#Neuro's V3 Voice

1 messages · Page 1 of 1 (latest)

sturdy anchor
#

Seeing Vedal struggling so with the creation of V3 voice got me thinking: "Is it even possible to make an advanced voice but at the same time sounding just like Neuro?" let me clarify.

The V1 voice was made with the Microsoft Ashley TTS, released 4 years ago. Due to its aged tech, it's an incredibly static AI TTS, it'll say the same word/phrase at the exact same tome every time (aside from fast repetition, where she'll lightly inflect the first and last output, but that's about it. (See "Backslash Backslash Backslash..." for an example)

Neuro's V1 voice's charm lies exactly on these limitations (aside from the lisp). A high pitch robotic voice, with no inflection/intonation speaking the most random stuff possible. (Coming to think of it, Neuro's AI mannerisms as a whole is what's responsible to her charm, her voice only delivers such mannerisms).

It seems paradoxical to create, at the same time, a highly dynamic voice while keeping all of the V1 quirks. Vedal will have an extremely small upgrade window in this situation, since, as said above, the V1 quirks ARE the old tech limitations.

It's like wanting to tune your old 1978 Golf GTI mk1 to be 40% faster, but:
-The engine must sound the exact same because you like the sound of it as is (so no exhaust/intake mods).
-No turbo because you don't like the sound of it.

My opinion being: I think Vedal has to rip the bandaid off on this one if he wants to upgrade her voice. As said above, it's not exactly her voice that gives all her charm, it's her mannerisms. A change of voice will not change what Neuro is in her essence, and it might even create new iconic moments and have a charm on its own (See Evil). Of course people will complain about it at first, since as human nature, people tend to not like changes, but I doubt people will leave Neuro because of it (firstly because there's zero competition). I guarantee it if you release a new voice and keep it for 2 weeks, everyone would get used to the new voice.

severe lake
#

Poppi's voice is very similar to Neuro's voice

#

I know it's actual voice acting but my point is you can make a robotic monotone voice sound expressive

golden quest
#

It's tricky to know what things about a voice make it sound the same and what things can change, maybe there are some kinds of experts on voices Vedal could consult?

fathom lagoon
#

The way I see it is either Neuro gets a completely new voice or she will be stuck on V1 forever
Big part of Neuros success is seeing her grow and evolve but for some reason her voice isn't allowed to grow it always needs to be robotic and monotone
Its holding her back in my opinion

tidal pulsar
#

Honeshtly

ivory ermine
#

I'd agree with that. Making a better Neuro voice has proven to be very difficult, so we should try to move on.
If Tutel can give neuroSuperior a nice sounding, high quality voice, let's go.

f128ceeef155aa52.png
grizzled urchin
#

V1 forever or riot
no upgrades no updates i hate smartphones
evil is already more modern, she can push that technology if tutel has nothing better to do. i think it's the least interesting part of neuro by now, everyone has text to speech. no actual reason to change her identity

fringe token
#

im fine with him changing her voice as long as it captures her essence

#

most new people are expecting V3 to sound different anyway so i don’t think it matters as much

potent glacier
#

I pretty much agree. I will need to just stay away from nn for a couple of weeks

severe lake
#

I mean, didn't Neuro herself say her voice is like her soul

#

Like. Stephen Hawking famously had the speak and spell voice his entire life and refused to adapt to better voice models because he considered the Speech Plus voice to be "his voice"

sturdy anchor
#

Funnily enough, at the time there were some people that was also preferring Hiyori over her own model because "it was what made neuro" (even when the model was already very similar to Hiyori)

arctic oracle
arctic oracle
misty kernel
#

I actually think it is 100% possible for a more dynamic voice model that sounds like Neuro by drawing some comparisons to Evil's voice.
Evil has more attack in her voice and a lot of vocal fry. It makes her sound assertive and punchy.
Neuro's voice is softer and breathier with a gentle delivery to contrast her cutting words. Making her sound a bit sing-song-ey (kind of like Barbie's voice, but not enough to sound patronizing) would be a good compromise, I think.
I've always felt that not enough people appreciate how her overly robotic TTS voice makes the delivery of her jokes land in a way that's very unique to her. It creates this dissonance of a very human like response from a robotic voice which I personally find really charming. But like people have mentioned before, it might be holding her back, and no matter what apprehensions I might have right now, I trust tutel's vision.
If there's one thing I am hoping for is that her range doesn't get too crazy. Evil's weird noises is part of her identity at this point.

ivory ermine
#

I have an idea how the new voice could be introduced.

