Offline Translator

Offline Translator by David Ventura uses downloadable open models instead of forwarding each sentence to Google, DeepL or another translation API. Its current language list covers more than 60 languages, including English, Portuguese, Spanish, Chinese, Japanese, Korean, Arabic, Hindi and many European and South Asian languages. You install only the packs you need; their storage use is shown in the language-pack screen.

What runs on the phone

• Text translation uses models derived from Mozilla’s Firefox Translations work. Automatic language detection is provided by CLD2, and transliteration is available for non-Latin scripts.
• Images and the live camera use PaddleOCR models. You can select recognised words for a dictionary lookup or edit recognised text before translating it.
• Documents include PDF, ODT, EPUB and TXT. Version 0.8.7 fixes some PDFs that failed to open and reduces memory use with long text and documents.
• Dictionary data comes from Wiktionary exports by Kaikki; Japanese also has a word dictionary for Kanji transliteration.
• Speech is optional and model-based. Version 0.8.7 enables higher-quality Kokoro voices for English, Spanish, French, Hindi, Italian and Portuguese.

Text can be typed in the app, shared from another app or sent through Android’s selected-text Translate action. The app also registers for ordinary HTTP and HTTPS links under “Translate link” and opens the chosen page in its own WebView; the website therefore receives a normal network request even though selected text is translated locally. A mini overlay and screen translator are available for material that is difficult to copy. There is also an AIDL interface through which another Android app can request a local translation.

Downloads and sensitive access

The APK includes a model catalogue, and every model-file URL in it points to the project’s offline-translator.davidv.dev host. Language management offers a manual catalogue refresh from that host; the catalogue address can be changed in Settings. Mozilla, PaddleOCR, Kaikki and the other named projects describe the data’s provenance, not extra download hosts contacted by the app. Once the required data is present, translation, OCR and dictionary lookup stay on the device.

An optional LibreTranslate-compatible HTTP interface is off by default and normally listens on port 5000. It can be changed from localhost-only to all interfaces; doing that exposes translation requests to devices able to reach the phone, so use it only on a network you trust. Our APK scan found no tracker SDKs, but screen capture, camera, Accessibility-assisted translation and display-over-other-apps can still see sensitive content while you actively use them.

What changed in 0.8.7

Besides fixes for some PDFs and memory use, the input transliteration can now be selected and copied; Android 8 image translation no longer crashes; selected image words gain a Dictionary action; and recognised image text can be reopened as editable text. These are the developer’s release notes, not claims from device testing here.

About this file

• This is the developer’s arm64 APK from the v0.8.7 GitHub release. It matches the SHA-256 digest GitHub publishes for the asset, and its signing certificate matches the fingerprint published in the project README. F-Droid’s 0.8.6 package uses the same developer key, so Android can update between those channels when version codes allow. The signer is pinned locally so a future file with another key will be rejected.
• The 44.9 MB file runs on Android 6.0 or later and contains only 64-bit ARM native libraries. The developer also publishes a separate 32-bit ARM asset; it is not this download.
• The manifest declares 14 permissions, including internet, camera, overlay, media projection, notifications, boot and storage-related permissions. Our scan found no tracker SDKs.
• Offline Translator is GPL-3.0-or-later. Its four retained project screenshots were not all made for 0.8.7: their last changes entered the 0.7.1, 0.8.2, 0.3.2 and 0.1.3 release lines respectively. Their captions identify the old interface and what each image actually shows.

CPU types in this build: arm64-v8a (64-bit ARM, almost every phone since 2017).

Recent releases on the project’s GitHub. We host and check only the newest one. Full changelog.