Show HN: An improved kokoro distillation in 9MB
Category: ai-ml
Tags: text-to-speech, model-distillation, kokoro, audio, machine-learning
Score: 7.0/10 (Innovation: 7, Technical: 8, Documentation: 4, Utility: 6)
This project distills the Kokoro text-to-speech model into a 9MB model with ~25x real-time factor, improving upon a previous paper by making the training recipe replicable and adding tweaks for better quality. It includes a browser demo, pip-installable package, and weights, targeting multiple languages starting with Italian.
Target audience: machine learning engineers, speech synthesis researchers, developers integrating TTS
Repository: https://huggingface.co/spaces/MithrilMan/teacher-in-the-loop
View on Hacker News