🤔 Ever wondered how to turn YouTube videos into concise summaries? We have created a Google Colab notebook to offer a behind-the-scene look at the necessary steps: 🎞️ Download video from YouTube 🎶 Isolate audio track 📄 Transcribe the audio into text 📑 Generate summary of key points (with Google Gemini) While this demonstration offers a basic framework, there's ample room for customization and enhancement.

This is great! If you want to expand on this and see the knowledge graph of the transcript, you can drop a YouTube link (or other data) into https://llm-graph-builder.neo4jlabs.com/. It will parse the transcript into a knowledge graph and create embeddings for gen-ai applications.

@Alison Cossette This notebook was really about highlighting the different steps required to generate a summary from a video. Obviously, the prompt could be modified to perform entity / relationship extraction to populate a knowledge graph. Neo4J provides several blogs on this topic; e.g. Entity Linking & Relationship Extraction with Relik and Llamaindex.

@Quentin Reul this is amazing thank you!! Just what I was looking for. Curious why use Vosk over Whisper?

@Daniel Hunter there were several reasons: • Most examples that I found using

SpeechRecognition

used Google, but were not working • Wanted to demonstrate the process without relying on external APIs TBH, I would have used Ollama and a local model, but didn't think that Colab would support it.