AI narration
AI narration has started to really really bother me.
So many AI-narrated videos feature fake British accents – why?
The tone is posh and often supercilious – why?
There’s monotony.
Many mispronunciations, like “desert” for “dessert” or vice versa. Or like “read” present tense for “read” past tense.
There’s the pause at the end of a line of text even if it’s in mid-sentence.
Some feature a little exhaled sigh – “ah” – in an effort to make the voice seem human. It’s an effort that fails utterly.
But most annoying of all is when AI narration uses the first person. Who is this “I” of whom you speak, kimosabe?
And one day they’ll get it right, which in this case means wrong. In a generation the sky will neither be blue or even up, but whatever the fake Richard Feynman says it is. That’s gonna leave a mark.
(inhale) I find your lack of faith disturbing. (exhale)
Annoying, occasionally most certainly. Betimes however hilarious, as when for instance the greek “Psi” is pronounced “piss-eye”.
Shibboleths of the world unite! You have nothing to lose but your covers!
When I called my pharmacy for a drug refill, I got Amy the AI Assistant. I said, Let me talk to a pharmacist, to whom I had to repeat my name and DOB.
And El-Sayed the Arab physician has closely won the Dem primary in MI for a seat in the Senate. This is how the National Socialists aka Nazis won in the 1930s, by stupid people casting protest votes.
Not a good morning.
Here,here!
Maybe my biggest beef (of many) with AI is how it has a way of getting stuff just plain &^%$#ing WRONG.
When there’s a topic with which I’m very familiar, and AI states something that I *know* (from long familiarity) is just plain &^%$#ing factually INcorrect, I have a small, silent fit. And it happens a lot.
[ M J R exhales ]
(Nothing useful to add, just venting. Carry on.)
“Here, here”…
Sharon, I hope that was a tongue-in-cheek homophonic quip. If not, be aware that the correct version is “Hear, hear.”, as if to command a crowd to “Hear, hear, these words of wisdom”!
I frequently enjoy viewing nautical history on YouTube, except for the all-too-frequent AI-generated homograph “bow”, where the pointy end of the boat is pronounced like it was one of those things that shoots arrows!
Some I find unbearable. Others I enjoy. I’m not sure what the difference is, but I think it has something to do with if the input sounds believable.
AI audio is terrible. Yet AI-generated video is what gets my goat. Flat-out wrong, wrong, wrong! Engine cowling cover mounted to the top of the canopy? Really? Browning machine guns mounted to the top of the wing of a P-47? Please just put AI out of its misery…
I guess I don’t view much on Youtube that uses AI narration. I do watch documentaries with something of a production company behind it, as well as TV news of various sorts, which are all humans.
With latter, I’ve noticed some radically different speaking styles, some of which are annoying. Some pundits use a slower style with tons of dramatic emphasis. A trailing syllable that is drawn out, followed by a pause. A tiny bit is OK, but it’s annoying when excessive.
But the one that really freaks me out, & maybe it is just me, are the rapid speakers who have no spaces or pauses what-so-ever. It’s rare to hear this, but it’s so strange. Where are the word breaks?? I find myself struggling to parse the words.
It reminds of traveling abroad and sitting next to a few Spaniards conversing. Of course, I don’t understand it, especially at that speed, but it sounds like a verbal machine gun at 1,000 rounds a minute.
Thanks so much, Ray Van Dune. I confess ignorance in this case and appreciate the correction!
And I thought Sharon W was pointing, “Here, here,” not “There, there” (it is). As opposed to “Here, hear.” Come “Here” and listen “hear.” And of course “They’re there.” Then again, “there, there” can be a phrase of consolation not a direction.
Am I clear? 🙂
Indirectly related, but this opinion piece from NYT makes a cautionary point about A.I. “The problem with writing with A.I. is that it’s mentally enfeebling — an escalator toward a result when you really need to make a daily habit of taking the stairs.”
Don’t like it, don’t trust it. Relating to pharmacies; my pharmacy had a fine phone system. I knew when to press 1 or 2 without listening to the whole spiel. Worked great. Too retro, I suppose. Now, not so great.
I usually laugh when the ubiquitous AI assistant informs me that ‘it’ is thinking. Thinking is certainly taking on a new meaning; and it remains to be seen how this electronic substitute for human thought will effect society as it becomes endemic.
British accents. After the world cup, which broadcast 12 hours or so a day in the house, I am so sick of British accents that I could be sick. Of course the broadcasts of women’s pro soccer have the obligatory accent as well. It may be obvious by now that there is a soccer fan in the house; but it isn’t me.
Strangely, when I worked in England I don’t recall having too much trouble understanding spoken English. I begin to suspect that on-air personalities have been schooled to speak in a way that they hope Americans will hear as sophisticated; and I hear as virtually unintelligible.
The recent video on Roman aqueducts is a case in point. Despite the clear meaning of the text about the metal pipes, the narrator continually pronounced “lead” as “leed”.
What burns my cookies about AI videos is the terminal vagueness and repetitiveness. I must admit I have problems listening to circumlocutory humans as well. I enjoy Pres. Trump’s rallies, but sometimes the weave is just too much. May the Lord forgive me for my enjoyment of its effects on the talking heads.
I am of the belief that we need a regulation requiring all ‘chatbots’ and other software never to use a personal pronoun in reference to the software. Instead of ‘I’ they should be required to say/print “this algorithm” or something along those lines.
I suspect it would at least help ameliorate the tendency for people who are already edge cases to mistake it for a person.
A lot of AI adds on TV. Often the background is out of focus. And the women have eyes that are unrealistically large.
“El-Sayed the Arab physician has closely won the Dem primary in MI for a seat in the Senate. This is how the National Socialists aka Nazis won in the 1930s, by stupid people casting protest votes.” Cicero
Indeed, though otherwise they are not analogous. Not that a civil war is desirable but the DSA is not going to be allowed to “fundamentally transform” America. If politics fails, “other means’ will be applied. Our unalienable rights are not subject to ‘revision’ by earthly authorities. .
I find AI narration so predictable it’s annoying. However, that’s AI’s vanilla voice. I asked ChatGPT and Chat agreed:
_____________________________
Yes. The “vanilla AI” sound is usually not a single tell but a bundle: balanced clauses, orderly transitions, mild uplift, repeated triads, abstract nouns, and an eagerness to summarize the significance of everything. In YouTube scripts it often sounds like:
“But here’s where things get interesting. This isn’t just about X—it’s about Y. And that raises an important question…”
Perfectly serviceable, but after a while you can hear the machinery.
As an improvement Chat recommends:
_____________________________
The strongest method is to supply samples of the desired voice and ask the model to infer the operating principles—not merely imitate vocabulary. Something like:
Study these three passages for rhythm, stance, level of compression, humor, and use of detail. Write new material with those same underlying tendencies, but do not reuse distinctive phrases.
And then revise in passes. “Make it 20% drier.” “Remove every sentence that explains the previous sentence.” “Replace generalizations with observed particulars.” “Let the ending stop rather than conclude.”
That last stage matters because vanilla AI style partly comes from its instinct to be maximally legible and complete. Distinctive human prose often includes preferences, omissions, asymmetries, hobbyhorses, abrupt turns, and things left hanging. Voice is not merely what a writer does. It is also what the writer repeatedly refuses to do.
_____________________________
AIs are spooky good at mimicry — if you give them something to imitate.
AI reminds me of those dolls where you pull a ring on the back of their necks and they “speak.”
AI reminds me of those dolls where you pull a ring on the back of their necks and they “speak.”