Video Transcription Workflow
Drop a video or podcast — faster-whisper transcribes with timestamps, LLMs add speaker labels and chapter markers, output SRT + DOCX + plain text.
What does this workflow do?
System flow
音频流预处理（降噪/增益归一化/VAD 语音活性检测）→ Faster-Whisper Large 高精度转写 → 说话人聚类与分离 → 时间戳对齐 → AI 语篇分割（按话题切换自动分段）→ 内容分类（观点/背景/行动/讨论）→ 金句与关键数据提取 → 可编辑 Word 输出
Pipeline breakdown
- Pillow
pillow— Image pre-processing — deskew, denoise, crop and normalise before analysis audio → audio - Faster-Whisper
faster-whisper— Local speech-to-text transcription with timestamps and speaker turns audio → json - LLM Reasoning
llm— Large-language-model step that extracts, classifies, validates or writes structured output json → json - LLM Reasoning
llm— Large-language-model step that extracts, classifies, validates or writes structured output json → json - Python-DOCX
python-docx— Renders the final Word document with headings, tables and styling json → docx
Workflow Reputation Score
Based on executions, success rate, update frequency, component quality, and user rating.
Based on executions, success rate, update frequency, component quality, and user rating.
Components used
Image pre-processing — deskew, denoise, crop and normalise before analysis
python-pillow/Pillow ↗Local speech-to-text transcription with timestamps and speaker turns
SYSTRAN/faster-whisper ↗Large-language-model step that extracts, classifies, validates or writes structured output
FlowSync proprietaryRenders the final Word document with headings, tables and styling
python-openxml/python-docx ↗How is this different from n8n or Zapier?
Unlike generic workflow builders, FlowSync workflows ship pre-configured with AI skills baked into every node — OCR, LLM reasoning, transcription, super-resolution — not just webhook triggers. Each node is white-box auditable: you see input, output, and config. Deployment is instant — no self-hosting, no API-key per node.
Use cases
- 视频内容多平台分发
- 播客逐字稿生成
- 采访录音整理
- 在线课程字幕
Author & Source
Field notes
Descript $24/月，我一个月只出 4 期视频。这个免费版效果一样，金句提取功能 Descript 还要加钱。
采访录音自动转文字+标时间戳，写稿效率提升太多了。
Data handling
FlowSync Workflow Marketplace · 69 reputation score · 25% execution · 96% success · 5-node white-box pipeline · maintained by FlowSync Official
Ready to run?
Drop a video or podcast — faster-whisper transcribes with timestamps, LLMs add speaker labels and chapter markers, output SRT + DOCX + plain text.