# TTS Personality Engine (Issue #494)

Context-aware text-to-speech with emotion-based rate/pitch modulation for SaltyBot.

## Features

### Context-Aware Responses
- **Time-based greetings**: Different greetings for morning, afternoon, evening, night
- **Person recognition**: Uses detected person names in responses
- **Emotion-driven speech**: Adjusts rate and pitch based on current emotion state
- **Personality identity**: Configurable personality name (default: "Luna")

### Priority Queue Management
- **Safety (Priority 3)**: Emergency/safety messages (immediate)
- **Social (Priority 2)**: Greetings and interactions (high priority)
- **Idle (Priority 1)**: Commentary and chatter (medium priority)
- **Normal (Priority 0)**: Default messages

### Emotion-Based Rate/Pitch Modulation
- **Happy**: Faster (1.1x) with higher pitch (1.15x)
- **Sad**: Slower (0.9x) with lower pitch (0.85x)
- **Neutral**: Normal speech (1.0x)

## Architecture

### Node: `tts_personality_node`

**Subscribes**
- `/saltybot/tts_request` - Text to synthesize with personality
- `/saltybot/emotion_state` - Current emotion ("happy", "sad", "angry", "neutral")
- `/saltybot/person_detected` - Detected known person name

**Publishes**
- `/saltybot/tts_command` - Formatted TTS command with personality parameters (JSON)
- `/saltybot/personality_state` - Current personality state

### Integration Flow

```
tts_request → personality_node → apply context/emotion → tts_service → audio device
                                                                        (Jabra SPEAK 810)
```

## Configuration

### Parameters (tts_personality_params.yaml)

```yaml
personality_name: "Luna"           # Robot's personality name
enable_context_aware: true         # Use time/person/emotion context

# Emotion modulation
speed_happy: 1.1                   # Rate multiplier when happy
speed_sad: 0.9                     # Rate multiplier when sad
pitch_happy: 1.15                  # Pitch multiplier when happy
pitch_sad: 0.85                    # Pitch multiplier when sad

# Time-based greetings
greetings_morning: [...]           # 5am-12pm
greetings_afternoon: [...]         # 12pm-5pm
greetings_evening: [...]           # 5pm-9pm
greetings_night: [...]             # 9pm-5am

# Known people for personalization
known_people:
  person_001: "Seb"
  person_002: "Alex"
```

## Usage

### Launch
```bash
ros2 launch saltybot_tts_personality tts_personality.launch.py personality_name:=Luna
```

### Send TTS Request
```bash
ros2 topic pub /saltybot/tts_request std_msgs/String "data: 'hello world'"
```

### Update Emotion
```bash
ros2 topic pub /saltybot/emotion_state std_msgs/String "data: 'happy'"
```

### Detect Person
```bash
ros2 topic pub /saltybot/person_detected std_msgs/String "data: 'Seb'"
```

### Monitor Personality State
```bash
ros2 topic echo /saltybot/personality_state
```

## Example Personalities

### Luna (Default)
- Morning: "Good morning! I hope you slept well."
- Happy speech: Faster, higher pitch
- Emotion context: Uses detected mood

### Alternative: Customize in Config

```yaml
personality_name: "Atlas"
greetings_morning:
  - "Systems online. Good morning, Commander."
  - "Wake-up sequence initiated."
speed_happy: 1.2        # Atlas speaks faster when happy
```

## Integration with Other Components

### Emotion Engine (Issue #429)
- Listens to `/saltybot/emotion_state`
- Adjusts rate/pitch based on emotion
- Integrates with personality for consistent expression

### TTS Service (Issue #421)
- Receives formatted TTS commands via `/saltybot/tts_command`
- Handles actual speech synthesis with Piper-TTS
- Outputs to Jabra SPEAK 810 audio device

### Person Tracking
- Subscribes to detected person names
- Personalizes responses with known person names
- Maintains recent person context (30-second window)

## Testing

```bash
# Run unit tests
colcon test --packages-select saltybot_tts_personality

# Manual testing
ros2 launch saltybot_tts_personality tts_personality.launch.py
ros2 topic pub /saltybot/tts_request std_msgs/String "data: 'test message'"
```

## Safety Considerations

- Safety messages cannot be overridden by emotions (always use neutral rate/pitch)
- Queue has maximum size of 16 items to prevent memory exhaustion
- Emergency/safety keywords trigger highest priority queue level
- Personality does not interfere with critical safety protocols

## Future Enhancements

- [ ] Machine learning-based emotion detection from voice analysis
- [ ] Multi-language support
- [ ] Voice variants (male, female, child)
- [ ] Custom personality training from recordings
- [ ] Integration with dialogue system for context-aware responses