We’ve learned Transcription (Whisper) and Summarization separately. Now, let’s combine them to solve a real problem: watching long YouTube videos. In this post, we’ll build a Python YouTube Summarizer to make this process much easier.
We will build a script that takes a YouTube URL and outputs a short text summary.
Step 1: Installation
We need yt-dlp to download audio, plus our AI libraries. Note: You need ffmpeg installed on your system.
pip install yt-dlp openai-whisper transformers torch
Step 2: The Code
This script has three stages: Download -> Transcribe -> Summarize.
import yt_dlp
import whisper
from transformers import pipeline
import os
# --- CONFIGURATION ---
VIDEO_URL = "https://www.youtube.com/watch?v=YOUR_VIDEO_ID" # Pick a short video (< 5 mins)
AUDIO_FILE = "audio.mp3"
# --- 1. DOWNLOAD AUDIO ---
print("Downloading audio...")
ydl_opts = {
'format': 'bestaudio/best',
'postprocessors': [{
'key': 'FFmpegExtractAudio',
'preferredcodec': 'mp3',
'preferredquality': '192',
}],
'outtmpl': 'audio', # Saves as audio.mp3
}
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
ydl.download([VIDEO_URL])
# --- 2. TRANSCRIBE (Speech-to-Text) ---
print("Transcribing audio (loading Whisper)...")
model = whisper.load_model("base")
result = model.transcribe(AUDIO_FILE)
full_text = result["text"]
print(f"Transcript length: {len(full_text)} characters")
# --- 3. SUMMARIZE (Text-to-Text) ---
print("Summarizing text...")
summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
# We might need to chunk the text if it's too long, but for short videos:
# (max_length is the length of the summary)
summary = summarizer(full_text[:3000], max_length=150, min_length=50, do_sample=False)
print("\n--- VIDEO SUMMARY ---")
print(summary[0]['summary_text'])
# Cleanup
os.remove(AUDIO_FILE)Step 3: The Result
You can now paste in a URL for a tech talk or news clip, wait a minute, and get a perfect paragraph explaining what happened!
Key Takeaways
- The article explains how to build a Python YouTube Summarizer to handle long video content.
- First, install necessary tools like yt-dlp and ensure ffmpeg is on your system.
- The script involves three stages: Download, Transcribe, and Summarize.
- Users can input any YouTube URL to receive a concise summary of the content.
- This solution is ideal for summarising tech talks and news clips efficiently.
