We just got our first project that will use an AI to read our voiceover script.
I’ve been running some experiments and the results are interesting.
Despite being far better than early AI voiceovers I’ve heard, it’s still wonky. Emphasis is put in odd places… for instance, if the word “now” is in the sentence, it often assumes an emphasis there for urgency, even if the sentence doesn’t require it.
On the other hand, in the sentence “Every now and then I’m right,” it adds a comma (a slight pause) after “now” and then puts emphasis on “and then I’m right” because it reads “and then” as an indication that something new has happened… As though “I’m wrong first, and then I’m right.” Which interestingly seems to infer that the software is looking ahead at the second half of the sentence, making an assumption about the intent of “and then I’m right,” and backfilling the reading of the first half to fit that intention (adding the invisible comma.)
This client uses a lot of acronyms. So writing H.U.D. gets you three distinct sentences… “The department of H. (pause) U. (pause) D.”
Without the periods, it reads it phonetically as HUD… which works for HUD but not for another org called NCHE, which it reads as “enchee.”
Writing it with spaces but no periods — N C H E — gets closer but the emphasis is odd in the context of the whole sentence… it doesn’t “phrase” the letters together since it doesn’t understand that it’s an acronym.
So… at least in the case of the system this client uses, we can safely say that A.I. is only artificially intelligent, not actually intelligent. 😉
But it’s a good opportunity to play with a tool that I’m sure more and more clients will be leaning on.
To the future!