OpenAI’s unreleased voice engine is too good:
…uses text input and a single 15-second audio sample to generate natural-sounding speech that closely resembles the original speaker.
I’m generally optimistic about AI’s potential in tools to help humans, but we are clearly going to need more laws around this.
@manton I can imagine this being used in tandem with a text-based audio editor like Descript to clean up speech (and presumably inserting words) and perfect the mimosas, assuming there's an inflection correction tool. Inflection correction tool. Those are three words I thought I'd never type.
Rene van Belzen Having had a person who lives in my hometown, looks like me, speaks similar to me, but has another personality, made different life choice, and who I've never met, I can attest this sounds not only creepy, but downright terrifying. I do not want anyone else to suffer through this Shakespearian confusion of identity by the means of GenAI, epecially if it's with criminal intent.
BTW, what I've learned from my ordeal is that people are general bad at identifying people outside the circle of family and close friends, or using reasoning to discern lookalikes amongst outsiders and strangers. So the AI tool doesn't need to be that good for use in business scams.
Numeric Citizen compare that to Apple's solution where you need to repeat 150 phrases to achieve similar(?) result.
