Skip to content

VideoLingo v3.1.0 🎉

Choose a tag to compare

View all tags
@Huanshere Huanshere released this 27 Sep 09:34
· 50 commits to main since this release
v3.1.0
21b1298

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.
  • The old Excel batch scripts have been removed. See the HTTP API guide 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。
  • 旧的 Excel 批处理脚本已移除；自动化用法见 HTTP API 文档。
  • 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

Assets 2
Loading