New App: TranSlander - Offline voice typing

  • The app complies with the inclusion criteria.
  • The app is not already listed in the repo or issue tracker.
  • The app has not already been requested
  • The upstream app source code repo contains the app metadata (summary/description/images/changelog/etc) in a Fastlane or Triple-T folder structure
  • The original app author has been notified, and does not oppose the inclusion.
  • Optionally donated to support the maintenance of this app in F-Droid.

APPLICATION ID: at.webformat.translander 

Categories:
 - Writing
 - Keyboard & IME

License: Apache-2.0

AuthorName: hatsch
AuthorEmail: hatsch@servus.at
AuthorWebSite: https://github.com/hatsch

WebSite: https://github.com/hatsch/TranSlander
SourceCode: https://github.com/hatsch/TranSlander

IssueTracker: https://github.com/hatsch/TranSlander/issues

RepoType: git

Repo: https://github.com/hatsch/TranSlander.git

Why do you want this app added to F-Droid:

Privacy-focused voice typing should be available to everyone. This app processes all speech locally using open-source neural models - no cloud, no tracking, no data leaving the device. F-Droid is the ideal home for privacy-respecting software.

Summary:

Offline voice typing using local neural speech recognition - no internet needed

Description:

TranSlander converts speech to text entirely on your device using the Parakeet neural model via sherpa-onnx. No internet connection is required after the initial model download.

Features:

  • Works in any app via Accessibility API text injection
  • Multiple input methods: floating mic button, accessibility button, keyboard integration
  • Voice message transcription from audio files (OPUS, AAC, M4A, MP3, etc.)
  • Folder monitoring to auto-transcribe new voice messages
  • Word corrections dictionary for fixing recognition errors
  • 25 languages with auto-detection
  • Material 3 UI with dark/light theme

Privacy: All speech processing happens locally. After downloading the ~600MB speech model (user-initiated, opt-in), the app works completely offline. No telemetry, no analytics, no cloud services.

Note: The app downloads the Parakeet speech model (~600MB) from HuggingFace on first use. This download requires explicit user action and is clearly communicated. The sherpa-onnx library (Apache 2.0) is used for on-device inference.

Edited by hatsch