Wake Word Detection

Wake word detection (like "Hey Siri" or "OK Google") allows your app to listen for a specific phrase and trigger an action entirely hands-free. VibeFast uses this to launch the AI Assistant, start a voice note, or open any feature just by speaking.

It relies on Vosk for offline recognition and exposes an auto-starting listener you can tie into any feature.

vf add wake-word

Core pieces

• Hook: apps/native/src/features/wake-word/hooks/use-wake-word.ts manages the lifecycle (start/stop/restart) and fires events via wakeWordBus.
• Services: wake-word/services/wake-word-manager.ts wraps react-native-vosk, and services/audio-session.ts prepares the microphone and audio mode.
• UI: Build your own trigger buttons or use the wake-word/components helpers for status chips.

Assets

Wake word detection requires a local language model. When you run vf add wake-word, the CLI automatically downloads this model for you.

If for any reason the assets are missing from apps/native/assets/vosk-model/model-en-us, you can run this script manually from the native folder:

node scripts/download-vosk-model.mjs

The default keywords live in apps/native/src/features/wake-word/constants.ts (KEYWORD_NAMES = ['vibefast']), and the DEFAULT_SENSITIVITY / cooldown values keep false positives in check.

Customization

• Provide your own WakeWordConfig (keyword, sensitivity, cooldown) whenever you call useWakeWord.
• Listen to wakeWordBus.on('trigger', ...) to react globally (e.g., open the chatbot modal when the word fires).
• Because Vosk is entirely local, no extra API key is required—just ensure the model file exists and the microphone permission is granted.