This project provides an Android Input Method Editor (IME) and system-wide voice input service powered by OpenAI's Whisper. It targets Android users seeking efficient, offline voice-to-text capabilities for seamless integration with other applications or as a standalone dictation tool, offering privacy and convenience.

How It Works

WhisperIME leverages TFLite models of OpenAI's Whisper for on-device, offline voice recognition. Users can select between a fast, English-only model or a slower, multilingual model. The system integrates as an Android RecognitionService, allowing it to be chosen as the default voice input method, and supports standard Android intents for speech recognition.

Quick Start & Requirements

• Install via F-Droid or direct APK download.
• Initial launch requires internet to download Whisper TFLite models (~435 MB).
• Voice recognition functions entirely offline after model download.
• For system-wide voice input, configure via Android settings: System > Languages > Speech > Voice Input.

Highlighted Details

• Functions as both an IME and a system-wide voice input service.
• Supports standalone app usage with translation to English.
• Offline processing ensures privacy and convenience.
• Offers a choice between English-only (faster) and multilingual (broader language support) Whisper models.

Maintenance & Community

The project is maintained by woheller69. Further details on community or roadmap are not explicitly provided in the README.

Licensing & Compatibility

Licensed under MIT. The project utilizes components also under MIT and Apache-2.0 licenses. Compatible with commercial use and closed-source applications.

Limitations & Caveats

Each recording is limited to 30 seconds. The multilingual model is noted as significantly slower than the English-only option.