Terms of Use

Olzaras: AI Voice PDF Reader · v1.2 · 2026-09-14

By installing or using Olzaras: AI Voice PDF Reader (“the App”) you agree to these Terms. If you do not agree, do not use the App.

1. Use restrictions (enforceable provision, BigScience OpenRAIL-M §5)

The neural voices in the App are powered by the Supertonic 3 model, licensed under the BigScience OpenRAIL-M license. Its Use Restrictions (Attachment A of that license, shown in full on the Licenses screen in the app) are incorporated into these Terms as a binding, enforceable provision. You agree not to use the App or any audio generated with it, among the other restrictions of Attachment A:

  • in any way that violates applicable law;
  • to exploit or harm minors;
  • to generate or disseminate verifiably false information intended to harm;
  • to defame, disparage or harass others;
  • to impersonate any person without consent (e.g. deepfakes);
  • to present machine-generated audio as human without disclosure.

Every exported audio file is tagged as AI-generated (machine-readable ID3 metadata) and additionally carries an inaudible AudioSeal watermark; if watermarking fails on a device (for example under memory pressure), the export still carries the AI-generated tag. You must not remove or circumvent these disclosures. These restrictions apply to all subsequent users of audio you share, and you must give notice of them to such users (OpenRAIL-M §4(a), §5).

2. Third-party notices (Apache-2.0, MIT, ICU)

The App bundles third-party components. Their attribution notices are listed in the NOTICE file on the Licenses screen, and the full Apache License 2.0 text is included. ICU4J is used under the Unicode License v3; its copyright and permission notice is reproduced in the NOTICE file. The following components are licensed under the MIT License: ONNX Runtime (© Microsoft Corporation); AudioSeal, code and weights (© Meta Platforms, Inc.); Supertonic 3 inference code (© Supertone Inc.); voiceorbs “Nebula Orb”, the speech-reactive sphere ported to Android GLES (© 2026 Alexis Munoz); webgl-noise Simplex noise, GLSL (© 2011 Ashima Arts). The MIT permission notice is reproduced on the Licenses screen; the software is provided “as is”, without warranty of any kind.

3. LAME MP3 encoder: prominent notice (LGPL-2.0 §6)

This App uses the LAME MP3 encoding library (version 3.100), dynamically linked as a separate library (libmp3lame.so). LAME and its use in this App are covered by the GNU Library General Public License, version 2 (LGPL-2.0), or at your option any later version; a full copy of version 2 is included on the Licenses screen. The complete corresponding source of LAME 3.100 is available at https://lame.sourceforge.io and is also pinned (with SHA-256) in the NOTICE file; on written request to the publisher a copy of the exact source archive will be provided. You may modify and reverse engineer the LAME library portion for your own use as permitted by the LGPL.

4. Model weights (OpenRAIL-M)

The Supertonic 3 model weights are used under the BigScience OpenRAIL-M license (full text on the Licenses screen). The Licensor claims no rights in the audio output you generate; you are accountable for the output and its uses (OpenRAIL-M §6), subject to the Use Restrictions above.

5. Reading aids

The optional Smart narrator may leave out non-narrative elements when reading aloud (page numbers, running headers, Project Gutenberg boilerplate, footnote marks, scene-break lines, hyphenation, duplicated lines, tables of contents and speaker labels). Your text is never altered: skipped parts stay visible on screen, each category can be switched off, and the printed text remains the authoritative version.

6. Publisher

DH Studio Technologies, contact: support@dhstudio.dev. The App has no network access and collects no data; all text and audio are processed on your device.