ReaLingo v0.6.0

新增

• 接入 OpenAI 与 Google Gemini — 设置 → 接口新增 OpenAI（gpt-realtime-translate，输出 13 种语言）和 Gemini Live Translate（gemini-3.5-live-translate-preview，54 种语言，全部可朗读），与阿里云百炼的 Qwen3.5 / 3、Qwen3.8 并列可选。感谢 @zixiaoluobai 的建议（#3）。
• 自定义模型和接入地址 — 模型可从预设里选，也可以填任意模型 ID；「高级」里可填自定义地址，代理或兼容同一接口的服务都能接。切换接口时，每个接口记住自己的模型、每家记住自己的地址。
• 各家的 API Key 分开保存 — 都存在系统凭据库里，两个 Qwen 接口共用百炼的 Key。
• 关于页显示更新内容 — 检查更新后直接看到新版本（或当前版本）的新增与修复，按界面语言显示。
• 音色带介绍 — 47 种音色都有中文名和一句简介，可以按名字或描述搜索。

改进

• 设置页重新整理：分「连接 / 翻译 / 高级」三组，每项一行；接口、模型、地域都改为下拉框；外观和界面语言只保留在标题栏。
• 标题栏 logo 下方显示当前模型，点它直接打开设置。
• 百炼的 API Key、可用模型、业务空间管理三个链接按当前地域跳到对应控制台；业务空间 ID 的提示按地域区分（北京 llm-、新加坡 ws-）。
• 关闭字幕窗时自动收起下面的样式控件；音色下拉框加宽；灰色提示不再有多余空隙。
• OpenAI / Gemini 下朗读时提示「该接口的音色由服务端决定」。

修复

• 服务端拒绝会话时（例如所选地域没有该模型）直接显示错误并结束，不再一声不响地挂到超时。
• 模型字段缺失时按接口取各自的默认模型，不再错把 Qwen 的模型名发给其他接口。

功能一览

• 三种音频来源：麦克风 / 系统声音 / 本地音频文件
• 四种接口：阿里云百炼 Qwen3.5 / 3、Qwen3.8，OpenAI，Google Gemini；模型和接入地址可自定义
• 最多 60 种语言互译，源语言可自动检测
• 朗读译文：Qwen 有 47 种音色、29 种目标语言有语音输出；OpenAI / Gemini 所有目标语言都能朗读
• 独立字幕窗：置顶、双语 / 仅译文 / 仅原文、字号颜色描边对齐可调、锁定后点击穿透、右键菜单直接改
• 系统托盘、翻译历史、导出 TXT / SRT、界面中英文与浅色 / 深色主题、启动自动检查更新

安装与打开

需要一个 API Key：阿里云百炼、OpenAI 或 Google AI Studio 任选其一。首次打开后点右上角齿轮，选好接口再填入 Key。

Windows 10 / 11

1. 下载 ReaLingo_0.6.0_x64-setup.exe（推荐）或 ReaLingo_0.6.0_x64_en-US.msi，双击安装。
2. 安装包没有代码签名，SmartScreen 可能提示「Windows 已保护你的电脑」：点 更多信息 → 仍要运行。
3. 从开始菜单打开 ReaLingo。Windows 11 会把托盘图标折叠进 ^，想常驻就拖到任务栏上。

macOS 12+（Intel 与 Apple Silicon 通用）

1. 下载 ReaLingo_0.6.0_universal.dmg，打开后把 ReaLingo 拖进「应用程序」。
2. 应用没有经过 Apple 公证，首次打开会被拦下（提示「已损坏」或「无法验证开发者」）。在终端执行一次：

   xattr -dr com.apple.quarantine /Applications/ReaLingo.app

   或者先打开一次，再到 系统设置 → 隐私与安全性，在页面下方点 仍要打开。
3. 首次用麦克风 / 系统声音时会弹权限申请，点允许。系统声音需要 macOS 14.4+。
4. 从旧版本升级后如果系统声音没有反应，到 系统设置 → 隐私与安全性 → 屏幕与系统录音（以及 麦克风）里把 ReaLingo 删掉，重新打开应用再授权一次。

