jknapp
6d4ae2e5c3
- OAuth authentication via Authentik - WebSocket connection to OpenClaw gateway - Configurable gateway URL with first-run setup - User preferences sync across devices - Multi-user support with custom assistant names - ElevenLabs TTS integration (local + remote) - FCM push notifications for alarms - Voice input via Google Speech API - No hardcoded secrets or internal IPs in tracked files
105 lines
2.6 KiB
Markdown
105 lines
2.6 KiB
Markdown
# Wake Word Implementation Plan
|
|
|
|
## Overview
|
|
Add "Hey Alfred" or "Alfred" wake word detection to activate voice input hands-free.
|
|
|
|
## Technology Choice: Porcupine by Picovoice
|
|
|
|
**Why Porcupine:**
|
|
- ✅ On-device processing (no internet required)
|
|
- ✅ Low battery usage
|
|
- ✅ Free tier available (1 wake word, non-commercial)
|
|
- ✅ Android SDK available
|
|
- ✅ Custom wake words supported
|
|
|
|
**Free Tier Limits:**
|
|
- 1 user account
|
|
- 1 custom wake word model
|
|
- Non-commercial use only
|
|
- On-device processing (unlimited uses)
|
|
|
|
## Implementation Steps
|
|
|
|
### 1. Sign up for Porcupine
|
|
- Go to https://console.picovoice.ai/
|
|
- Create account
|
|
- Get Access Key (free tier)
|
|
- Train custom "Alfred" wake word
|
|
|
|
### 2. Add Porcupine SDK
|
|
|
|
Add to `app/build.gradle.kts`:
|
|
```kotlin
|
|
dependencies {
|
|
implementation("ai.picovoice:porcupine-android:3.0.2")
|
|
}
|
|
```
|
|
|
|
### 3. Add to secrets.properties
|
|
```
|
|
PORCUPINE_ACCESS_KEY=<your-access-key-here>
|
|
```
|
|
|
|
### 4. Create WakeWordManager.kt
|
|
|
|
Similar to VoiceInputManager, this would:
|
|
- Initialize Porcupine with the wake word model
|
|
- Listen continuously in the background (low power mode)
|
|
- Trigger voice input when wake word detected
|
|
- Show visual feedback (e.g., pulse animation)
|
|
|
|
### 5. Add Background Service
|
|
|
|
Wake word detection needs to run in the background:
|
|
- Foreground service with notification
|
|
- Shows "Listening for 'Hey Alfred'..." notification
|
|
- Can be toggled on/off from app
|
|
- Respects battery optimization settings
|
|
|
|
### 6. UI Updates
|
|
|
|
Add toggle in settings or status bar:
|
|
- "Wake Word Detection" switch
|
|
- Shows when listening
|
|
- Visual feedback when triggered (e.g., microphone icon pulses)
|
|
|
|
## User Experience Flow
|
|
|
|
1. User enables "Wake Word Detection" in app
|
|
2. App starts background service
|
|
3. Notification shows: "Listening for 'Hey Alfred'"
|
|
4. User says **"Hey Alfred"** or **"Alfred"**
|
|
5. App shows visual feedback (screen lights up, icon pulses)
|
|
6. Voice input starts automatically
|
|
7. User speaks their message
|
|
8. Message auto-sends when done
|
|
9. Alfred responds with voice
|
|
|
|
## Considerations
|
|
|
|
**Battery Impact:**
|
|
- Porcupine is optimized for low power
|
|
- Typically <1% battery drain per hour
|
|
- Can be toggled off when not needed
|
|
|
|
**Privacy:**
|
|
- All processing on-device
|
|
- No audio sent to cloud
|
|
- Only activates when wake word detected
|
|
|
|
**Implementation Complexity:**
|
|
- Medium complexity (1-2 hours work)
|
|
- Requires Porcupine setup
|
|
- Need background service implementation
|
|
- Permission handling (microphone always-on)
|
|
|
|
## Next Steps
|
|
|
|
If you want this feature:
|
|
1. Create Picovoice account
|
|
2. Get Access Key
|
|
3. Train "Alfred" wake word model
|
|
4. I can implement the code integration
|
|
|
|
Let me know if you want to proceed with this! 🎤
|