OpenWhispr Brings Local and Cloud AI Dictation to Android

Summary

OpenWhispr is a free, open-source Android app that provides push-to-talk dictation through a floating overlay and inserts the result into the currently focused text field. It can transcribe speech on the device with sherpa-onnx models, requiring no API key or internet connection, or send audio to Groq's Whisper-large-v3 endpoint using the user's own API key. An optional Groq-hosted language model removes filler words, corrects punctuation and grammar, formats emails, and can follow custom cleanup instructions. The project is a fork of kafkasl/phone-whisper: this version replaces OpenAI cloud services with Groq, changes transcription to Whisper-large-v3, uses openai/gpt-oss-120b for cleanup, and adds reliability and interface improvements. These include focus-aware overlay visibility, foreground-service and battery-optimization handling, crash defenses, in-app release updates, organized setup settings, and a pause switch that does not revoke Accessibility permission. An optional Voice Commands mode recognizes a configurable trigger phrase and supports only summarizing, flow enhancement, translation, tone changes, and list formatting, including chained operations. Text insertion depends on Android Accessibility and standard text fields; unsupported apps may receive the result through the clipboard instead. Local model choices range from 100 MB to 465 MB, while cloud mode requires a Groq key. The repository provides release APKs and build instructions requiring JDK 17 and the Android SDK.