LocalVoiceMode

# Voices and Skills Guide ## Quick Start ### Switching Voices 1. Place your voice file as `C:\AI\localvoicemode\voice_references\default.wav` 2. Restart the app - it will use your new voice ### Switching Skills/Personalities 1. Edit `C:\AI\localvoicemode\v2\ml-server\.env` 2. Add: `DEFAULT_SKILL=hermione-companion` 3. Restart the app --- ## Voice Requirements Pocket TTS clones voices from reference audio. For best results: | Requirement | Optimal | Acceptable | |-------------|---------|------------| | **Duration** | 10 seconds | 5-30 seconds | | **Format** | WAV (16-bit PCM) | WAV (any bit depth) | | **Sample Rate** | 24000 Hz | 16000-48000 Hz | | **Channels** | Mono | Stereo (will be converted) | | **Content** | Clear speech | Minimal background noise | ### Recording Tips 1. Record in a quiet environment 2. Speak naturally (avoid reading robotically) 3. Include varied intonation (questions, statements) 4. Keep consistent microphone distance 5. Use [ai-coustics](https://ai-coustics.com/) to enhance recordings (optional) ### Voice File Locations Priority order (first found is used): 1. `voice_references/default.wav` - Default voice override 2. `skills/<skill-id>/reference.wav` - Per-skill voice 3. `voice_references/<skill-id>.wav` - Global skill voice 4. Built-in voices: `alba`, `marius`, `javert`, `fantine` ### Swapping Voices **Method 1: Replace default.wav** ``` C:\AI\localvoicemode\voice_references\default.wav ``` **Method 2: Rename your file** ``` # Your voice file -> default.wav my_voice.wav -> default.wav ``` **Method 3: Per-skill voice** ``` C:\AI\localvoicemode\skills\hermione-companion\reference.wav ``` --- ## Skills/Personalities Skills define character personalities, system prompts, and optional voice files. ### Available Skills | Skill ID | Character | Description | |----------|-----------|-------------| | `assistant-default` | Default Assistant | General-purpose voice assistant | | `hermione-companion` | Hermione Granger | Roleplay companion at the Hog's Head pub | ### Skill Structure ``` skills/<skill-id>/ ├── SKILL.md # Character definition (required) ├── reference.wav # Custom voice (optional) ├── avatar.png # Character image (optional) └── references/ # Lore/knowledge files (optional) ``` ### Creating a New Skill 1. Create folder: `skills/my-character/` 2. Create `SKILL.md`: ```yaml --- id: my-character name: My Character display_name: "My Character" description: A custom character voice: reference.wav metadata: setting: "Where the character is" greeting: "Hello! How can I help you?" --- # My Character ## System Prompt You are My Character. [Full personality description here...] ``` 3. Add voice file (optional): `reference.wav` 4. Restart and select the skill in Settings ### SKILL.md Format The file uses YAML frontmatter + Markdown body: ```yaml --- id: skill-id # Unique identifier name: Character Name # Display name display_name: "Name" # Optional display override description: Brief desc # For skill listing voice: reference.wav # Voice file (optional) avatar: avatar.png # Avatar image (optional) metadata: setting: "Scene description" greeting: "First message" personality_traits: - Trait 1 - Trait 2 speech_patterns: - "Tends to say..." --- # Character Name ## System Prompt Full character instructions here... ``` --- ## Configuration ### Environment Variables (.env) Edit `C:\AI\localvoicemode\v2\ml-server\.env`: ```bash # LLM Provider OPENROUTER_API_KEY=sk-or-v1-... OPENROUTER_MODEL=deepseek/deepseek-chat-v3-0324 # Default skill to load DEFAULT_SKILL=assistant-default # Force LLM provider: lm_studio, openrouter, openai # VOICE_PROVIDER=openrouter ``` ### System Prompt Override To use a custom system prompt without creating a skill: 1. Edit `.env` 2. Add: `SYSTEM_PROMPT="You are a helpful assistant..."` --- ## Troubleshooting ### Voice not changing - Ensure file is named exactly `default.wav` - Check file format (must be WAV) - Restart the ML server ### Skill not loading - Check SKILL.md YAML syntax - Verify skill folder exists in `skills/` - Check ML server logs for errors ### Voice quality issues - Use higher quality source audio - Ensure 10+ seconds of speech - Remove background noise - Try different reference audio