If done well, I’ll usually enjoy traditional voice acting over the absence of it, but I’m extremely happy for games with lots of dialogue to only voice the bits where they want to spend the budget because of the heightened emotion of the moment. RPGs where most of the dialogue isn’t voiced is totally good with me when it enables a wider range of writing and stories. Plus the lines that are voiced (especially important: voice the very first line by every character) give you the tone for reading the rest of the stuff. I think that’s where any of the short barks etc stuff can work well - it helps define the characters according to the authorial intent even if the only thing you get is a few “huh”, “ahh”, “hmm” lines that play while characters are talking.
It also heavily depends on genre/structure. I can’t imagine a Life is Strange game without VO but I could totally play such a game with good VO in a language I don’t understand and subtitles. It’s the performance not the actual spoken words that matter for that sort of short drama. I think it’s a minority play style but I know at least some of us enjoy some games with alternative language options (lots of Japanese games with the original Japanese direction, some French games in French, using the voice options in Assassin’s Creed games to get more into the setting etc).
From there (players enjoying VO without comprehending it) it is another step to just building out a language that isn’t actually an established language. I’m interested in where games might go with voice synthesis based not around the incredibly hard task of recreating VO actors giving believable performance you can understand but creating performances that use metadata to evoke emotional notes while the actual language can be gibberish. Characters defined as tones, ranges, and inflections; dialogue written like those classic RPGs you read yourself but with enough metadata to construct the emotional tones intended. One extremely cool thing such a system could do: make the spoken dialogue longer or shorter depending on the player’s desired (reading) speed so they always roughly fit to the pace they wish to read the actual text at (and roughly fit).