I Cloned My Own Voice to Waste Telemarketer's Time

Follow me for updates!
Twitter: @davepl1968 davepl1968

[JonnyJKF]

"I'm sorry Dave, I'm afraid I can't do that."

[michaelangellotti]

Great ... but I got a lot of stuff to do right now.

[CarterCrews]

the fake “click” from the AI had me absolutely ROLLING on the floor laughing

[tim]

I would love to see an episode of start to finish how you made this model. Excellent work.

[crtune]

It is a convincing "Dave" voice. Good work. Repeats how busy he is a lot.

[theherrmann]

The model overfits at “I’ve got a lot of Stuff to do” 😂

[zyxxy]

Ironically most of the telemarketing calls i receive are (at least initially) AI generated calls. With this idea you would end up with AI to AI calls happening. I wonder if they would be friends.

[rdreher]

I study linguistics and one area of research I've been particularly interested in is prosody, the rhythmic and tonal features of speech. Because I think about that kind of thing, my ears honed in on this particular quirk of the AI voice: although individual sentences had a very natural intonation pattern, when those sentences were strung together the overall prosody felt disjoint. You could feel this long hard stop in between each sentence, which just would not be what normal natural speech sounds like. Even when reading a written script out loud, we anticipate things, our minds saying the next thing before our tongues get to it, and each sentence becomes part of a larger pattern of pace, tone, and intensity, even as they repeat the same internal motifs. It's very much like a piece of music.

[Zaf]

The "I've got a lot of stuff to do" AI. 🤣

And those sounds at the end.

[MikeB_UK]

Sounds just like you. The amount of repeating of "how busy I am right now" might start to give it away. The strange noises at the end really cracked me up.

[garanceadrosehn]

One thing to consider: Some telemarketers will start out saying "This call is recorded for <whatever> purposes". One thing I make sure to do on those is make sure I never given them a straightforward "Yes" or "Good" or "True" reply. I'll reply with something appropriate for the question they're asking, but not give them a "Yes" that they could pull out of the audio recording and match up with a different question. There was one call I had where the guy was *really* trying hard to get me to say "Yes" or "Good".
Another tactic is to say "Just a minute", and to play some random music or TV show.

[kimabg]

I’m somehow left with the impression that you have a lot of stuff to do.

[theantipope]

It definitely sounds legit, Dave. The only thing that would've tipped me off was the repetitive responses.

[benskev]

dude you crushed it as a telemarketer, like wow.

[charlvandermerwe]

I know you are busy and that you have a lot to do, but I think you need to create a tutorial on how to do this. It's priceless!

[xdram]

The ending was perfect 😂 I loved the sounds. Voice was spot on btw great job.

[humphshumphs]

Absolutely bloody brilliant. You really need to video a 'real' call.

[garthhowe]

I think it is even better if the telemarketer thinks they have been "had" by the end of the call. Then they are pissed about it.

[tpaine]

What a great idea! An idea for an episode would be to show how you did this. Keep up the great work! Greetings from a fellow Canuck!

[pyography]

Dave, I'm totally blind, I see literally nothing. I have used hardware and software TTS synthesizers for about 40 years. Let me be among those of us who didn't tell you first, it was very challenging to tell the two voices in that conversation apart.
Will you show us how you cloned your voice and how you connected it to the AI model? SAPI (Speech API), or something else? I'm curious about how much compute power is needed to speak after the hard work of training your voice is done? That would be a very interesting video for me, you explain things in a way I can easily understand.
I always enjoy your videos, thanks for sharing your experiences and ideas!