Offline Voice Input (Android)

An offline, privacy-focused voice input keyboard and live subtitle tool for Android, built with Rust.

Features

• Offline Transcription: Uses deep learning models (Parakeet TDT) to transcribe speech entirely on-device.
• Supported Languages: Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Slovak, Slovenian, Spanish, Swedish, Russian, Ukrainian
• Voice Input Keyboard Use your voice as a text field input method.
• Live Subtitles: Get real-time captions for any audio or video playing on your device (e.g., YouTube, Podcasts, Zoom).
• Privacy-First: No audio data leaves your device.
• Rust Backend: Efficient and safe native code using transcribe-rs (included).
• Native Android UI: Clean Material Design interface.

Screenshots

Building Prerequisites (Linux)

To build this app, you need the following system packages:

sudo pacman -Syu jdk-openjdk rustup unzip zip base-devel cmake

(Adjust for your distribution: e.g., apt install openjdk-17-jdk build-essential cmake unzip on Ubuntu)

Ensure you have the aarch64-linux-android target for Rust:

rustup target add aarch64-linux-android

Download and extract the parakeet model

curl -L https://blob.handy.computer/parakeet-v3-int8.tar.gz | tar -xz -C assets

Android SDK Setup (Manual)

Follow these steps to set up the SDK and NDK:

1. Setup Directory:

   mkdir -p android-sdk/cmdline-tools
   cd android-sdk

2. Download Tools: Download the command-line tools from Android Developers or use wget:

   wget https://dl.google.com/android/repository/commandlinetools-linux-11076708_latest.zip -O cmdline-tools.zip
   unzip -q cmdline-tools.zip
   mkdir -p cmdline-tools/latest
   mv cmdline-tools/bin cmdline-tools/lib cmdline-tools/NOTICE.txt cmdline-tools/source.properties cmdline-tools/latest/
   rm cmdline-tools.zip

3. Install SDK Components:

   export ANDROID_HOME=$(pwd)
   export PATH=$PATH:$ANDROID_HOME/cmdline-tools/latest/bin
   
   yes | sdkmanager --licenses
   # Install Platform, Build Tools, and Platform Tools
   sdkmanager "platforms;android-35" "build-tools;35.0.0" "platform-tools"

4. Install NDK r28:

   # Inside android-sdk/ndk/
   mkdir -p ndk
   cd ndk
   wget https://dl.google.com/android/repository/android-ndk-r28-linux.zip
   unzip -q android-ndk-r28-linux.zip
   rm android-ndk-r28-linux.zip

   Note: Ensure your ANDROID_NDK_HOME environment variable points to this directory (e.g., .../android-sdk/ndk/android-ndk-r28).

Building

APK (Debug/Release)

You can build a standalone APK for testing:

# Ensure ANDROID_HOME and ANDROID_NDK_HOME are set
export ANDROID_HOME=$(pwd)/android-sdk
export ANDROID_NDK_HOME=$ANDROID_HOME/ndk/android-ndk-r28

./build.sh
# Output: android_transcribe_app_release.apk

Release AAB (Google Play Ready)

1. Run the AAB Build Script:

   ./build_aab.sh

   This will:

   • Build the Rust library in release mode.
   • Create a base module for the app code.
   • Create a model_assets module for the large model files (Install-Time Asset Pack).
   • Generate android_transcribe_app.aab.
   • Sign it with a generated release.keystore

   Output: android_transcribe_app.aab

2. Testing the AAB on a Device: You cannot install an .aab directly. Use bundletool (downloaded to libs/ by the script):

   # 1. Generate APKs from the bundle
   java -jar libs/bundletool.jar build-apks \
       --bundle=android_transcribe_app.aab \
       --output=android_transcribe_app.apks \
       --ks=release.keystore \
       --ks-pass=pass:password \
       --ks-key-alias=release \
       --key-pass=pass:password \
       --overwrite
   
   # 2. Install to connected device
   # Ensure ADB is in your PATH or provide it via --adb
   java -jar libs/bundletool.jar install-apks \
       --apks=android_transcribe_app.apks \
       --adb=android-sdk/platform-tools/adb

Acknowledgments

• Speech Model: Parakeet TDT 0.6b v3 by NVIDIA.

  • ONNX quantization provided by istupakov.
  • Licensed under CC-BY 4.0.

• Inference Backend: transcribe-rs by CJ Pais.

License of this Project

MIT