Lexi-Keyboard (言犀键盘) is an Android input method application that integrates Large Language Model (LLM) and Automatic Speech Recognition (ASR) technologies to provide a more natural and efficient voice input experience. It targets Android users seeking advanced voice-to-text capabilities, offering features like AI-powered text correction, cross-input method voice input via a floating ball, and intelligent text editing.

How It Works

Built with Kotlin for the Android platform, Lexi-Keyboard leverages multiple ASR engines, including support for local, offline models, ensuring privacy and flexibility. It processes spoken input and then applies LLM-based post-processing for corrections and enhancements. A key architectural choice is the dual-mode operation: a full keyboard input method and a floating ball that allows voice input to function seamlessly with any other Android keyboard or application, preserving existing user habits.

Quick Start & Requirements

• Installation: Download and install the latest APK from the Releases page.
• Requirements: Android 10.0 (API 29) or higher. Requires microphone permission. Floating window and accessibility permissions are optional but recommended for full functionality.
• Setup: Enable the keyboard in Android system settings. Configure an ASR service provider; VolcEngine is recommended due to its free tier.
• Links: Telegram Chat, Zread, DeepWiki.

Highlighted Details

• Hybrid Input Method: Offers a floating ball for voice input that works across any application and existing keyboard, alongside a dedicated keyboard input method.
• Advanced ASR & LLM Integration: Supports over seven mainstream ASR services, including local offline models, and utilizes LLMs for intelligent text post-processing and AI-driven editing commands.
• AI Editing Panel: A dedicated interface for voice-based text manipulation, supporting cursor movement, text selection, copy/paste, and applying AI prompts for rewriting or formatting.
• Smart Recording & Input: Features intelligent auto-stop based on silence, segmented recording for long audio inputs, and automatic insertion of recognized text.

Maintenance & Community

The project is actively maintained by BryceWG, with contributions from other developers listed in the README. A Telegram chat is available for community interaction.

Licensing & Compatibility

The project is licensed under the Apache 2.0 License, which permits free use, modification, and distribution, including for commercial purposes, provided attribution is maintained.

Limitations & Caveats

A "compatibility mode" is provided for specific applications (e.g., Telegram, Douyin) to address potential issues with background text interference during automated text insertion via accessibility services. Some ASR streaming features are noted as being in early development. The reliance on microphone, floating window, and accessibility permissions may be a consideration for some users.