audio splices , jinriki , and machine generation

[kossai]

when think about , really ... wonder why not see people have so much problem with things like sentence mix and jinriki voicebanks , when do hate machine generate voices .

one big problem with machine generate that people say is almost always lack permission ... but then , true with sentence mix and jinriki too . but . sentence mix and jinriki have skill that people can understand - to splice , mix , smooth transitions , understand language to get best sound within limits of source audio .
these things also have limits to use and presentation , which means not likely to mistake for real . even best efforts with either will have audio artifacts … spongebob jinriki voicebank can sing party junkie , but doubt anyone will think tom kenny is out there sing vulgar japanese songs .

still possible to use these for bad and cruel things , but lots of effort and skill to get there , let alone pass off as real .
machine generate unfortunately make this part much easier - can directly feed in real reference audio and work from that , warp ( in theory ) any voice to fit . think that is why more people take issue with lack permission than with splice methods . of course , some people still do - but never see near as much storm .

another aspect - splice can never replace that real person … machine generate , in some contexts , unfortunately could . good quality splice can at times inspire person to do for real - good quality machine generate remove purpose of this .

not fully sure how to navigate ... sentence mix is most often comedic tool , and have big part in things like youtube poops . jinriki voicebanks exist to fulfill silly thoughts of , " this song is perfect for character , and need to hear sing ! "
is that just as wrong as machine generate voices , for lack of permission ? is this less wrong , because can not pass for real thing to spread cruelty , lies , or replace ?

could also argue that unfortunately , any enforcement against this for individuals could hamper any other creation that rely on sample , reference , and use other people's audio - and permission can be tricky thing to prove , when also possible to make fake proof . so feel … not sure there is clean answer .

as technology , do think machine generation impressive . however do not prefer - especially because of relative ease to misuse .
kossai admire sentence mix and jinriki , and work with these for quite some time now , though mostly for private amusement . suppose this robotic quality and artifacts give odd comfort in this , that if work somehow did leak , would never be possible to mistake for real thing .

so , right now ... that is about where stand - nice as would be to have permission at times and know person approve , do not know would help overall to enforce this . permission is of less concern to kossai than misuse and ability to pass for real , which machine generation tick both much more often than typical splice methods .
but , maybe miss something in this view . would be curious to hear .