BiBi Keyboard

AI-powered Smart Voice Input Method | Make Voice Input More Natural and Efficient

🌐 Official Website • 📖 Documentation

English | 简体中文

Features • UI Showcase • Quick Start

📱 UI Showcase

✨ Features

🌟 Pro Version Released

💎 BiBi Keyboard Pro is now officially available on the Play Store for a one-time purchase of just $5.49!

The Pro version offers more advanced features and a better experience (hotwords, Simplified/Traditional conversion, continuous dictation, WebDAV auto-backup, and more). You can learn more in About → Learn about Pro, or in the Pro Features documentation. We welcome your feedback to help us polish a better product!

If you are interested in BiBi Keyboard, join our Telegram Group for more information.

🚀 Quick Start

Provider Configuration Guide

📋 System Requirements

• Android 8.0 (API 26) or higher
• Microphone permission (for voice recognition)
• Overlay permission (optional, for floating ball feature)
• Accessibility permission (optional, for automatic text insertion)
• LSPosed / LSPatch (optional, for integration with any third-party IMEs)

📥 Installation Steps

1. Download & Install

   • Download the latest APK from Releases page
   • Install on your Android device

2. Enable Input Method

   Settings → System → Language & Input → Virtual Keyboard → Manage Keyboards → Enable "BiBi Keyboard"
   

3. Configure ASR Service

   • Open BiBi Keyboard settings
   • Select ASR vendor (recommended: Volcengine)
   • Enter your API key

4. Start Using

   • Switch to BiBi Keyboard in any input field
   • Long press the microphone button to start voice input

💡 Tip: First-time users are recommended to configure Volcengine for 20 hours of free quota!

🎨 Tech Stack

Kotlin 2.3.21
Android SDK (Compile SDK 37, Target SDK 35, Min SDK 26)
Jetpack Compose + Material Design 3 / Miuix
Coroutines (async processing)
OkHttp 5.3.2 (network requests)
SharedPreferences (data storage)
sherpa-onnx (local ASR models)

📄 License

This project is licensed under the Apache 2.0 License, see the LICENSE file for details.

Apache 2.0 License - Free to use, modify, distribute, must retain copyright notice

Star History

☕ Support & Appreciation

If this project helps you, please give it a Star ⭐️ and feel free to buy me a coffee ☕️

_{Scan with WeChat to appreciate}

🙏 Acknowledgments

Thanks to the following open-source projects for their technical support:

• sherpa-onnx - Provides technical support for local ASR models, making offline voice recognition possible
• TEN-VAD - Provides the current VAD model support
• SyncClipboard - Provides clipboard synchronization backend service (not running locally, requires server)
• Phosphor - Provides almost all icons used in the software
• miuix - Provides Miuix-style Compose UI components, icons, and preference components
• WaveLineView - Provides beautiful wave animation implementation for audio visualization, making recording experience more fluid and visually appealing
• Thanks to the author NURKE(南㲺) of the "Completion Project" icon pack for designing the new application icon for this project

📝 更新日志 / Changelog (v4.5.0)

中文

新增功能

• AI 自动选择润色预设：润色前可根据识别内容自动匹配最合适的 Prompt 预设；可为预设填写 Skill 描述，支持独立分类模型，并用历史记录预览效果；内容已足够好时也可自动跳过润色
• 录音时屏幕常亮：键盘或悬浮球录音时可保持屏幕常亮，避免中途熄屏；也可通过键盘扩展按键开关
• DashScope Qwen3.8-Omni-Flash：DashScope 新增 Qwen3.8-Omni-Flash 非流式模型
• StepAudio 新模型：StepAudio 新增 stepaudio-3-asr-max，并支持自定义识别模型
• [Pro] 助手与热词分析独立模型：AI 助手和热词纠正分析可分别指定模型，不必跟随默认润色模型

改进优化

• 在线非流式分段识别：在线文件识别也会按静音分段渐进识别，长录音更快出字；失败片段会自动重试
• Prompt 全屏编辑：润色 Prompt 支持全屏编辑
• AI 更多参数：温度等高级参数收入独立面板，设置页更清爽
• [Pro] 输入框上下文：开启后，自动选择润色预设也会参考光标前后文本

问题修复

• 折叠屏与旋转：修复旋转或折叠后悬浮球、悬浮键盘位置与可用区域不准的问题
• 录音音源：缓存已验证的录音源，减少部分机型录音失败或反复探测音源的问题
• 备份导入：提高识别历史备份导入兼容性，避免新旧版本字段不兼容导致导入失败

English

New Features

• AI Auto Prompt Selection: Before polishing, automatically pick the best Prompt preset for the transcript. Presets can include a Skill description; you can choose a separate classifier model and preview with history. High-quality transcripts can skip polishing
• Keep Screen On While Recording: Keeps the screen on during keyboard or floating-ball recording so it will not sleep mid-session; also available as a keyboard extension toggle
• DashScope Qwen3.8-Omni-Flash: DashScope adds the Qwen3.8-Omni-Flash non-streaming model
• StepAudio Models: StepAudio adds stepaudio-3-asr-max and custom recognition models
• [Pro] Separate Assistant and Hotword Models: AI assistant and hotword-correction analysis can each use their own model instead of the default polish model

Improvements

• Online Non-Streaming Chunking: Online file recognition now chunks on silence for faster first results on long recordings; failed segments are retried automatically
• Full-Screen Prompt Editor: Polish prompts can be edited full screen
• AI Extra Parameters: Temperature and other advanced parameters move into a dedicated sheet
• [Pro] Input Field Context: When enabled, automatic prompt selection also uses text around the cursor

Bug Fixes

• Fold and Rotation: Fix floating ball and floating keyboard position and usable area after rotation or folding
• Recording Source: Cache the validated capture source to reduce recording failures or repeated source probing on some devices
• Backup Import: More compatible recognition-history backup import when newer enum fields are missing