Video Summarizer

A containerized, microservices-based video processing pipeline that extracts audio from videos and transcribes them using AI. Services are orchestrated via Docker Compose with N8N handling workflow automation.

Architecture

Google Drive (video)
    → N8N: write to /tmp
    → Converter (FFmpeg) → .mp3 → /tmp
    → Transcript (Whisper) → JSON segments
    → N8N: AI Agent (Groq) → Markdown summary
    → Google Drive (Markdown file)

N8N Workflow

The pipeline runs as an N8N workflow with the following steps:

#	Step	Description
1	Download file	Downloads the source video from Google Drive
2	Write to disk	Saves the binary to /tmp for shared volume access
3	Convert video to mp3	Calls the Converter service to extract audio
4	Transcribe video	Calls the Transcript service to produce a text transcript
5	Edit Fields	Maps and reshapes the transcript response fields
6	AI Agent (Groq)	Sends the transcript to a Groq Chat Model to generate a structured summary
7	Create Markdown in Drive	Writes the AI-generated summary as a Markdown file to Google Drive

Services

Service	Port	Technology	Role
N8N	5678	N8N	Workflow orchestration
Converter	5001	Go 1.25, Gin, FFmpeg	Extracts mono MP3 audio from video
Transcript	5002	Python 3.11, FastAPI, faster-whisper	Transcribes audio to text

Prerequisites

• Docker and Docker Compose
• A HuggingFace account and access token (for Whisper model downloads)
• A Groq account and API key (for the AI summarization step)
• A Google account with Google Drive access (configured as N8N credentials)

Getting Started

1. Clone the repository

git clone <repository-url>
cd video-summarizer

2. Set up environment variables

Create a .env file in the project root:

HF_TOKEN=your_huggingface_token_here

3. Start the services

docker compose up --build

4. Access N8N at http://localhost:5678 to configure your workflow.

API Reference

Converter Service — GET /convert/:filename

Reads a video file from /tmp, extracts audio, and writes a mono MP3 at 16kHz/32kbps back to /tmp.

Example:

GET http://localhost:5001/convert/video.mp4

Response:

{
  "output": "video.mp3",
  "path": "/tmp/video.mp3"
}

---

Transcript Service — GET /transcribe/{filename}

Reads an MP3 from /tmp and returns a timestamped transcription with language detection.

Example:

GET http://localhost:5002/transcribe/video.mp3

Response:

{
  "filename": "video.mp3",
  "language": "Portuguese",
  "language_iso": "pt_BR",
  "language_probability": 0.9981,
  "transcript": "Segment one text.\nSegment two text."
}

Supported languages: Portuguese, Spanish, English, French.

Configuration

Variable	Service	Default	Description
HF_TOKEN	Transcript	—	HuggingFace token for model download
WHISPER_MODEL	Transcript	medium	Whisper model size
API_PORT	Converter	5001	HTTP port override

Project Structure

video-summarizer/
├── converter/          # Go-based video-to-audio converter
│   ├── main.go
│   ├── go.mod
│   └── Dockerfile
├── transcript/         # Python-based audio transcription
│   ├── app.py
│   ├── requirements.txt
│   └── Dockerfile
├── videos/             # Local video file storage
├── docker-compose.yml
└── .env