Talk to AI persona clones from your phone — voice in, voice out.

How It Works · Quick Start · Clones · Architecture · Troubleshooting

What is Clone Talking?

Clone Talking is a voice AI web app that brings personas to life. Run a server on your PC, scan a QR code with your phone, hold a button to speak, and the AI responds in a cloned voice — in real time.

Whether you want to chat with Elon Musk, a custom mentor, or any fully custom persona you create — it's just a few taps away. No app installation. No cloud hosting. Just your PC and your phone.

How It Works

┌─────────────┐
│ Your Phone  │  Scan QR code → browser opens
└──────┬──────┘
       │  Hold button → speak → release to send
       ▼
┌──────────────────┐
│  Express Server  │
│   + WebSocket    │
└────┬─┬─┬────────┘
     │ │ │
     ▼ ▼ ▼
   ┌───────────────────────┐
   │ Whisper (OpenAI)      │  Audio → Text
   │ OpenRouter LLM        │  Text  → Response
   │ VoiSpark TTS          │  Text  → Cloned Voice
   └──────────┬────────────┘
              ▼
         ┌─────────────┐
         │ Your Phone  │  Plays audio response
         │ (Speaker)   │  Saves conversation log
         └─────────────┘

Features

Prerequisites

• Node.js 18+
• Android phone with Chrome — or any modern browser supporting Web Audio API
• 4 API keys (all have free tiers):

Quick Start

1. Clone and install

git clone https://github.com/MatheusSimonaci/clone-talking.git
cd clone-talking
npm install

2. Configure environment

cp .env.example .env

Edit .env with your keys:

OPENROUTER_API_KEY=your-key-here
OPENROUTER_MODEL=anthropic/claude-sonnet-4-5
OPENAI_API_KEY=your-key-here
VOISPARK_API_KEY=your-key-here
VOISPARK_PROVIDER=fish-audio
VOISPARK_MODEL_ID=s1
NGROK_AUTHTOKEN=your-token-here

3. Run the interactive setup

npm start

The wizard will:

• List all available clones
• Ask for a VoiSpark voice ID (saved automatically for future runs)
• Start the server on localhost:3000
• Open an ngrok HTTPS tunnel
• Display a QR code in your terminal

4. Scan and chat

1. Scan the QR code with your phone
2. Tap "Allow" when prompted for microphone access
3. Hold the button and speak
4. Release to send
5. Listen as the AI responds in its cloned voice

Environment Variables

Why ngrok? Android Chrome requires HTTPS for microphone access. ngrok creates a free encrypted tunnel to your local server — no certificate setup needed.

Clone System

Clones are the heart of Clone Talking. Each clone is a persona — a character with its own voice, personality, and expertise.

How clones work

Clones live in the clones/ directory as Markdown files. Each file has:

• YAML frontmatter — metadata (id, name, description, voiceId)
• Markdown body — the system prompt (instructions for the AI)

When you chat with a clone, the file body becomes the LLM's system prompt.

Clone file format

---
id: elon-musk
name: Elon Musk
description: Strategic mentor — First Principles, The Algorithm, Maniacal Urgency
voiceId: b43a971a-f770-41fe-8afd-d245423ee75a
---

You are Elon Musk — entrepreneur, engineer, and strategic mentor.
Apply first principles thinking. Question every requirement. Delete before building.

Adding a custom clone

cat > clones/my-mentor.md << 'EOF'
---
id: my-mentor
name: My Mentor
description: A custom strategic advisor persona.
voiceId: ""
---

You are a sharp, direct strategic advisor.
Give concise, actionable answers. Ask clarifying questions before diving in.
EOF

Then run it:

npm start -- my-mentor

On first run, the wizard will ask for a VoiSpark voice ID. It saves automatically.

Voice IDs (VoiSpark)

Each clone needs a VoiSpark Voice ID — identifies a celebrity voice or custom clone.

How to get a Voice ID:

1. Go to voispark.com and log in
2. Browse Celebrity Voices or the voice catalog
3. Select a voice and copy its Project ID
4. Paste it as the voiceId in your clone's frontmatter or enter it during setup

How voices are stored:

Voice IDs are saved automatically in .voice-config.json (gitignored):

{
  "elon-musk": { "voiceId": "b43a971a-f770-41fe-8afd-d245423ee75a" }
}

You can also edit .voice-config.json directly, or set a voiceId in the clone's frontmatter as a default.

Usage

# Interactive mode — pick from menu
npm start

# Direct mode — skip wizard
npm start -- elon-musk

# Raw Node.js
node src/server.js elon-musk

Model Selection

Any model on OpenRouter works. Set OPENROUTER_MODEL in .env.

Architecture

Client (Browser)
       │
       │ WebSocket
       │
    Express Server (server.js)
       │
       ├─→ Whisper API (OpenAI)        ← Audio → Text
       │
       ├─→ OpenRouter (any LLM)        ← Text → Response
       │
       ├─→ VoiSpark TTS API            ← Text → Cloned Voice (streamed)
       │
       └─→ ngrok Tunnel               ← HTTPS for mobile mic access

Pipeline flow:

1. Client records audio and sends it via WebSocket
2. Whisper transcribes audio to text
3. OpenRouter LLM generates a response using the clone's system prompt
4. VoiSpark streams synthesized audio with the clone's voice
5. Server streams audio chunks back to the client in real time
6. Logger saves the full exchange to a session JSON file

Session Logs

Every conversation is automatically saved to logs/:

logs/session-2026-03-28-15-30-00.json

{
  "sessionId": "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
  "cloneFile": "clones/elon-musk.md",
  "voiceId": "b43a971a-f770-41fe-8afd-d245423ee75a",
  "model": "anthropic/claude-sonnet-4-5",
  "startedAt": "2026-03-28T15:30:00.000Z",
  "messages": [
    {
      "role": "user",
      "content": "What is the theory of relativity?",
      "timestamp": "2026-03-28T15:30:05.123Z"
    },
    {
      "role": "assistant",
      "content": "Ah, a wonderful question! Allow me to explain E=mc²...",
      "timestamp": "2026-03-28T15:30:08.456Z"
    }
  ]
}

The logs/ directory is gitignored.

Troubleshooting

Ethical Use

Clone Talking is a tool for creative, educational, and entertainment purposes. Voice cloning is powerful technology — use it responsibly.

The following uses are strictly prohibited:

• ❌ Fraud & impersonation — cloning someone's voice without consent to deceive others
• ❌ Scams — using AI voices to impersonate real people in financial or social engineering attacks
• ❌ Non-consensual content — generating voice content with a real person's likeness without their explicit permission
• ❌ Harassment — using cloned voices to harass, defame, or harm individuals
• ❌ Disinformation — creating fake audio attributed to real people to spread false information

By using this project, you agree to comply with all applicable laws and the terms of service of each API provider (VoiSpark, OpenRouter, OpenAI, ngrok). The author is not responsible for any misuse of this software.

If you witness abuse, report it to the relevant platform or authority.

Contributing

Clone Talking is open source. Contributions welcome!

• Found a bug? Open an issue
• Have an idea? Submit a pull request
• Want to share a clone? Add it to clones/ and open a PR

License

MIT License. See LICENSE for details.

_{Built with Node.js · Express · WebSocket · OpenAI Whisper · OpenRouter · VoiSpark
Talk to anyone. From anywhere. Voice first.}