Commonwealth-Bank-of-Australia/project-echo
Commonwealth Bank of Australia open-sources a pipeline that turns old VHS archive footage into searchable AI-annotated text
Commonwealth Bank of Australia (CBA) has been digitising over a century of archival video stored on degrading VHS, Betamax and other magnetic tapes, and Project ECHO is the pipeline it built to make that footage searchable. It transcribes spoken audio with speech-to-text models and describes keyframes from the video using multi-modal vision-language models (VLMs), optionally summarising everything into a narrative. The repository is published for research, educational and community purposes only and is explicitly not CBA's production system.
What it does
- Built a pipeline that converts audio and visual content from aging archival tapes (VHS, Betamax) into searchable text as the tapes are digitised
- Uses a Whisper-compatible speech-to-text server to transcribe audio, ffmpeg to extract representative keyframes at scene changes, a vision-language model (VLM) to describe those frames, and optionally an LLM to write a narrative summary combining transcript and descriptions
- Storage (local filesystem or S3) and database (SQLite or DynamoDB) backends are configured independently, letting it run from a laptop to production-scale infrastructure
- Works with any OpenAI-compatible endpoint, including self-hosted vLLM deployments, not just OpenAI's API; results are always saved to the database and optionally also written as transcript/summary files
- Includes an opt-in cluster-aware scene detection mode that groups rapid scene-change bursts (e.g. a title card flashing before the real content) and describes them together in one call, instead of possibly describing the wrong frame
Why it matters
It's a concrete example of digitising a large, decaying archive while simultaneously making its contents discoverable, rather than just preserving files. The combination of speech recognition and vision-language models for automatic archive indexing is a pattern useful for any organisation, broadcaster, or library sitting on large volumes of unsearchable video.
Terms in this repo
- VLM (Vision-Language Model) · a multi-modal AI model that can look at an image and describe it in text
- Whisper · a speech-to-text AI model that converts spoken audio into text
- vLLM · an open-source tool for hosting and serving AI models efficiently
- ffmpeg · a widely used open-source tool for cutting and converting audio/video
- DynamoDB · Amazon's cloud database service
Repository description (English)
Codebase for Project ECHO: Video Transcription and Annotation Pipeline
Open on GitHubTrending repos
- vorssaint/vorssaint-utilsOne free menu bar app replaces a dozen paid Mac utilities
- Alishahryar1/free-claude-codeA local proxy that lets coding AI agents run on 49 free or cheap model providers instead of one paid service
- freestylefly/awesome-gpt-image-2A library of 532 reverse-engineered prompts that turn GPT-Image2 into a predictable image-making tool
- block/buzzAn open-source workspace where humans and AI agents chat, code, and review in the same rooms
- NousResearch/hermes-agentNous Research's Hermes is an AI agent that gets smarter the more you use it
- virgiliojr94/book-to-skillA tool that turns technical book PDFs into on-demand reference skills for AI coding agents
- VoltAgent/awesome-agent-skillsA single hub collecting over 1000 'how-to' manuals that make AI coding assistants act like experts
- anthropics/claude-plugins-communityA shared shelf where anyone's Claude add-ons get listed for install