# Aixora Studio 0.7.6

## 中文

- 新增实时翻译：支持本地语音识别、翻译和朗读，可选择输入与输出设备、译文声音，并下载原声或译文 WAV 音频。
- 同步上游后端，新增 LFM3-Audio、Gradium Phonon、Magenta RealTime 2、Earshot VAD，以及 Underdog Bark 3.0 系列语音识别模型，完善工作台中的模型参数与功能适配。
- 更新 CPU、CUDA 和 Vulkan 三套运行后端，保留原有显卡支持范围，合入多项推理性能优化及兼容性修复。
- 修复首次打开工作台默认显示中文的问题：根据浏览器首选语言选择中文或英文，保留已有语言设置，并新增侧栏 EN／中文快捷切换按钮。
- 优化实时翻译的设备选择与结果显示，将“实时翻译”卡片移至场景应用最后。
- 修复已知问题，改善使用体验。

已安装 0.7.0 及以上整合包的用户，关闭程序后运行根目录的 `update.bat` 即可升级。更新会保留已下载模型、自定义声音、生成结果和个人配置。0.6.x 及更早版本需先安装 0.7.0 完整包，再运行更新。

## English

- Added live translation with local speech recognition, translation and spoken output, selectable input/output devices and translation voices, plus WAV downloads for original and translated audio.
- Synced upstream changes and added LFM3-Audio, Gradium Phonon, Magenta RealTime 2, Earshot VAD, and the Underdog Bark 3.0 ASR models, with updated model parameters and workbench integration.
- Updated the CPU, CUDA and Vulkan runtimes while retaining the existing GPU support range, including inference performance improvements and compatibility fixes.
- Fixed the workbench always starting in Chinese on first visit. It now selects Chinese or English from the browser's preferred language, preserves existing language preferences, and provides an EN / 中文 shortcut in the sidebar.
- Improved live translation device selection and result display, and moved the Live Translation card to the end of Scenarios.
- Fixed known issues and improved the user experience.

For portable installations based on 0.7.0 or later, close the app and run `update.bat` in the package root to upgrade. Downloaded models, custom voices, generated outputs and personal settings are preserved. Versions 0.6.x and earlier must first install the complete 0.7.0 package, then run the updater.