How sven-0414/n8n works
This repository documents a local n8n-driven media transcription pipeline. New media is detected by n8n, processed by a FastAPI service through audio extraction and speech recognition, then cleaned by an Ollama language model. Results are saved under out/ and the original is moved to done/. The tree contains outputs and configuration but no application source, so component paths are null where no implementation path is present; relationships are based on the README only.
Main entry points: out Raw Transcript
Components 11
Orchestration
- n8n Watcher
- FastAPI Service
Media Processing
- ffmpeg Extraction
- whisper.cpp
- Silero VAD
- Ollama Cleanup
Transcript Outputs
- Raw Transcript
out - SRT Subtitles
out - Cleaned Transcript
out - Processed Media
Other components
- Incoming Media
Connections 10
- Incoming Media → n8n Watcher: arrives in inbox
README.md - n8n Watcher → FastAPI Service: calls
README.md - FastAPI Service → ffmpeg Extraction: extracts audio
README.md - FastAPI Service → whisper.cpp: sends audio
README.md - whisper.cpp → Silero VAD: skips silence
README.md - whisper.cpp → Raw Transcript: writes text
README.md - whisper.cpp → SRT Subtitles: writes timestamps
README.md - FastAPI Service → Ollama Cleanup: sends transcript
README.md - Ollama Cleanup → Cleaned Transcript: writes cleaned text
README.md - n8n Watcher → Processed Media: moves original
README.md