v3.1.0

# VideoLingo v3.1.0 🎉

Easier setup, Qwen speech recognition by default, and a local API for automation.

### 🚀 What’s new

- **Qwen3-ASR is now the default local speech recognition engine**, with Qwen3-ForcedAligner for word timing. The 1.7B model is selected by default; a lighter 0.6B model is also available.
- **Getting started is simpler.** On Windows, download and extract the Release ZIP, then double-click `OneKeyStart.bat`. It prepares uv, Python 3.12, dependencies, and FFmpeg before opening VideoLingo in your browser. For source installations, run `uv run start.py`.
- **FFmpeg and ffprobe are prepared automatically.** No separate download or PATH setup is needed.
- **A local HTTP API replaces the Excel batch workflow.** Start it with `uv run start.py --api` to let scripts and agents use the same processing pipeline as the web interface.
- **Speech recognition and subtitle timing are more reliable.** This release improves language detection, word timing, punctuation handling, and error messages. It also fixes an issue where intro music could make the first subtitle appear several seconds too early.
- The example LLM settings now use OpenLux and GPT-6 Luna; you can change them in the sidebar.

### ⚠️ Before upgrading

- VideoLingo now uses **Python 3.12**. An older environment may need to be recreated; the first setup requires an internet connection.
- WhisperX is no longer installed automatically. Configurations without an explicit ASR backend now use Qwen. To keep using WhisperX, follow the [manual installation guide](https://github.com/Huanshere/VideoLingo/blob/main/docs/pages/docs/whisperx-manual.en-US.md).
- The old Excel batch scripts have been removed. See the [HTTP API guide](https://github.com/Huanshere/VideoLingo/blob/main/docs/api.md) for automation.
- Intel Macs use CPU speech recognition. By default, dubbed videos on Intel Macs contain the new voice without the original background audio.

---

# VideoLingo v3.1.0 🎉

安装更简单，默认使用千问语音识别，也能通过本地 API 交给脚本或 Agent 使用。

### 🚀 这次更新

- **默认使用 Qwen3-ASR 进行本地语音识别**，并通过 Qwen3-ForcedAligner 对齐词语时间。默认选择 1.7B 模型，也可以切换到更轻量的 0.6B。
- **安装步骤更简单了。** Windows 用户下载并解压 Release 的源码 ZIP，双击 `OneKeyStart.bat` 即可。它会准备 uv、Python 3.12、依赖和 FFmpeg，随后在浏览器中打开 VideoLingo。从源码安装则运行 `uv run start.py`。
- **FFmpeg 和 ffprobe 会自动准备**，无需单独下载或配置 PATH。
- **本地 HTTP API 取代了原来的 Excel 批处理。** 使用 `uv run start.py --api`，脚本和 Agent 就可以调用与网页相同的处理流程。
- **语音识别和字幕时间更稳了。** 改进了语言判断、词语时间戳、标点处理和错误提示；同时修复了片头音乐可能导致第一条字幕提前数秒出现的问题。
- 示例 LLM 配置更新为 OpenLux 和 GPT-6 Luna，也可以在侧边栏自行更换。

### ⚠️ 升级前留意

- 现在统一使用 **Python 3.12**。旧环境可能需要重新创建，首次准备环境需要联网。
- WhisperX 不再自动安装。未明确指定识别后端的旧配置现在会使用千问；想继续使用 WhisperX，请参考[手动安装指南](https://github.com/Huanshere/VideoLingo/blob/main/docs/pages/docs/whisperx-manual.zh-CN.md)。
- 旧的 Excel 批处理脚本已移除；自动化用法见 [HTTP API 文档](https://github.com/Huanshere/VideoLingo/blob/main/docs/api.md)。
- Intel Mac 使用 CPU 识别。默认情况下，Intel Mac 生成的配音视频只保留新配音，不保留原视频背景音。

Full Changelog / 完整改动：[v3.0.4...v3.1.0](https://github.com/Huanshere/VideoLingo/compare/v3.0.4...v3.1.0)