voice-generation

Files (4)

• SKILL.md
• elevenlabs.py · scripts/
• gemini_tts.py · scripts/
• openai_tts.py · scripts/

SKILL.md (first lines)

# Voice Generation Skill

Generate realistic speech using AI (Google Gemini TTS, ElevenLabs, OpenAI TTS).

## Prerequisites

At least one API key is required:

- `GOOGLE_API_KEY` - For Google Gemini TTS (same key as video/image/music) ✅
- `ELEVENLABS_API_KEY` - For ElevenLabs high-quality voice synthesis
- `OPENAI_API_KEY` - For OpenAI TTS voices

## Available APIs

### Google Gemini TTS (Recommended - Same API Key)
- **Best for**: Podcasts, dialogues, audiobooks with style control
- **Voices**: 30 voices with natural language style control
- **Multi-speaker**: Up to 2 speakers for dialogues ✅
- **Languages**: 24 languages (auto-detected)
- **Features**: Control style, accent, pace via prompts
- **Output**: 24kHz WAV
- **API Key**: Same `GOOGLE_API_KEY` as video/image/music ✅

### ElevenLabs (Best Quality)
- **Best for**: Natural-sounding voices, voice cloning, long-form content
- **Voices**: 100+ pre-made voices + custom voice cloning
- **Languages**: 29+ languages
- **Models**: Eleven Multilingual v2, Eleven Turbo v2

### OpenAI TTS (Simplest)
- **Best for**: Quick, reliable text-to-speech with consistent quality
- **Voices**: alloy, echo, fable, onyx, nova, shimmer
- **Models**: tts-1 (fast), tts-1-hd (high quality)
- **Output**: MP3, Opus, AAC, FLAC

## Workflow

### Step 1: Understand the Request

Parse the user's voice request for:
- **Text content**: What should be spoken?
- **Voice type**: Male, female, specific character?
- **Tone**: Professional, casual, dramatic, cheerful?
- **Use case**: Narration, voiceover, audiobook, notification?
- **Language**: English, Spanish, other?
- **Speed**: Normal, slow, fast?

### Step 2: Select Voice and API

Choose based on requirements:

| Use Case | Recommended API | Reason |
|----------|----------------|--------|
| **Default / Same key as video** | Gemini TTS | Same `GOOGLE_API_KEY` ✅ |
| **Multi-speaker dialogue** | Gemini TTS | Up to 2 speakers built-in |
| **Style/accent control** | Gemini TTS | Natural language prompts |
| **Voice cloning** | ElevenLabs | Only API with cloning |
| **100+ voice options** | ElevenLabs | Widest selection |
| **Audiobook/podcast** | ElevenLabs or Gemini | Both excellent for long content |
| **Quick narration** | OpenAI TTS | Fast, reliable |
| **Budget-conscious** | OpenAI TTS | Lower cost |

### Step 3: Prepare the Text

Optimize text for speech:

1. **Add pauses**: Use commas, periods for natural rhythm
2. **Spell out numbers**: "1,234" → "one thousand two hundred thirty-four" (if needed)
3. **Handle acronyms**: "NASA" vs "N.A.S.A." depending on pronunciation
4. **Mark emphasis**: Some APIs support emphasis markers

**Example transformation:**
- Original: "The Q4 2024 results show a 15% YoY increase."
- Optimized: "The Q4 2024 results show a fifteen percent year-over-year increase."

### Step 4: Generate the Audio

Execute the appropriate script from `${CLAUDE_PLUGIN_ROOT}/skills/voice-generation/scripts/`:

**For Google Gemini TTS (single speaker):**
```bash
python3 ${CLAUDE_PLUGIN_ROOT}/skills/voice-generation/scripts/gemini_tts.py \
  --text "Welcome to our podcast!" \
  --voice "Charon"
```

**Gemini TTS with style direction:**
```bash
python3 ${CLAUDE_PLUGIN_ROOT}/skills/voice-generation/scripts/gemini_tts.py \
  --text "Have a wonderful day!" \
  --voice "Puck" \
  --style "Say cheerfully with a British accent:"
```

**Gemini TTS multi-speaker (dialogue):**
```bash
python3 ${CLAUDE_PLUGIN_ROOT}/skills/voice-generation/scripts/gemini_tts.py \
  --multi \
  --speaker "Host:Charon" \
  --speaker "Guest:Aoede" \
  --text "Host: Welcome to the show!
Guest: Thanks for having me!"
```

**For ElevenLabs:**
```bash
python3 ${CLAUDE_PLUGIN_ROOT}/skills/voice-generation/scripts/elevenlabs.py \
  --text "Your text here" \
  --voice "Rachel" \
  --model "eleven_multilingual_v2"
```

**For OpenAI TTS:**
```bash
python3 ${CLAUDE_PLUGIN_ROOT}/skills/voice-generation/scripts/openai_tts.py \
  --text "Your text here" \
  --voice "nova" \
  --model "tts-1-hd"
```

From michaelboeding/skills, MIT.

More from michaelboeding/skills