WhisperIME
An open-source Android voice typing app that runs Whisper on-device, works offline after downloading models, and supports system-wide speech input.
WhisperIME turns speech into text on your Android device using OpenAI Whisper. It's for people who want private dictation in their keyboard or a separate app for transcription and translation. Speech recognition runs entirely offline after an initial model download from Hugging Face, so audio stays on your device during recognition.
The app works as an Android input method and can provide dictation through the microphone button in HeliBoard. It also supports Android's system-wide voice input service, letting other apps use it for speech recognition. Apps that request speech input through Android's standard interface can call it too.
You can choose a compact English-only model for faster recognition or a multilingual model that supports more languages but runs much more slowly. The model choice applies across the standalone app and keyboard use. In the standalone app, WhisperIME can also translate speech in supported languages into English.
WhisperIME is open source under the MIT license and builds on the Whisper-Android project. It uses Whisper models in TFLite format and Android VAD for voice activity detection. Its Chinese text conversion component uses Opencc4j.
