Skip to content

v2.0.0 — Local transcripts, speaker labels, and Ask AI

Choose a tag to compare

View all tags
@nexmoe nexmoe released this 22 Aug 11:15
· 10 commits to main since this release
v2.0.0
2137523

VidBee 2.0 — from a downloaded file to a searchable local library

Download → transcribe locally → search a quote → ask AI for a summary.

In 1.x, VidBee saved the file and stopped there. 2.0 keeps going: speech recognition, speaker labels, clickable timestamps, and Ask AI — in the same app.

Privacy first: speech and files stay on your computer.
Audio is never uploaded to a transcription service. When you run Ask AI, only the transcript text goes to the provider you chose (OpenAI, Ollama, LM Studio, …).

VidBee 2.0 workflow: download, transcribe, search keywords, ask AI

Before v1.x: a downloaded file. After v2.0: a searchable local library

中文说明在下方。 · Download VidBee

What's new

Four steps: download, local transcription, keyword search, Ask AI

1. Local transcription — Whisper, SenseVoice, and more

Turn a finished download — or a file already on disk — into a transcript on this computer.

  • Whisper — multilingual
  • SenseVoice — Chinese, Japanese, Korean, English (including Cantonese)
  • Parakeet — strong English
  • Qwen3-ASR — highest Chinese accuracy

VidBee can recommend a model for your machine and language. Download or switch models from Settings. After the model is on disk, transcription can run offline.

2. Speaker labels

VidBee separates overlapping voices and shows who is talking, with speaking-time bars under the player.

  • Detect speakers automatically
  • Or set a fixed speaker count and re-label only — the recognized words stay as they are
  • Click a speaker to jump to their next line

3. Timestamps you can search and click

Every line has a timestamp aligned with playback.

  • Search spoken text or a speaker name
  • Matches highlight in place
  • Click a line (or a word) to play from that exact moment
  • Copy a passage, export plain text / Markdown / subtitles, or burn captions into a new video

4. Ask AI on the transcript

Open a transcript and run a prompt without leaving VidBee.

Built-in prompts: Bullet Points, Improve Grammar, Generate FAQ, Extract Statistics, Paraphrase, Mind map, Translate. You can also write your own prompts and keep them in the app.

Providers: OpenAI, Anthropic, Google, DeepSeek, Groq, Azure, Hugging Face, OpenRouter, xAI, Ollama, LM Studio, or any OpenAI-compatible endpoint. API keys stay on your computer.

Speech stays local. Ask AI sends text, not the media file. Use Ollama or LM Studio if you want that step on this computer too.

Also in 2.0

  • Import local audio or video (drag and drop onto Transcripts)
  • Mini player stays available while you move between screens
  • The download engine can update itself in the background so more sites keep working
  • Transcription works on Intel Macs as well as Apple Silicon
  • Portable mode no longer resets a custom download folder outside the portable directory
  • More reliable downloads and self-hosted setups

Get 2.0

Download VidBee · Windows, macOS, and Linux builds are attached below.

Docs: Transcribe · Ask AI

Star this repo if VidBee is useful

A GitHub star is the simplest way to support continued updates — and you’ll get release notifications without delay.

⭐ Star nexmoe/VidBee

Follow along:

If you post about VidBee, mention @nexmoe so I can see it and help amplify it.

中文

VidBee 2.0 — 从「下载一个文件」到「本机可搜索的资料库」

下载 → 本地转录 → 搜索关键词 → 问 AI 总结。

1.x 下载完文件就结束了。2.0 继续往下做：语音识别、说话人标注、可点击的时间戳，以及 Ask AI — 都在同一个应用里。

隐私优先：语音和文件都在本地。
音频不会上传到任何云端转录服务。只有在你主动运行 Ask AI 时，才会把文稿文字发给你自己选择的模型（OpenAI、Ollama、LM Studio 等）。

VidBee 2.0 流程：下载、转录、搜索关键词、问 AI

2.0 新能力

1. 本地转录（Whisper / SenseVoice 等）

下载完成，或导入磁盘上已有的音视频，在这台电脑上转成文稿。

  • Whisper：多语言
  • SenseVoice：中 / 日 / 韩 / 英（含粤语）
  • Parakeet：英语更强
  • Qwen3-ASR：中文准确率最高

应用会按你的电脑和语言推荐模型，也可在设置里下载或切换。模型装好后可以离线转录。

2. 说话人识别

自动分开重叠人声，在播放器下方显示每个人的发言时间轴。

  • 默认自动识别说话人
  • 也可以指定人数，只重新标注、不改已经识别出的文字
  • 点击某个说话人，跳到他的下一句

3. 时间戳：能搜，也能点

每一行都带时间戳，和播放进度对齐。

  • 搜索说过的词，或说话人名字
  • 命中会高亮
  • 点击一行（或一个词）即可从那一时刻播放
  • 支持复制、导出纯文本 / Markdown / 字幕，或生成带字幕的新视频

4. Ask AI 工作流

打开文稿，直接跑提示词，不用离开 VidBee。

内置：要点摘要、润色语法、生成 FAQ、抽取数据、改写、思维导图、翻译。也可以自己写提示词，保存在应用里反复用。

可选 OpenAI、Anthropic、Google、DeepSeek、Groq、Azure、Hugging Face、OpenRouter、xAI，以及本地的 Ollama / LM Studio。API 密钥只存在你的电脑上。

语音留在本地。Ask AI 发送的是文稿文字，不是音视频文件。想让这一步也完全本机完成，选 Ollama 或 LM Studio 即可。

同一版本里还有

  • 可导入本地音频或视频（拖到文稿页即可）
  • 切换页面时迷你播放器会继续保留
  • 下载引擎可在后台自行更新，以便继续支持更多网站
  • Intel Mac 现在也可以转录
  • 便携版不再把你选在便携目录外的下载文件夹改回去
  • 下载与自建部署更稳定

下载 2.0

vidbee.org/download · Windows / macOS / Linux 安装包在本页下方。

文档：本地转录 · Ask AI

欢迎 Star，支持我们继续更新

如果 VidBee 对你有帮助，点一颗 Star 是最直接的支持。Star 之后也能第一时间收到新版本通知。

⭐ Star nexmoe/VidBee

如果在 X 或 Threads 提到 VidBee，请带上 @nexmoe，我看到会帮忙转发。

What's Changed

  • fix(settings): keep external download path in portable mode (#427) by @xianjianlf2 in #434
  • fix: resolve actionable download and self-hosting issues by @nexmoe in #452
  • fix(desktop): pack matching sherpa natives per platform arch by @nexmoe in #457

New Contributors

Full Changelog: v1.3.14...v2.0.0

Contributors

nexmoe and xianjianlf2
Assets 23
Loading