Last updated: October 12, 2026
Summary
OpenVoice PDF processes PDF, text, Markdown and HTML files, OCR results and spoken text on your device. This content is not sent to the developer or a cloud service.
Android: on-device processing
Document rendering and cleanup, text recognition, language identification and speech synthesis run locally on the Android device. Document input, OCR output and narration text are not sent to the developer or to a speech service. The app contains no advertising, user accounts or developer-operated tracking.
Android: ML Kit diagnostics
The Android app uses Google ML Kit for on-device text recognition and language identification. ML Kit SDKs may send encrypted device information, application information, per-installation identifiers, performance metrics and API-utilization metrics to Google for diagnostics, analytics, maintenance, improvement and abuse prevention. Document images, document text and ML Kit results are processed on-device and are not sent to Google by these APIs.
Optional model download
When you choose to install a voice pack and accept the applicable model terms, the app downloads model files over HTTPS. Supertonic files are downloaded from an official GitHub release. The optional Android Qwen voice model is downloaded from Hugging Face and its download infrastructure. Your documents, OCR results and narration text are not transmitted during these downloads. Download providers receive ordinary connection information, including your public IP address. After installation, speech synthesis runs on your device; selecting Qwen does not send text to a cloud speech service.
Android: local storage and permissions
Selected settings and downloaded voices are stored in the app's private storage. You can remove them through Android's app settings or by uninstalling the app. Internet access is used for the optional model download and ML Kit diagnostics described above. Notification, foreground-service and wake-lock permissions enable user-initiated background reading and lock-screen playback controls.
iPhone and iPad
The iOS version uses Apple Vision for on-device OCR and Supertonic 3 through ONNX Runtime for local speech synthesis. The Android ML Kit diagnostics and Android permissions above do not apply to iOS. The iOS app has no advertising, accounts, developer-operated analytics or tracking. It does not upload documents, OCR results or narration text to the developer or a speech service. On iOS, preferences, imported documents, downloaded voices and the reading position are stored locally. The share extension transfers documents to the main app through an on-device App Group container. System backups and file-provider behavior depend on your iOS settings. Delete App removes local app data; original documents and existing backups may remain separately. Background audio is used only for user-initiated reading. The app does not request microphone, camera, location or advertising-identifier access.
Speech model and licenses
Supertonic 3 is used under the OpenRAIL-M license. Before downloading it, users must accept its use restrictions. The app contains a Legal, privacy & licenses screen with model and runtime notices. Supertonic inference uses the MIT-licensed ONNX Runtime. The optional Android Qwen model uses a local native runtime; its model and runtime license notices are included in the app. The app does not include Piper or eSpeak.
Support and contact
For support or privacy questions, contact nikki@gmx.ch or +41 76 358 33 33. Please include your device model, system version and a description of the issue. Do not send private documents unless you explicitly wish to share them for troubleshooting.
Getting started on iPhone and iPad
Use Open document to import PDF, TXT, Markdown or HTML, or use More > Paste text. Choose Listen, accept the voice-model license and install the voice pack. After this initial download, document processing, OCR and speech work on your device. To begin at a chosen word, select text and choose Read from here. Reading settings include language, voice, speed, dialogue mode and text following.