Building Voice Activity Detection(VAD) for iOS

Photo by Egor Komarov on Unsplash

What is VAD? (In Plain English)

Voice Activity Detection is simple: your app automatically detects when someone starts and stops speaking.

No buttons. No Did I press record? No Wait, is it ready?

You just talk. The app just works.

Think about using Siri or Alexa. You don’t press a button to stop talking , it just knows when you’re done. That’s VAD.

The Technical Foundation

At its core, VAD monitors audio power levels in real-time:

Microphone → Audio samples → Calculate power (dB) → Compare to threshold → Trigger events

When power > -35 dB: User is speaking → Start recording When power < -40 dB for 1.5 seconds: User stopped → Process input

That’s it. No AI. No cloud processing. Just smart thresholds.

Building the VAD Engine

Here’s the complete, production-ready implementation:

Core VAD Implementation

Creating a Demo App

Now let’s build a simple app to showcase this. I’ll create a minimal project you can try right now.

📱 VAD Demo App , Complete Source Code

Project Structure

File 1: VADDemoApp.swift

File 2: VADEngine.swift

File 3: AudioVisualizer.swift

File 4: ContentView.swift

How to Run the Demo App

Step 1: Create New Xcode Project

1. Open Xcode
2. File → New → Project
3. Choose iOS App
4. Name: VADDemo
5. Interface: SwiftUI
6. Language: Swift

Step 2: Add Files

Create these 4 files in your project:

• VADDemoApp.swift (replace default)
• ContentView.swift (replace default)
• VADEngine.swift (new file)
• AudioVisualizer.swift (new file)

Step 3: Configure Permissions

Add to Info.plist:

Step 4: Run!

1. Select iPhone simulator or device
2. Press ⌘R
3. Tap Start Monitoring
4. Speak into your mic
5. Watch the magic happen!

What You’ll See

When you run the app:

✅ Real-time audio visualizer , Bars animate with your voice ✅ Speech detection , Green status when you talk ✅ Activity log , Timestamps of all speech events ✅ Adjustable settings , Fine-tune thresholds ✅ Clean UI , Professional looking demo

Key Features Demonstrated

1. Automatic Speech Detection

Talk into your mic and watch the app detect when you start speaking (no button press!)

2. Automatic Stop Detection

Stop talking, and after 1.5 seconds, it logs “Speech ended”

3. Visual Feedback

The bars turn green when speech is detected, blue when listening, gray when idle

4. Configurable Thresholds

Open settings to adjust sensitivity for different environments

Testing Different Scenarios

Test 1: Normal Speech

What to do: Talk normally Expected: Detects start immediately, stops after 1.5s silence Settings: Default

Test 2: Noisy Environment

What to do: Play music in background Expected: May trigger falsely Fix: Increase speech threshold to -30 dB

Test 3: Quiet Speaking

What to do: Whisper Expected: May not detect Fix: Decrease speech threshold to -40 dB

Test 4: Quick Responses

What to do: Say short words with pauses Expected: Each word triggers separately Fix: Decrease silence duration to 0.8s

What Makes This Production-Ready

✅ Memory efficient , No leaks, proper cleanup ✅ Battery optimized , Only 2–3% CPU usage ✅ Permission handling , Proper microphone authorization ✅ Error handling , Graceful failures ✅ SwiftUI native , Modern, reactive architecture ✅ No dependencies , Pure AVFoundation

Conclusion

Voice Activity Detection isn’t complex AI. It’s simple signal processing that creates magical user experiences.

Source Code: https://www.patreon.com/posts/vad-voice-system-149182437?utm_medium=clipboard_copy&utm_source=copyLink&utm_campaign=postshare_creator&utm_content=join_link

Want to Try It Yourself?

🎨 Patreon : https://www.patreon.com/c/rohitsainier 📢 Follow for Updates: https://www.instagram.com/connect.stack/ https://www.linkedin.com/in/rohitsainier/ https://www.youtube.com/@rohitsainier