Accents make people more unique and interesting and there are a lot of lovely accents. They help to suggest someone’s culture, where they originate from and background.
Some observations related to English and Accents.
- The Scottish (English) accent is ok but could be better. The Scotts are bound not be impressed. It sounds like a Londoner trying to sound Scottish having only heard Scotts on TV. Oh dear. NB Scottish accents have at least 3 main flavours (SE, SW and Highlands)
- Welsh - oh dear!
- Newcastle English sounds like Scottish - rubbish! It claims that it can do it.
- It does a reasonable Yorkshire accent but could be improved! Sean Bean would not be impressed.
- it is unable to speak English with a Chinese accent at all even though it claims to be able to do it.
- German English is terrible but at least it’s clearly trying
(terrible as in its attempt, English spoken by a German sounds lovely).
- I gave up at this point and will wait for version 1.5.
I’m sure you can do a similar test for variations across the US or Australasia. Or any region or country for that matter and all our many languages.
Some background listening for Open AI researchers:
Some iteration needed to get accents up to a decent level, but it’s an interesting start.
I also like the idea of regional accents. I mean, they are everywhere.
Off the top of my head, in the US, there is the standard “midwestern” or neutral accent. But even then the upper Midwest, you start sounding like Fargo. Then there’s the New York accent, Boston accent, and Southern accents (and variations in there too).
Controlling the regional accent would be a cool feature. In AVM, you probably can prompt it, as I can say, say X to me in English with a German accent, and it will do it and sound convincing.
So maybe try the regional variations in Advanced Voice Mode by requesting it, and see if it works!
1 Like
I’m very used to hearing Germans speak English, and I’m not convinced … (I’m part of the Star Citizen community and there are loads of Germans who love it just like the Brits (and Americans!) and are kind enough to talk to us in English).
Yes, I have been prompting it. Perhaps what it is trying to do is make the current speaker speak as if they were trying to speak the accent from the prompted region. What we need is more voices from genuine local speakers.
Let’s have Sean Bean, some great rugby commentator from Wales … a Scottish James Bond … so many options!
Are you sure you really are in AVM?
The only way I know you can tell, is if the “dot” is has the blue/white swirlies. But if the dot is black, then you are in normal voice mode, and there is not much flexibility there.
Yep!
It’s the new blue watery thing.
And the new kick-off icon.
Maybe I’m just very sensitive and choosy? I am also a musician with a critical ear.
No, it’s probably not just you.
I have noticed that it “breaks accent” after just one turn. And if you have it rotate accents in a session, it will randomly morph between the accents, which is weird.
So giving it a fixed regional accent, at the “voice selection” level should fix this. But it’s long term regional accent retention just isn’t there. I mean, I could get it to persist by prefacing everything with “speak with X accent: “ but this gets old.
Also, there is a big vocabulary shift too when changing accents … because the accent is only half of the persona, the other half is what is being said.
For example, a southern accent might say:
“How y’all doin’?”
But a neutral accent would say:
“How are you doing?”
So to really pull off the accent, the model would have to go through a fine-tune of sorts, developed for that accent, that would also switch or alter the words according to that region.
So a lot of improvements need to occur for it to really work. This would require the accent and a morphing of the words. So it’s not entirely trivial.
I’m thinking they just have the initial “voice” part down, but not the words? Maybe I will have to test different accents to see if the words are slightly altered for that accent. If so, that’s good news, since it means everything is in-place to properly regionalize.
1 Like
Yes, completely.
Vocabulary is very key too!
1 Like
It also does quite a good London road man accent!
1 Like
