MUCKYWORX // PROJECT
MuckyTTS
PROJECT

MuckyTTS

I'm training Qwen3-TTS on my own voice, and maybe a Piper-TTS voice will even come out of it.

So, for those who didn't know yet: I've written my own novel and I want to turn it into an audiobook sometime soon. And since I can't be bothered to read my own book out loud again, I'll just do it using my own voice via generation with Qwen3-TTS. To do that, I first have to train my voice. Over the course of several months, I used my Voice Chaser pipeline to not only transcribe my voice but also record it and simultaneously create a dataset. The original dataset comprised 67 hours of pure speaking time, and after distillation – removing noise and so on – at least 43 hours remained. The whole thing is still in development and ongoing, but I'll show you the fun part here, where you can download the dataset from Hugging Face. It's only 28 GB...