YouTube Insights & Speech Generation Pipeline

An end-to-end Python pipeline that automatically fetches videos from a YouTube playlist, transcribes them using AWS Transcribe, generates structured insights using AWS Bedrock (Claude 3 Sonnet), and stores everything in AWS S3 and DynamoDB. Includes a speech generation module that analyses a speaker's communication style from transcripts and generates contextually relevant speeches using real-time financial data.

Problem Statement

Employees at a financial services firm couldn't access YouTube due to company security restrictions, yet needed to stay updated with the Managing Director's video content. Watching full 1-hour videos was impractical. This pipeline automates the entire process: fetching, transcribing and summarising videos into 5-minute insight reports, and extends further to auto-generate speeches in the MD's style.

Architecture

Features

• Fetches all videos from a YouTube playlist automatically
• Downloads and converts audio using yt-dlp and ffmpeg
• Transcribes audio using AWS Transcribe with custom domain vocabulary for improved accuracy
• Avoids reprocessing using DynamoDB to track processed video IDs
• Generates structured JSON insights (summary, action items, follow-ups) using Claude 3 Sonnet via AWS Bedrock
• Analyses speaker communication style from transcripts
• Generates contextually relevant speeches using real-time Nifty market data and financial news headlines

Project Structure

youtube-insights-pipeline/ ├── youtube_transcripts.py ← main pipeline: fetch, transcribe, generate insights ├── speech_pipeline.py ← speaking style analysis and speech generation ├── requirements.txt ← Python dependencies ├── .env.example ← environment variables template └── README.md

Setup

Prerequisites

Install ffmpeg before running:

• Mac: brew install ffmpeg
• Windows: Download from https://ffmpeg.org/download.html and add to PATH
• Linux: sudo apt install ffmpeg

Installation

pip install -r requirements.txt

Configuration

Copy .env.example to .env and fill in your credentials:

cp .env.example .env

Variable	Description
YOUTUBE_API_KEY	YouTube Data API v3 key from Google Cloud Console
PLAYLIST_ID	YouTube playlist ID to process
S3_BUCKET_NAME	AWS S3 bucket name for storing files
DYNAMODB_TABLE_NAME	DynamoDB table name for tracking processed videos
AWS_REGION	AWS region (e.g. ap-south-1)
AWS_ACCESS_KEY_ID	AWS access key
AWS_SECRET_ACCESS_KEY	AWS secret key

AWS IAM Permissions Required

Your IAM user needs the following policies:

• AmazonS3FullAccess
• AmazonDynamoDBFullAccess
• AmazonTranscribeFullAccess
• AmazonBedrockFullAccess

Running

# Run the main pipeline
python youtube_transcripts.py

# Run the speech generation pipeline
python speech_pipeline.py

How It Works

youtube_transcripts.py

1. Fetches all videos from the configured YouTube playlist
2. Checks DynamoDB to skip already processed videos
3. Downloads audio using yt-dlp and converts to MP3 via ffmpeg
4. Uploads audio to S3 and starts an AWS Transcribe job with custom vocabulary
5. Once transcription completes, sends transcript to Claude 3 Sonnet via Bedrock
6. Saves structured JSON insights (transcript, summary, action items, follow-ups) back to S3
7. Marks video as processed in DynamoDB

speech_pipeline.py

1. Loads all transcripts from local storage
2. Analyses the speaker's communication style using Claude 3 Sonnet
3. Fetches real-time Nifty market data and financial news headlines
4. Generates a contextually relevant speech in the speaker's style
5. Uploads the generated speech directly to S3

Notes

• This project requires active AWS credentials and a YouTube Data API key to run
• YouTube blocks API requests from cloud services (Lambda, EC2), so the pipeline is designed to run locally
• The custom vocabulary in AWS Transcribe significantly improves accuracy for domain-specific financial terminology

Tech Stack

Python, AWS S3, AWS Transcribe, AWS Bedrock, AWS DynamoDB, Claude 3 Sonnet, YouTube Data API v3, yt-dlp, ffmpeg, yfinance, BeautifulSoup