Linux（Ubuntu 22.04+）

1. 下载 ReaLingo_0.6.0_amd64.deb，用 apt 安装（会自动拉依赖，dpkg -i 不会）：

   sudo apt install ./ReaLingo_0.6.0_amd64.deb

2. 从应用菜单打开 ReaLingo，或在终端运行 realingo。
3. 系统声音需要 PulseAudio 或 PipeWire（Ubuntu 桌面默认就有）。托盘图标不响应左键，用右键菜单里的「显示主窗口」。

EN

New

• OpenAI and Google Gemini — Settings → API now offers OpenAI (gpt-realtime-translate, 13 output languages) and Gemini Live Translate (gemini-3.5-live-translate-preview, 54 languages, all spoken) alongside Alibaba Cloud Model Studio's Qwen3.5 / 3 and Qwen3.8. Thanks to @zixiaoluobai for the suggestion (#3).
• Custom model and endpoint — pick a preset or type any model ID; under Advanced, a custom URL lets you use a proxy or any service speaking the same API. Switching APIs remembers each one's model and each vendor's URL.
• A separate API key per vendor — all kept in the OS credential store; both Qwen APIs share the Model Studio key.
• What's new in About — after an update check, the new (or current) version's features and fixes show right there, in the interface language.
• Voice descriptions — all 47 voices have a name and a one-line description, searchable by either.

Improvements

• Settings reorganised into Connection / Translation / Advanced, one row per setting; API, model and region are dropdowns; theme and interface language stay in the title bar only.
• The current model shows under the title-bar logo; click it to open Settings.
• The Model Studio links (API key, available models, workspace management) follow the selected region; the workspace ID hint matches it (llm- for Beijing, ws- for Singapore).
• Turning the subtitle overlay off folds its style controls away; the voice picker is wider; grey hints no longer float in extra space.
• Reading aloud with OpenAI / Gemini notes that the service picks the voice.

Fixes

• When the server refuses a session (e.g. the model is not offered in that region), the error shows and the session ends, instead of hanging silently until a timeout.
• A missing model field now falls back to each API's own default, instead of sending a Qwen model name to another API.

Features

• Three audio sources: microphone / system audio / a local audio file
• Four APIs: Alibaba Cloud Model Studio Qwen3.5 / 3 and Qwen3.8, OpenAI, Google Gemini; model and endpoint are customisable
• Up to 60 languages, with automatic source detection
• Read the translation aloud: Qwen offers 47 voices and speaks 29 target languages; OpenAI / Gemini speak every target language they offer
• Standalone subtitle overlay: always on top, bilingual / translation only / source only, adjustable size, colours, outline and alignment, click-through when locked, a right-click menu for quick changes
• System tray, translation history, TXT / SRT export, Chinese / English interface, light / dark themes, update check at startup

Install

You need an API key from Alibaba Cloud Model Studio, OpenAI or Google AI Studio. On first launch, click the gear in the top-right, pick the API and paste the key.

Windows 10 / 11

1. Download ReaLingo_0.6.0_x64-setup.exe (recommended) or ReaLingo_0.6.0_x64_en-US.msi and run it.
2. The installer is not code-signed, so SmartScreen may say "Windows protected your PC": click More info → Run anyway.
3. Open ReaLingo from the Start menu. Windows 11 folds new tray icons under ^; drag it onto the taskbar to keep it visible.

macOS 12+ (universal: Intel and Apple Silicon)

1. Download ReaLingo_0.6.0_universal.dmg, open it and drag ReaLingo into Applications.
2. The app is not notarised, so the first launch is blocked ("damaged" or "unidentified developer"). Run once in Terminal:

   xattr -dr com.apple.quarantine /Applications/ReaLingo.app

   or open it once, then click Open Anyway at the bottom of System Settings → Privacy & Security.
3. Allow the microphone / system audio permission when asked. System audio needs macOS 14.4+.
4. If system audio stays silent after upgrading, remove ReaLingo under System Settings → Privacy & Security → Screen & System Audio Recording (and Microphone), then reopen the app and grant it again.

Linux (Ubuntu 22.04+)

1. Download ReaLingo_0.6.0_amd64.deb and install it with apt (which pulls in dependencies; dpkg -i does not):

   sudo apt install ./ReaLingo_0.6.0_amd64.deb

2. Open ReaLingo from the app menu, or run realingo in a terminal.
3. System audio needs PulseAudio or PipeWire (standard on the Ubuntu desktop). The tray icon ignores left clicks; use "Show main window" in its right-click menu.

Full Changelog: v0.5.1...v0.6.0