🎙️ KawiriTTS Inference Playground

ℹ️ About Me & The Model

Hi there! I am the developer behind KawiriTTS. This model is using a 44.1 kHz VAE vocoder combined with a Flow-Matching architecture. This 2-stage architecture is a stunning 100M parameters. I trained the entire model for $50 on 1x RTX 5090 using 880 hours of English speech.

I don't have a job right now, so if you want to hire me, feel free to DM!

Note: It currently has some Word Error Rate (WER) issues, but these will be improved in future iterations.

📖 Try a Random Story