HEBREW "NATURAL VOICE" MODEL - README AND LICENSE NOTICE

Files in this folder:
  he.onnx   the trained text-to-speech model (ONNX format; runs with ONNX Runtime and Piper-style input)
  he.json   its settings (phoneme/character id map, sample rate, inference defaults)
  README.txt  this notice

LICENSE
This model is licensed under Creative Commons Attribution-ShareAlike 3.0 (CC BY-SA 3.0)
https://creativecommons.org/licenses/by-sa/3.0/

ATTRIBUTION
Training audio: Tanakh narration read by Dan Be'eri, Internet Archive collection "Tanakh Narration"
https://archive.org/details/Tanakh_Narration_01Genesis (CC BY-SA 3.0).
Starting checkpoint: LJ Speech English "lj-med" model (public domain data) from the Piper project
(rhasspy/piper-checkpoints, Open Home Foundation).
Training tool: Piper (OHF-Voice/piper1-gpl, GPL) was used only as software to train the model.
Training text/pronunciation: written for the Binyanim Lab website from Hebrew text of the MACULA Hebrew
project (Clear Bible, Westminster Leningrad Codex; Public Domain / CC BY 4.0).
Credit for this model: "Binyanim Lab Hebrew voice, trained on recordings by Dan Be'eri (CC BY-SA 3.0)".

CHANGES
The model was trained (fine-tuned) from the starting checkpoint on short clips of the recordings above,
which were aligned to a phonetic spelling of modern Israeli Hebrew pronunciation. Input is that
spelling (one character per sound), not ordinary Hebrew letters.

YOU MAY copy, share and adapt this model, including commercially, if you give the credit above, indicate
changes, and release your adapted version under CC BY-SA 3.0 (or a compatible license).
Provided as is, without warranty.