ElevenLabs' new V3 model can generate full conversations with emotions like laughter and shouting. Matt Wolfe trained it on his own voice using a 30-second sample and got back a dialogue that sounded plausible but not quite like how he hears himself.
June 17, 2025
A quick option requiring ~30 seconds of audio and a higher-quality option using ~30 minutes of audio.
11 labs also rolled out a new voice feature this week. It can even do, like, conversations back and forth with people, and you can add in things like chuckles and laughing and excited and shouting. Here's one of their dialogue examples. Hey, Chris. Knock, knock. I'm not doing this again. Come on, man. Please. I promise you'll love this one. I mean, it sounds really, really good. So I went ahead, jumped over to 11 labs and trained my own voice into it. When you train your voice, they really give you two options. They have the quick option where they just need, like, 30 seconds of audio, and then they have the better quality option where you train on, like, 30 minutes of audio. I just did the 30 seconds of audio, so you probably can make this sound a lot better. But here's what it made from my voice, and you can see that it put enthusiastically here and then put laughter at the end. Everyone should subscribe to Matt Wolf and check out future tools. In my own head, that doesn't really sound like me, but I'm sure people in the comments are gonna say, that sounds just like you. I have this, like, weird cognitive dissonance when it comes to hearing my own voice, where I'm like, is that what I sound like? But then other people are like, that is so. I don't know. Did that sound like me?
AI News: Dots, GPT-6.1 Sol, Sonnet 5.5, Gemini 4, and everything you need to know
