In late 2013, the Spike Jonze film Her imagined a future where people would form emotional connections with AI voice assistants. Nearly 12 years later, that fictional premise has veered closer to reality with the release of a new conversational voice model from AI startup Sesame that has left many users both fascinated and unnerved.
It feels weird, like maybe over-practiced, but I agree that it sounds human enough to fool me.
It feels reminiscent of the way narrators used to do books on tape. Modern ones are better imho, but all the pausing and intonation definitely seems “professional” more than conversational. Still extremely good.
I listened to an audiobook by Levar Burton a few days ago, and this sounds similar enough to his pattern of speech during the intro that I wouldn’t have known there was anything unusual about the AI voice. If I’d heard it read a book, I would have just assumed that the pauses were a style choice.
Yeah, it sounds like it was recorded in a recording studio, but not like it’s a robot. Very creepy.
Oh! Interesting! So for you this is in the uncanny valley?
I’m on the fence about that.
No, I think it sounds very normal, I just think the idea that I can no longer distinguish between fake and real voices creepy.
It sounds a bit off. But if you’re not looking for it, you won’t find it. That, I believe, is enough to fool most everyone, which is arguably a bad thing.
For reference, this is what Maya reminds me of, Merle Dandridge, VA for Half Life 2’s Alyx Vance.
I’ve skipped to some slower commentary, just so that you can kinda see what a human and AI can sound like with similar pacing while reflecting on a question:
At first, I thought it said it likes peanut butter & people sandwiches, and was like, WTF??? 🫥
Yeah I immediately disregard all AI propaganda that references “Her.” This is for rubes and suckers.
I totally get what you’re saying but did you listen to the recordings? This might be a breakthrough.
But is it that different than the podcasts voices Google already generate with NotebookLM since a while ago?
I used Notebook LM to create a ten minute podcast but it took a lot of repeated attempts with tweaks of the prompt to make sure there was no stupid mispronouncing. Even the final product required editing out two words which were not even human (just glitches).
Not understanding pronunciation of words is to be expected, especially when they are acronyms but sometimes it was really dumb grammar.
Still very impressive but not quite there yet.
Or go try the demo. It IS eerily uncanny. It’s not at the point where it would fool you for long, but it’s close enough to get caught up in it from time to time.
Lmfao, perish the thought
Wow, Jesus. Maya has the same conversational style that Merle Dandridge has in the Half Life 2 commentary tracks (in commentary mode).
That is, even for AI, eerily good. Honestly, I’m not sure if I could always tell them apart if I was fed a bunch of voice clips blind, and asked which ones were human, and which were AI.
