H
Huggingface
Kokoro TTS distillation in ~9MB, explored as an experiment for Italian voices
Visit Huggingface ↗
link: rel="ugc noopener"
This Hugging Face Space by MithrilMan presents an improved Kokoro distillation in about 9MB. It builds on a paper describing a ~9MB distilled Kokoro model with a reported real-time factor (RTF) of around 25x, which was tested on a single English voice.
The author, an Italian developer, set out to adapt the approach to Italian after finding the original training recipe was not reproducible. The page's description is cut off, so this listing does not state specific results.
Discussion