Video Subtitle Pipeline (Go + Whisper + LibreTranslate)

A simple backend project that processes a local video and produces a subtitle-burned output video.

Pipeline steps:

1. Extract audio from input video with FFmpeg
2. Transcribe audio with OpenAI Whisper (Python script)
3. Generate SRT subtitles
4. Translate SRT lines using LibreTranslate-compatible API
5. Burn translated subtitles into the output video with FFmpeg

Architecture

The project follows a lightweight clean architecture split:

• cmd/subtitle: CLI entrypoint and wiring
• internal/usecase: pipeline orchestration
• internal/ports: service interfaces (video/transcription/translation/SRT)
• internal/infrastructure: concrete implementations
  • video: FFmpeg integration
  • transcription: Whisper script wrapper
  • translation: LibreTranslate script wrapper
  • srt: SRT writer
• scripts/: Python scripts used by Go services

Prerequisites (local run)

• Go 1.22+
• Python 3.11+
• FFmpeg installed and available in PATH

Python dependencies:

pip install -r requirements.txt

CLI usage

go run ./cmd/subtitle \
  -input ./samples/input.mp4 \
  -output ./samples/output_es.mp4 \
  -source-lang auto \
  -target-lang es

Available flags:

• -input (required): input video path
• -output: output video path (default output_subtitled.mp4)
• -source-lang: source language code for translation (default auto)
• -target-lang: target language code (default es)
• -workdir: temp working directory (default uses system temp)
• -keep-artifacts: keep intermediate WAV and SRT files
• -whisper-model: Whisper model name (default base)
• -translate-endpoint: translation API endpoint (default https://libretranslate.com/translate)

Docker

Build:

docker build -t subtitle-pipeline .

Run (single command):

docker run --rm \
  -v "$PWD:/data" \
  subtitle-pipeline \
  -input /data/input.mp4 \
  -output /data/output_es.mp4 \
  -target-lang es

Note: Whisper downloads model weights on first run, which can take time.

Notes

• Translation is line-based on subtitle text lines to keep implementation simple.
• You can replace -translate-endpoint with your own LibreTranslate-compatible service.