说点啥
正式版
7
查看
0
讨论
20
收藏
photo_library 应用截图
new_releases 更新日志
📝 更新日志 / Changelog (v4.4.2)
中文
新增功能
Gemini Transcribe:Gemini 新增 Transcribe 调用模式
识别历史时间轴:历史详情展示录音、识别、润色与文字写入各阶段耗时;弹窗支持按高度自适应
改进优化
LLM 连通性测试:测试结果分段展示连接、首字、输出耗时与连接复用,并标明是否支持流式
识别与键盘响应:异步提交历史、缓存键盘首帧 UI,并更广泛地复用网络连接池,识别延迟更低、上屏更少卡顿、打开更快
AI 润色文案:界面将「后处理」统一为「润色」
打字机效果:关闭后改为润色完成后一次性上屏
Qwen3-ASR 1.7B 结果清洗:去掉模型标记,避免内部标签写入正文
收起键盘中断润色:隐藏键盘时取消进行中的 AI 润色并提交原文,避免下次按麦无响应
润色超时与取消:为 AI 润色加上超时上限;取消后不再错误走非流式回退
[Pro] 热词预加载:启动时预构建音素索引,降低首次识别的后处理延迟
问题修复
Bridge 模块:优化与 Bridge 模块的通信管理
后台通知语言:保活与悬浮等通知将跟随应用语言
悬浮球性能:未录音时不再持续刷新悬浮球光环特效
English
New Features
Gemini Transcribe: Gemini adds a Transcribe call mode
History Timing Timeline: History details show time spent on recording, recognition, polish, and text commit; the dialog adapts to available height
Improvements
LLM Connectivity Test: Test results show connect, first-token, and output latency, connection reuse, and whether streaming is supported
Faster Recognition and Keyboard: Async history commit, cached first-frame UI, and broader HTTP connection-pool reuse for lower recognition latency, less commit jank, and faster keyboard open
AI Polish Wording: UI now uses “polish” instead of “post-processing”
Typewriter Effect: When off, the full result is applied at once after polish finishes
Qwen3-ASR 1.7B Cleanup: Strips model markers so they are not committed as text
Cancel Polish on Keyboard Hide: Hiding the IME cancels in-flight AI polish and commits the raw transcript, so the next mic tap is not ignored
Polish Timeout and Cancel: AI polish now has a timeout cap; cancelled requests no longer fall back to non-streaming
[Pro] Hotword Preload: Builds phoneme indexes at startup to reduce first-recognition post-processing delay
Bug Fixes
Bridge Module: Improved communication management with the Bridge module
Notification Language: Keep-alive, floating, and similar notifications now follow the app language
Floating Ball Performance: Aura effects no longer keep refreshing when not recording
description 应用介绍
Lexi‑Keyboard 是一款开源 Android 键盘应用,集成语音识别(ASR)与大语言模型(LLM)文本编辑功能。用户可通过悬浮球或键盘直接进行语音输入,并享受智能文本处理与编辑体验。支持本地离线识别、多语言输入、AI 文本润色与纠错,同时保留你熟悉的输入法操作习惯。
核心功能:
- 长按录音键语音输入,支持智能静音判停
- 悬浮球模式,可在任意输入法下使用语音输入
- AI 文本后处理:润色、纠错、智能编辑
- Material 3 风格设计,支持主题色彩自定义
- 支持中英文及多语言输入,键盘高度和振动反馈可调
适用场景:
- 日常聊天、写文档、发邮件时快速语音输入
- 希望结合 AI 辅助提升文字质量
- 想在 Android 上体验语音 + 智能文本编辑混合输入
📦 开源地址:https://github.com/BryceWG/Lexi-Keyboard