The Beige of Audio

Okay, so you know the voice. You hear it on a video and within a quarter of a second, you just know.

Oh, the overly enthusiastic, slightly robotic, Here are three things you didn't know about—

Yes! Exactly that one. Why is it always that one? Like, a listener wrote in asking why AI has such a known voice and I know exactly what they mean. It's always this chipper, mid-twenties, usually American...

Right, and completely frictionless. Like they have never had a bad day in their life.

Never skipped a coffee, never got stuck in traffic. It's so weird. Why do tech companies all gravitate to—

Well, I mean, think about what they're trying to achieve. When you're making an AI assistant or a text to speech tool, the brief is basically to sound helpful. You want a voice that is completely non-threatening.

Okay but see, I find it incredibly threatening.

Ha, well, because it hits the uncanny valley, right?

Yeah! It's like a pod person. It sounds like someone who is pretending to be human to trick me into, I don't know, giving them my credit card or joining a cult.

Well they actually hire real voice actors for the base models usually. And the direction they give them is corporate customer service. So it's not even a natural human speaking, it's a human putting on their best, most sanitized phone voice, and then the AI averages it out from there.

Oh, so it's a copy of a copy of a fake voice.

Exactly. It's the audio equivalent of stock photography. But you know what's really wild? The reverse is happening now.

Wait, what do you mean?

Okay, so you know how influencers on social media will do these explainer videos? They're starting to mimic the AI cadence. They do that weird staccato rhythm where every sentence goes up at the end?

Oh my god, yes. Today we are making, an avocado toast! It's like they're pacing it for the algorithm so they don't get swiped away.

Right! So humans made AI sound like a fake corporate human, and now real humans are trying to sound like the AI because the AI retains viewer attention better.

That is— I hate that so much. It's a terrifying feedback loop.

It really is. But coming back to the actual AI generation, part of the reason it sounds so known is just the data. If you feed a machine a thousand hours of professional audiobooks and news broadcasts, it learns a very specific, broadcast-standard pronunciation.

It irons out all the wrinkles. No regional slang, no weird breathing, no vocal fry.

Unless you specifically program the vocal fry in! Which some companies are trying to do now, to make it sound relatable. They literally add little sighs and um's and ah's.

See, that just makes it worse. When the chipper pod-person suddenly goes um, I want to throw my phone across the room. Just be a robot!

Ha, I think we're a few years away from them figuring out actual messiness. Until then, we're stuck with the chipper stock photo voice.

Great. Well, I'm gonna go practice my monotone just to rebel. Catch you guys next time.