# Media rights and privacy review
Original scripted lessons, code-drawn illustrations and examples; no client records, stock photos, music or personal voice cloning.
Kokoro-82M stock af_heart voice, model publisher https://huggingface.co/hexgrad/Kokoro-82M (declared Apache-2.0); kokoro-onnx runtime declared MIT. Notices included. Build records downloaded model/voice hashes in voice-provenance.json; this is not an upstream training-data rights audit or a warranty of exclusive output ownership.
DejaVu font notice: /assets/academy/dejavu-font-notice.txt.

Media uses same-origin delivery after Play; page/images/captions may request earlier. The host receives ordinary request metadata. Hosting retention and account-level logging remain operator-review limitations. Module 4 adds no analytics, third-party embeds or learner uploads. Optional localStorage valistera_module4_v1 stores studied lesson IDs, passed lesson checks and final practice score; defaults off. Opt-out and Clear remove only this module key. Workbook responses are not persisted. No learner data is a controlled GMP training record.

Human audio, physical Android/iOS, screen-reader and rights/privacy acceptance remain separate review gates. No paid generation API or personal voice upload is used.