y20k/voxscribe
1
0
Fork
You've already forked voxscribe
0
Voxscribe - Offline Voice-Input Keyboard for Android
  • Kotlin 90.5%
  • C++ 8.2%
  • CMake 1.3%
2026年07月12日 23:55:24 +02:00
.ai Finish styling, nav-bar tint, and V1 close-out (P7.4) 2026年07月12日 18:05:55 +02:00
app Give the input panel its own surface-container background 2026年07月12日 23:55:24 +02:00
assets Add app launcher icon and F-Droid metadata icon 2026年07月10日 23:14:37 +02:00
external Vendor whisper.cpp and wire native build (P6.1) 2026年07月10日 12:04:53 +02:00
gradle Add JNI bridge: model load from SAF fd and transcribe entry point (P6.2) 2026年07月10日 15:45:24 +02:00
metadata Add F-Droid store metadata and screenshots 2026年07月12日 18:06:03 +02:00
.gitignore Ignore app/release build output 2026年07月10日 13:12:50 +02:00
.gitmodules Vendor whisper.cpp and wire native build (P6.1) 2026年07月10日 12:04:53 +02:00
build.gradle.kts Initial commit 2026年07月09日 10:14:10 +02:00
gradle.properties Initial commit 2026年07月09日 10:14:10 +02:00
gradlew Upgrade Gradle wrapper to 9.6.1; bump compile/target SDK to 37 2026年07月09日 11:55:49 +02:00
gradlew.bat Upgrade Gradle wrapper to 9.6.1; bump compile/target SDK to 37 2026年07月09日 11:55:49 +02:00
LICENSE.md Initial commit 2026年07月09日 10:14:10 +02:00
README.md Finish styling, nav-bar tint, and V1 close-out (P7.4) 2026年07月12日 18:05:55 +02:00
settings.gradle.kts Initial commit 2026年07月09日 10:14:10 +02:00

README

Voxscribe - Offline Voice Input Keyboard

Version 1.0 ("Boing Boom Tschak")

Voxscribe is an offline voice-input keyboard (input method) for Android. Hold the mic button to record; your speech is transcribed entirely on-device with whisper.cpp using a Whisper model file you supply yourself, and the result is written into the focused text field.

Voxscribe is free software. It is published under the MIT open source license.

Setup

Voxscribe does not ship a speech model. You download one yourself and point the app at it.

  1. Download a Whisper model file in GGML format from the Hugging Face repository ggerganov/whisper.cpp:

    • ggml-base-q5_1.bin (~60 MB), the recommended default
    • ggml-small-q5_1.bin (~190 MB), better quality but slower
    • ggml-tiny-q5_1.bin (~32 MB), for older or weak devices

    Use a multilingual model only. The .en (English-only) variants break both language modes.

  2. Open Voxscribe, tap Model file, and pick the file you downloaded.

  3. Under Setup, tap Enable Voxscribe keyboard and turn Voxscribe on in the system keyboard settings.

  4. Under Setup, tap Microphone permission and allow recording. You can also grant it the first time you hold the mic button.

  5. In any text field, switch to Voxscribe with the keyboard's mic button or the system input-method switcher, then hold the mic button to record and release to insert the transcript.

Features

  • Offline, on-device transcription, so nothing leaves your phone
  • Push-to-talk: hold to record, release to transcribe
  • Two favourite languages to switch between on the panel, or automatic language detection
  • Light and dark themes, plus Material You dynamic colors

A word on privacy

Voxscribe works entirely on your device. Your speech is recorded while you hold the mic button, transcribed locally by whisper.cpp. There is no web service behind Voxscribe. The Whisper model stays wherever you saved it. The only permission Voxscribe requests is microphone access, needed to record your voice. Voxscribe does not hold the INTERNET permission. It cannot transmit data.

Screenshots