AI 原生开放教材

外观

EdgeAI for Beginners系统课程·英文原文 ↗原件 ↓仓库 ↗

Act 3: Bring Your Podcast to Life 🎤 ​

lab3

The Grand Finale ​

You've researched topics. You've written scripts. Now for the cherry on top: turning your text into actual podcast audio with realistic voices!

Enter VibeVoice — Microsoft Research's open-source TTS (text-to-speech) magic that creates:

  • 🎭 Natural-sounding conversations
  • 👥 Multiple speakers (up to 4!)
  • ⏱️ Long-form audio (up to 90 minutes!)
  • 🎵 Expressive delivery (not robot voices!)

This is the tech behind synthetic podcasts. Let's make yours!

What's VibeVoice? (The Cool Stuff) ​

VibeVoice is Microsoft Research's gift to the world. It's specifically designed for podcast-style conversational audio.

Why It's Awesome 🔥 ​

  • ⏱️ Marathon Sessions: Generates up to 90 minutes of continuous speech (that's a full podcast episode!)
  • 👥 Multi-Speaker Magic: Up to 4 different voices with consistent personalities
  • ⚡ Crazy Efficient: Uses ultra-low 7.5 Hz frame rate to save compute power
  • 🧠 Smart Audio: Combines an LLM (understands context) with a diffusion model (creates realistic audio)
  • 🎭 Natural Flow: Handles turn-taking, pauses, and conversational rhythm automatically

Translation: VibeVoice doesn't just read your script — it performs it like real people talking.

Before You Start 🚀 ​

What You Need:

  • 🐍 Python 3.10+ (you already have this from Acts 1 & 2)
  • 🚀 uv (a fast Python package manager — we'll install it)
  • 📝 Your Script: The podcast.txt file from Act 2 (in ../03.Application/)

Pro Tip: This step requires a good internet connection for downloading the pre-trained model. Grab a coffee! ☕

Let's Go! The Easy Way 🎬 ​

We've made this stupid simple. One shell script does everything.

The Process ​

  1. Make it executable:
bash
chmod +x run_vibe_voice.sh
  1. Run it:
bash
./run_vibe_voice.sh
  1. Wait for magic (this might take a few minutes on first run)

What Happens Behind the Scenes 🎭 ​

The script is basically your automated sound engineer:

  1. 📥 Downloads VibeVoice: Clones the official repo from GitHub
  2. 📦 Installs Dependencies: Uses uv pip for lightning-fast package installation
  3. 🎬 Generates Audio: Runs the inference script with:
    • --model_path: The pre-trained VibeVoice-7B model
    • --txt_path: Your podcast.txt script
    • --speaker_names: Assigns voices (Xinran & Anchen by default)

Result: Your script becomes a real podcast episode! 🎉

Your Mission 🎯 ​

Let's make this interesting:

Task 1: Create Content ​

Edit ../03.Application/podcast.txt with a conversation between two people. Make it about tech, hobbies, whatever! Just make it conversational.

Format Example:

Speaker 1: Hey! Did you hear about the new AI model?
Speaker 2: No way! Tell me more!
Speaker 1: It's called...

Task 2: Generate Audio ​

Run the script and watch the magic happen. First time will take longer (downloading the model).

Task 3: Listen & Analyze ​

  • Does it sound natural?
  • Do the speakers have distinct voices?
  • Is the turn-taking smooth?
  • Any robotic moments?

Task 4: Experiment (For the Brave) ​

Edit run_vibe_voice.sh and change --speaker_names to try different voice combinations. VibeVoice has multiple pre-trained voices!

Bonus Challenge: Try a 3-speaker conversation! 🎆

Learn More 📚 ​

⚠️ Responsible AI Reminder: VibeVoice is powerful. Use it ethically! Don't create deepfakes or misleading content. Create cool stuff that helps people. 🙏

🏆 Congratulations! You Did It! ​

You just completed the full pipeline:

  1. ✅ Act 1: Built AI agents with custom tools
  2. ✅ Act 2: Orchestrated a multi-agent workflow
  3. ✅ Act 3: Generated real podcast audio

You now have:

  • A working AI research assistant
  • A full podcast production workflow
  • An actual audio file you can share

What's Next? 🚀 ​

Launch your podcast!

  • Upload to podcast platforms
  • Share on social media
  • Iterate and improve

Keep building!

  • Try different topics
  • Experiment with more speakers
  • Add background music
  • Build a web interface
  • Automate the whole thing

Share your work! Tag us! Show the world what you built. The AI podcast revolution starts with you. 🎙️

Questions? Ideas? Success stories? Drop them in the workshop chat!

Welcome to the future of content creation. 🌟