# Kokoro phoneme ALBERT weights

Source: [hexgrad/Kokoro-82M](https://huggingface.co/hexgrad/Kokoro-82M),
the phoneme ALBERT portion of `kokoro-v1_0.pth`.
The upstream model card explicitly licenses the weights under Apache-2.0;
the license is reproduced in [LICENSE](LICENSE).

This component contains the phoneme ALBERT tensor subset in `model.safetensors`,
with parameter names adapted to PL-Flow. Its SHA256 is recorded in
[bundle.json](../bundle.json).

Kokoro is published by hexgrad. Its model card credits `@rzvzn` with training and
[Li et al. / StyleTTS 2](https://github.com/yl4579/StyleTTS2) with the architecture.
It also attributes training audio to [Koniwa](https://github.com/koniwa/koniwa)
(`tnc`, [CC BY 3.0](https://creativecommons.org/licenses/by/3.0/)) and
[SIWIS](https://datashare.ed.ac.uk/handle/10283/2353)
([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)).

License and attribution reference:
[upstream model card at f3ff3571791e39611d31c381e3a41a3af07b4987](https://huggingface.co/hexgrad/Kokoro-82M/blob/f3ff3571791e39611d31c381e3a41a3af07b4987/README.md).