Post

Copying a voice in less than ten seconds of speech. You're gonna want to develop non-speech habits that are abnormal that models don't understand, as well as challenge questions which seem completely innocuous. I hate this so much.

Tough luck copying my regular speech habits, I flick between accents depending on anything. Switching Australian mid-sentence will surely upset the AI. And I hope it runs on ChatGPT, that's set to go under soon.

Also, it probably copies American rather well, but I doubt it copies other languages or accents anywhere near as well. It's called accidental bias, if I remember correctly.

Bold to assume it would listen to me talking enough to track the full range, especially when I might only go Scottish every few days for a few seconds. And nothing can predict my accent changes. Besides, it could be easily tricked by me only being boisterously American.

look I'm not trying to advertise a product to you here. all I'm saying is that you're asking questions answered within the link in the original post. question: even if it doesn't predict your few seconds of scottish, wouldn't just your common dialect be enough to potentially scam those near you?