![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png)
kossai is not very much fan of current uses for " AI " generations as means to replace someone - whether as artist of picture , as replacement to not pay voice actor for request ( which some do offer ! ) , or otherwise .
but with audio in particular , think this is stickier than just " need explicit knowledge and permission from voice provider " - any enforcement of such rules can extend much further than just AI models , and do damage that would hurt everyone .
inability to sample and mashup musics on one end , and on other end , even further restriction on people who intend to review , document , and archive very important information .
( not to mention that companies who already loophole and disobey laws would simply continue to regardless ... )
what concern kossai much more is when these things sound very real - sentence mixes a la youtube poops and jinriki UTAUloids just do not have this realism factor ... but well put together AI models can , at least enough to fool average person .
that in mind ... do have weak spot : sometimes will stumble across character AI model , and wonder exactly how that voice come through .
today ? bill cipher .
and answer here is : poorly . these models often do not carry across vocal filters very well - as if remove " noise " - and tone is often wrong , with voice cracks to boot .
well , one cover is rick astley greatest hit , dQw4w9WgXcQ .
as with others , voice struggle to sound quite like bill .
but here is where comedy come in : auto captions . youtube auto captions tend to struggle with music , which lead to lines like :
but with audio in particular , think this is stickier than just " need explicit knowledge and permission from voice provider " - any enforcement of such rules can extend much further than just AI models , and do damage that would hurt everyone .
inability to sample and mashup musics on one end , and on other end , even further restriction on people who intend to review , document , and archive very important information .
( not to mention that companies who already loophole and disobey laws would simply continue to regardless ... )
what concern kossai much more is when these things sound very real - sentence mixes a la youtube poops and jinriki UTAUloids just do not have this realism factor ... but well put together AI models can , at least enough to fool average person .
that in mind ... do have weak spot : sometimes will stumble across character AI model , and wonder exactly how that voice come through .
today ? bill cipher .
and answer here is : poorly . these models often do not carry across vocal filters very well - as if remove " noise " - and tone is often wrong , with voice cracks to boot .
well , one cover is rick astley greatest hit , dQw4w9WgXcQ .
as with others , voice struggle to sound quite like bill .
but here is where comedy come in : auto captions . youtube auto captions tend to struggle with music , which lead to lines like :
- i'm going to get you up and i'm going to let you down
- i'm going to make you cry, i'm going to say goodbye
- i'm going to share the light and gra you
- i go around around around
- i'm going to get you
no subject
Date: Sep. 30th, 2024 01:23 pm (UTC)I laughed out loud when I read those auto-captions. Truly incredible.
Someone needs to make a Bill Cipher cover of Never Gonna Give You Up where he actually sings this. XD
no subject
Date: Sep. 30th, 2024 04:57 pm (UTC)