Wake Word Detection | Model Comparison & Demo

🎙️ Web Speech API Native Browser

Built-in browser speech recognition. Best for simple wake words like "Hey Computer" or "OK Google".

⚡ Speed: Fast 📦 Size: 0KB 🌐 Offline: ❌

✅ Zero setup, works instantly

✅ No external dependencies

❌ Requires internet (Google servers)

❌ Limited customization

🤖 Porcupine (Picovoice) Edge AI

Lightweight, privacy-focused wake word engine. Supports custom keywords and runs 100% offline.

⚡ Speed: Very Fast 📦 Size: ~500KB 🌐 Offline: ✅

✅ Privacy-first, no cloud

✅ Highly accurate

❌ Needs API key for custom words

❌ Slightly complex setup

🧠 TensorFlow.js + Speech Commands ML Model

Pre-trained model recognizing 20+ keywords like "Yes", "No", "Stop", "Go". Runs entirely in-browser.

⚡ Speed: Medium 📦 Size: ~4MB 🌐 Offline: ✅

✅ Open source, customizable

✅ No internet required

❌ Larger model size

❌ Requires GPU for best perf

🚀 Silero VAD + Custom Voice Activity

Voice activity detection combined with simple keyword matching. Great for custom triggers.

⚡ Speed: Fast 📦 Size: ~2MB 🌐 Offline: ✅

✅ Very flexible

✅ Low latency

❌ Manual keyword extraction needed

❌ Less accurate than dedicated models

🎯 Wake Word Detection Models