5.3MB Japanese voice model — license, credits and changes

Covered file: student_polish.convpost_r58.onnx (5,314,553 bytes)
Adaptation and student model: yasmorita, 2026

This published student model is shared under Creative Commons
Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).
The license applies to yasmorita's copyright and similar rights, to the
extent they exist, in the contributions to this model. Existing upstream
rights and notices are retained.
License text: LICENSE.txt
https://creativecommons.org/licenses/by-nc-sa/4.0/

When sharing this model or an adaptation, retain the supplied credits,
license and disclaimer notices, link to the source, and identify changes.
Use the licensed material for noncommercial purposes. Share adaptations
under CC BY-NC-SA 4.0, without additional restrictions on licensed rights.
The license text, rather than this summary, governs.

Credits and source
------------------
Base voice: ja_JP-hi_fi_captain-medium, published by rhasspy/piper-voices.
https://huggingface.co/rhasspy/piper-voices/tree/main/ja/ja_JP/hi_fi_captain/medium
Upstream model card (also retained as UPSTREAM_MODEL_CARD.txt):
https://huggingface.co/rhasspy/piper-voices/blob/main/ja/ja_JP/hi_fi_captain/medium/MODEL_CARD

The upstream card identifies the Hi-Fi-CAPTAIN dataset and its license:
https://ast-astrec.nict.go.jp/en/release/hi-fi-captain/
https://creativecommons.org/licenses/by-nc-sa/4.0/deed.en
It also states that the base voice was fine-tuned from a LibriTTS-R base model.
These upstream credits do not imply endorsement of this adaptation.

Changes by yasmorita
-------------------
The teacher was fine-tuned on 122 recordings of yasmorita's voice (393 seconds).
A smaller VITS student was trained on 3,023 teacher-generated utterances
(149.7 minutes), then exported to ONNX with selective int8 quantization.
Some decoder layers remain in floating point.

Publication scope
-----------------
Only the named 5.3MB student model is distributed here. The 22.8MB teacher
model, training checkpoints and original voice recordings are not included.
This notice does not apply a license to the website, article text or runtime
software. The runtime's license and third-party notices are supplied separately
under /vendor/onnxruntime-web/.

No additional permission regarding personality, publicity, privacy or trademark
rights is granted by this notice. The model is provided as-is, without
warranties, as described in the license. This license notice does not resolve
all questions about upstream rights or establish unrestricted reuse of
upstream materials or generated audio.