BryceWG/BiBi-Keyboard

说点啥（BiBi Keyboard）:一个基于 Kotlin 的 Android 平台的 LLM 与 ASR 语音输入法键盘应用 An LLM ASR voice input method keyboard application for the Android platform based on Kotlin

What it solves

BiBi Keyboard is an AI-powered voice input keyboard for Android. It solves the problem of slow and error-prone typing by letting users speak naturally, especially in Chinese and other languages. It also addresses unreliable network conditions by supporting on-device speech recognition, and it helps clean up rough speech-to-text results with AI polishing. Additionally, it allows voice input to work inside any app via a floating ball, not just within its own keyboard.

How it works

The keyboard listens through a long-press on the microphone button and automatically stops when you pause speaking. It sends the audio to one of 18 supported speech recognition (ASR) providers, of which 12 are cloud-based and 6 run locally on the device (e.g., SenseVoice, FunASR Nano, Qwen3-ASR, Parakeet). Offline models are powered by sherpa-onnx. The recognized text can then be refined by an LLM to fix grammar or make it more natural. Results go straight into whatever text field you are using. The app also provides a floating ball mode, so you can record voice over any input method, and an IME Bridge (via LSPosed/LSPatch) that lets third-party input methods send audio to this app for recognition. An AI editing panel lets you use voice commands to edit the text you have typed.

Who it's for

This project is for Android users who want faster, hands-free text input, especially speakers of Chinese (simplified/traditional), English, Japanese, and Arabic. It appeals to privacy-conscious users who want local, offline ASR options, power users who want a voice input layer that works with any input method or app, and developers or enthusiasts interested in the underlying open-source speech recognition and LLM integration stack.

Highlights

  • Supports 18 ASR vendors, including 6 local/on-device models for offline use.
  • Automatic silence detection stops recording without any extra tap.
  • Optional LLM-based text polishing to fix recognition errors and make text read better.
  • Floating ball mode lets you dictate in any app without switching keyboards.
  • Compatibility with Telegram, TikTok, and other special apps via a compatibility mode.
  • IME Bridge lets third-party input methods (like modified Sogou or Tongwen) call its recognition directly.
  • AI editing panel with voice instructions to modify text.
  • Syncs clipboard history with SyncClipboard and supports backup/restore.
  • Modern UI built with Jetpack Compose / Material3 / Miuix, with Monet and Miuix theming.
  • Works on Android 8.0 and above, and is open-sourced under Apache 2.0.
  • Includes a Pro version with extra features like hotwords, simplified/traditional conversion, and WebDAV auto-backup.

Related

  • Project
    DevEmperor/DictateKeyboard

    An AI-powered Android keyboard that enables high-accuracy dictation and transcription across any app using customizable speech engines and on-device models.

  • Project
    futo-org/voice-input

    FUTO Voice Input is an open‑source Android app that provides offline speech‑to‑text using OpenAI Whisper. It integrates with Android’s generic speech‑to‑text intents and supported keyboards, offering transcription in 17 languages. The project is maintained alongside the FUTO Keyboard app and can be built from source with Gradle.

  • Project
    Akshayykadam/PixelBoard

    PixelBoard is a modified version of Gboard that enables advanced AI features like natural voice dictation, AI writing assistance, and custom UI elements on Android devices without requiring root access.

  • Project
    woheller69/whisperIMEplus

    An Android input method editor (IME) that uses the Whisper engine to provide private, offline voice recognition and translation.

  • Project
    ximeiorg/Xime

    An Android input method based on the Rime engine that provides Wubi and Pinyin support with AI-enhanced predictive text and offline speech recognition.