VoiceBox: The Open-Source Voice Synthesis Studio That Keeps Your Data Local

Most voice synthesis tools work the same way. You type text, upload a voice sample, and an API somewhere in the cloud generates audio. It works. But your voice data leaves your machine, your costs scale with every character, and you have limited control over what happens to the models trained on your input.

VoiceBox takes a different approach. It runs entirely on your hardware. No API calls, no per-character billing, no data leaving your machine.

Built by Jamie Pine, VoiceBox is a local-first AI voice studio that handles both sides of the voice loop: text-to-speech output and speech-to-text input. It has accumulated over 24,500 stars on GitHub and is currently on version 0.5.0, released April 2026. The project is open source under the MIT license.

What VoiceBox Does

• Qwen3-TTS and Qwen CustomVoice for natural-language delivery control
• LuxTTS for lightweight generation (~1GB VRAM, 150x real-time on CPU)
• Chatterbox Multilingual and Chatterbox Turbo for expressive speech with paralinguistic tags like [laugh], [sigh], and [gasp]
• HumeAI TADA for emotional speech synthesis
• Kokoro for 50+ curated preset voices

Beyond basic TTS, VoiceBox includes post-processing effects powered by Spotify's pedalboard library: pitch shift, reverb, delay, chorus, compression, and filters. There are four built-in presets (Robotic, Radio, Echo Chamber, Deep Voice) and you can create custom ones.

The Privacy Argument

VoiceBox sidesteps this entirely. Models download once and run locally. Voice profiles, generated audio, and capture recordings stay in your data directory. Nothing phones home.

For creators building content at scale, or anyone cloning a voice they care about protecting, this distinction matters.

How to Install VoiceBox

macOS (Apple Silicon)

• Download the DMG from voicebox.sh/download/mac-arm
• Drag VoiceBox to your Applications folder
• On first launch, grant the required Accessibility and Input Monitoring permissions when prompted

macOS (Intel)

• Download the DMG from voicebox.sh/download/mac-intel
• Same installation and permissions process

Windows

• Download the MSI installer from voicebox.sh/download/windows
• Run the installer and follow the prompts

Docker

docker compose up

Building From Source

git clone https://github.com/jamiepine/voicebox.git
cd voicebox
just setup    # creates Python venv, installs all deps
just dev      # starts backend + desktop app

Prerequisites: Bun, Rust, Python 3.11+, Tauri prerequisites, and Xcode on macOS. Install just via brew install just or cargo install just.

How to Clone Your First Voice

• Open VoiceBox and navigate to the Profiles section
• Click Create Profile
• Either upload an audio file of the voice you want to clone, or record directly in the app
• Give the profile a name and optional description
• Select a TTS engine (Qwen3-TTS is a good starting point for general use; LuxTTS if you want fast, lightweight generation)
• Type your text in the generation box and click Generate
• Preview the output, apply effects if needed, and export

Working With the REST API

Generate Speech

curl -X POST http://127.0.0.1:17493/generate \
  -H "Content-Type: application/json" \
  -d '{ "text": "Hello world", "profile_id": "abc123", "language": "en" }'

Agent Voice Output

curl -X POST http://127.0.0.1:17493/speak \
  -H "Content-Type: application/json" \
  -H "X-Voicebox-Client-Id: my-script" \
  -d '{ "text": "Deploy complete.", "profile": "Morgan" }'

Transcribe Audio

curl -X POST http://127.0.0.1:17493/transcribe \
  -F "audio=@recording.wav" \
  -F "model=whisper-turbo"

List Voice Profiles

curl http://127.0.0.1:17493/profiles

Full API documentation is available at http://127.0.0.1:17493/docs when the app is running.

MCP Server for AI Agents

Claude Code Setup (one line)

claude mcp add voicebox \
  --transport http \
  --url http://127.0.0.1:17493/mcp \
  --header "X-Voicebox-Client-Id: claude-code"

Cursor / Windsurf / VS Code

{
  "mcpServers": {
    "voicebox": {
      "url": "http://127.0.0.1:17493/mcp",
      "headers": {
        "X-Voicebox-Client-Id": "cursor"
      }
    }
  }
}

Four MCP tools are available: voicebox.speak, voicebox.transcribe, voicebox.list_captures, and voicebox.list_profiles. You can bind specific voice profiles to specific agents in Settings, so Claude Code uses one voice and Cursor uses another.

The Stories Editor

This is useful for anyone producing audio content that involves more than one voice or requires precise timing control.

How VoiceBox Compares to Cloud Alternatives

VoiceBox is not a direct replacement for every use case. Cloud services offer higher-fidelity voices out of the box and require zero hardware configuration. But for workflows where data ownership, cost control, and API flexibility matter, VoiceBox fills a gap that cloud providers do not address.

Who This Is For

• Content creators producing narrated articles, podcasts, or video voiceovers at scale without per-character costs
• Developers integrating voice I/O into applications via REST API or MCP
• AI agent builders who want their agents to speak in cloned voices
• Privacy-conscious teams that cannot send voice data to third-party servers
• Accessibility projects building voice synthesis tools for people who can't speak in their original voice

Source: github.com/jamiepine/voicebox
Website: voicebox.sh
Docs: docs.voicebox.sh