The God Pill
Well-known member
Experimented with an AI audio program called Eleven Labs to see how easy one could make a deepfake of people's voices I used five minutes of audio clips from a voice actor I worked with on an unfinished motion comic project in 2016. It exceeded my expectations a little you could still tell it was software but the gap between normal text to to speech and real human output with those samples and just three short paragraphs of text for it to to work off of was 80% narrower the only tell was that the tone and pacing was a bit to even for the context so I spent barely a minute tweaking the settings and in the second run it was only half as not natural. With just ten megabytes of audio samples and ten minutes of effort I think you could really fool people with where software is at. We're definitely approaching an era where there will be enormous deception using machine learning.
Last edited: