It would be hilarious that if they get around to talking about my paper, they would be like "Wow! That seems like such a misguided idea! Yeah Joe Prime, I wonder if this author just needed to push out a half-baked paper because he's worried that he's about to be fired. I don't think so Ferris Prime. It appears that this guy spent two years working on this crap, based on his recent publications. Wow, that's sad. Interlude baby.
This is great! I'd love to use this method to generate something similar. I have a couple of questions on the process:
- How do you split the conversation in two for each of the speakers? Is that in the ChatGPT prompt or do you filter out each speaker question/answer before putting it through tts?
- Where did you get the voices you use from? Did you train them yourself? I don't see these in the list of available out-of-the-box voices in Tortoise;
I might be more interested in hearing straight reading of the abstract + gpt summary of each paper (could go section by section) rather than the conversation format
What strikes me right away is that in any other podcast listening experience I automatically place the context of the speaker around the content, i.e. I give Joe Rogan's opinions a certain weight due to what I know about him (or think I know). These voices seem to have opinions ("wow!" has a positive connotation when speaking about these papers) and I realize I have no idea what the characters base their opinions of the papers on. This is super important work though - it's almost inevitable that we'll have AI generated content delivery - and what humans look for in content delivery is a perspective on that content, an opinion, where AI by default will only have the opinion it is told to have by a human or script. I don't know how things will go, but this is an early step into what will be the defining of relationships between AI and humans, as well as between different AIs. Awesome.
It would be hilarious that if they get around to talking about my paper, they would be like "Wow! That seems like such a misguided idea! Yeah Joe Prime, I wonder if this author just needed to push out a half-baked paper because he's worried that he's about to be fired. I don't think so Ferris Prime. It appears that this guy spent two years working on this crap, based on his recent publications. Wow, that's sad. Interlude baby.
This is great! I'd love to use this method to generate something similar. I have a couple of questions on the process:
- How do you split the conversation in two for each of the speakers? Is that in the ChatGPT prompt or do you filter out each speaker question/answer before putting it through tts?
- Where did you get the voices you use from? Did you train them yourself? I don't see these in the list of available out-of-the-box voices in Tortoise;
Thanks!!
The third voice is hilarious to me
I might be more interested in hearing straight reading of the abstract + gpt summary of each paper (could go section by section) rather than the conversation format
What strikes me right away is that in any other podcast listening experience I automatically place the context of the speaker around the content, i.e. I give Joe Rogan's opinions a certain weight due to what I know about him (or think I know). These voices seem to have opinions ("wow!" has a positive connotation when speaking about these papers) and I realize I have no idea what the characters base their opinions of the papers on. This is super important work though - it's almost inevitable that we'll have AI generated content delivery - and what humans look for in content delivery is a perspective on that content, an opinion, where AI by default will only have the opinion it is told to have by a human or script. I don't know how things will go, but this is an early step into what will be the defining of relationships between AI and humans, as well as between different AIs. Awesome.
i've love to learn how to do this. so powerful.