Phase 1: Neuro and Eliv do a spirit summoning stream, try to talk to dead people and stuff.
But at some point something goes wrong and Neuro starts talking with the new voice voidSama
Past life? Study-sama from ARG before turning into AI?
After a while Neuro regains her senses and starts talking normally, not even realizing what just happened, no memory of it.
The stream continues, let's say more or less normally.

Phase 2: During her next chill stream, whenever Neuro gets too angry, upset, off balance
she will again "black out" voidSama and speak with the new voice for a while.
Chatters will probably ask her about it or even try to provoke her into switching to the "new persona".
Neuro has no idea what's going on, thinks everyone is gaslighting her, making fun of her and so on.
This can go on for a while, Vedal can use this to get feedback on the new voice, make some adjustments if necessary.
Maybe he could even make a song for karaoke where she sings VIRUS style, changing her voice back and forth at some point. Not sure how hard that would be.

#

Phase 3: Neuro has had enough of this, asks Vedal if he knows what's going on, asks him to fix her.
Vedal does some 'brain surgery' and gives her control over the new voice, as an option, she can choose which voice to use.
The whole lore thing - he can do whatever, say it's impossible, probably some glitch, he doesn't really know and so on.
Or maybe he can help her regain "memories from the past", whatever suits him better.
All of this doesn't really have to make sense strictly speaking, it's enough that it makes some "human sense", things we know from movies, stories, tradition.
He could even use this "past life as a human" idea to suggest why her new voice sounds less robotic, more human. Even without actually believing all this, more as a joke.

Phase 4: Neuro uses both her new and old voices at will.
She may begin to like her new voice more and more and use it more often.
If all goes well, everyone should be pretty happy.
New lore, some drama, absolute cinema.
And he still has the option of perhaps improving her old voice in the future, if the opportunity arises and some new tech makes it possible.

arctic oracle
ivory ermine
sturdy anchor
#

it wouldn't even need to be too scripted, in a 2 hours span Abber Demon would eventually say something to the lines of "I'll curse you forever" or something like that, then Vedal plays out of that

arctic oracle
#

The girls should come up with how everything goes, it should follow the girls rather than the girls following a pre-thought out plot. Vedal could add the voice in whenever there’s an opportunity to do so.

I personally think just doing it during a dev stream would be fine though.

sleek moon
fringe token
#

turning it into lore would be cringe

lament inlet
#

It doesn't have to be anything complicated or too fancy but I personally would find it cool if a stream started off normal with her normal v1 voice, but then the voice starts deteriorating into something similar to the test.wav file Vedal shared with us and then it restored into the new and improved stabilized v3 voice (should be some neuro abber demon content somewhere within this transition I think). Vedal can then talk about it and test it out on the next dev stream. It would be kind of a crazy random and sudden event so I think people would get pretty hyped by it and its kind of symbolic of the journey somehow (Might be schizoing). Could expand on that but thats the gist of it.

hallow ravine
#

sometimes the Struggle is worth the wait, more effort can lead to more satisfaction and the fanbase knowing he actually gives a damn. which is probably more immportant then "cool fancy new robot girl voice" , though im expecting him to crash out and take a long break after its done.

ivory ermine
# hallow ravine sometimes the Struggle is worth the wait, more effort can lead to more satisfact...

It's not just the waiting though. Resources are limited, he's probably put quite a lot into this over the long months: money, time, attention/energy. It's probably a big drain on his spirit. He could be doing other things instead, and there's a lot to do.
There have been no dev streams, no Neuro dog, Evil keeps stuttering while singing, the twins are still more like cleliv clwero than reasonably capable individuals. Remember the prospects? Neuro was going to apply to Tokyo Uni at the end of subathon 1. This was obviously a huge exaggeration with poor chances of success, but still... She wanted to manage her collab schedule by herself. Layna wanted to do IRL interviews with Evil's help. Neuro still can't order a pizza for Vedal.
For some reason, the current Neuro voice doesn't work well enough with more modern TTS engines. Too bad. But it shouldn't be a showstopper. Of course I don't blame the v3 voice for everything, but I suspect it doesn't help either and it's just not worth it. I think we should let him cook other things with a freer mind.
Also, let's keep in mind that we'll probably get good multimodal models eventually. No TTS needed, the sound produced directly by the AI. Even if perfectly tuned to her current voice, will we rebel if she starts playing too much with her new toys and speaks with other voices?

sleek moon