[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-75128":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":8,"htmlUrl":8,"language":8,"languages":8,"totalLinesOfCode":8,"stars":9,"forks":10,"watchers":11,"openIssues":12,"contributorsCount":13,"subscribersCount":13,"size":13,"stars1d":14,"stars7d":11,"stars30d":15,"stars90d":13,"forks30d":13,"starsTrendScore":16,"compositeScore":17,"rankGlobal":8,"rankLanguage":8,"license":18,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":21,"hasPages":19,"topics":22,"createdAt":8,"pushedAt":8,"updatedAt":23,"readmeContent":24,"aiSummary":25,"trendingCount":13,"starSnapshotCount":13,"syncStatus":14,"lastSyncTime":26,"discoverSource":27},75128,"voice-input-src","yetone\u002Fvoice-input-src","yetone",null,2202,233,4,8,0,2,26,6,29.11,"MIT License",false,"master",true,[],"2026-06-12 02:03:33","**[中文](README_CN.md) | English**\n\n## Source Code\n\n```bash\nclaude \\\n  --dangerously-skip-permissions \\\n  --output-format=stream-json \\\n  --verbose \\\n  -p \"Please implement a macOS menu-bar voice input app (Swift, macOS 14+) with the following requirements:\n\n1. Hold the Fn key to record, release to inject the transcribed text into the currently focused input field. Use streaming transcription (Apple Speech Recognition framework) as preferred approach. Monitor Fn key globally via CGEvent tap, suppressing the Fn event to prevent triggering the emoji picker.\n2. Default language must be Simplified Chinese (zh-CN), ensuring Chinese input recognition works out of the box. Also provide language switching options in the menu bar (English, Simplified Chinese, Traditional Chinese, Japanese, Korean). Language selection is stored in UserDefaults.\n3. While recording, display an elegant frameless capsule-shaped floating window centered at the bottom of the screen — no traffic lights or titlebar. Use NSPanel (nonactivatingPanel) + NSVisualEffectView (.hudWindow material), sufficient height (56px, corner radius 28px), containing:\n   - 5 vertical bar waveform animation on the left (44×32px), driven by real-time audio RMS levels (no hardcoded fake animations) — louder speech produces larger waveforms, quiet moments produce smaller ones. Bar weights are [0.5, 0.8, 1.0, 0.75, 0.55] creating a natural center-high, sides-low effect. Smooth envelope (attack 40%, release 15%), add ±4% random jitter per bar for organic feel. Waveforms should be large enough to be clearly visible.\n   - Text label on the right (elastic width 160-560px) showing real-time transcription, capsule elastically widens as text grows\n   - Entry spring animation (0.35s), text width smooth transition (0.25s), exit scale animation (0.22s)\n4. Text injection uses clipboard + simulated Cmd+V paste. Before injection, detect the current input method: if it is a CJK input method, temporarily switch to an ASCII input source (ABC\u002FUS keyboard) before pasting, then restore the original input method after paste completes — this prevents CJK input methods from intercepting Cmd+V. Restore original clipboard contents after injection.\n5. Integrate LLM to improve speech recognition accuracy, especially for mixed Chinese-English scenarios. Use an OpenAI-compatible API (configurable API Base URL, API Key, Model) to refine transcribed text. The LLM system prompt must be very conservative in corrections: only fix obvious speech recognition errors (e.g., Chinese homophone errors, English technical terms mistakenly converted to Chinese like 配森→Python, 杰森→JSON). Never rewrite, polish, or remove any content that appears correct — if the input looks correct, return it as-is.\n6. Provide an LLM Refinement submenu in the menu bar with an enable\u002Fdisable toggle and a Settings entry. The Settings window contains three input fields: API Base URL, API Key, Model — the API Key field must support being fully cleared — plus Test and Save buttons. After releasing Fn, if LLM is enabled and configured, the floating window shows a Refining... status, waiting for the LLM response before injecting the final text.\n7. The app runs in LSUIElement mode (menu bar icon only, no Dock icon). Build with Swift Package Manager, provide a Makefile (build\u002Frun\u002Finstall\u002Fclean), build output is a signed .app bundle.\"\n```\n\n## Dist\n\nhttps:\u002F\u002Fgithub.com\u002Fyetone\u002Fvoice-input-dist\n","该项目开发了一款适用于macOS 14+的菜单栏语音输入应用。核心功能包括通过长按Fn键进行录音，并在释放时将转录文本注入当前焦点输入框，使用Apple Speech Recognition框架实现流式转录。支持简体中文默认识别及其他多种语言切换，实时显示优雅的无边框胶囊状浮动窗口展示波形动画和转录文本。该应用还集成了LLM来提高语音识别准确性，特别是在中英文混合场景下。适合需要高效语音输入且对界面美观有要求的用户使用，如程序员、作家等经常需要快速录入文字的专业人士。","2026-06-11 03:52:25","high_star"]