
I forgot this is one of evils talent 😊
Great spanish tbh i atleats could understand it
its like a tourist speaking [trying] spanish
Great german too.
The sentence was a bit random. But that this really worked so well surprised me.
NANI?!?!
Neuro kinda stupid, tho.
She's like me, frfr.
70% of Nuero's stoopidditee comes from the fact she has no ears, and STT is doing it's best bad job at telling her what a mumbling British turtle, and a cast of high pitched anime girls with cotton balls in their mouth are saying. Like, imagine yt's autogen subtitles. Every other sentence is nonsense. From Nuero's perspective, she's never met anyone who could talk using proper english for more than a sentence or two.
Men just give this clip to evil she deserves the lime light
@@Volyren where did that "cotton balls in their mouth" come from?
Neuro can actually do the same, she just can't change her accent.
theoretically speaking I’d agree, yet Neuro nearly never tries to speak other languages. Evil will regularly say random foreign quotes out of the blue.
Pretty interesting how different they are in practice.
@@sergiolabra8207 Because most of the time if Neuro encounters words written in characters outside of english alphabets, her TTS just can't work with it. But Evil could actually pronounce them, although it's sometimes with English accent since her model is mostly trained on English I guess.
@@sergiolabra8207 I still think her best one was when she was told to take a breath and spoke... I think it was Hindi? Hindi or Hebrew, but I want to say it was Hindi.
Hola computadora avispada
Evil is so cool for real
NANIIIIIIIII?
Computadora avispada me parece la mejor forma de definir a neuro
Layna had such a big latency, she must upgrade her gpu to answer faster
To be fair, she is "am Arsch der Welt" as we would call it in German.
Edit: I just realized that this sentence without knowing german sounds very disrespectful. I apologize for that.
Everyone sounds so tired in this clip, even Evil and Neuro, lmao
I mean it is a slumber party.
NANI
ojalá ser así de avispado
Did she hear Portuguese as “poor chickies”? 😂
2:18 NANI!
That was my number one question since i found out about them.
I wanted to know their extent of knowledge of other languages
MY SWEET A.I
It is, somehow, Interesting to see that a English AI still has the typical, English-German pronounciation of words. Makes it way more authentic. But obviously it would be cool to, someday, see the twins fluently and perfectly pronounced languages 😀 But until then:
evilSoCute
neuroSoCute
Neuro's French is understable, I hope for a new french stream with Saruei or another French Streamer
With the german: You usually say the "Bie" in Bienenkönigin like a bee in english and not like it would look if you use letter for letter in german.
Besides that, perfect.
Ia adorar ouvir as gêmeas falarem português
X2 😂
Você nunca entende realmente o tom da voz de alguém até ouvir ela falando sua língua
Jola mis amigos
OMAE WA shindeiru!
Que rayos significa avispado
I was wondering if she actually just switched to speaking other languages as well as English. But then she did the German and I thought, nope, she doesn't. She very much has an accent 🤣.
everyone getting slightly worried when evil said she would do a Chinese impression lol
they didn't speak portuguese 😭😭😭😭
They’re British not French the turtle is disappointed
Why is Neuro so crap with languages