Voicebox: Open‑Source Voice Studio Powered by Qwen3‑TTS

Voice synthesis is no longer the domain of a handful of cloud‑based services. With Voicebox—a free, local‑first application built on Qwen3‑TTS—developers and creators can own their voice data, edit multi‑track audio like a digital audio workstation, and generate natural speech faster than ever on Apple Silicon.

What is Voicebox?

• Local‑first: All inference, cloning, and editing run on your hardware—no internet needed, no subscription fees.
• Open source: MIT licensed, fully community‑driven.
• Multi‑track editing: Think DAW meets text‑to‑speech.
• Built with modern stack: Tauri (Rust) for the desktop, React/TS for the UI, FastAPI for the API, MLX/Metal for GPU acceleration.
• Powered by Qwen3‑TTS: Alibaba’s breakthrough model that can clone a voice from just a handful of seconds, producing high‑fidelity, expressive speech.

Core Features at a Glance

Architecture Snapshot

graph TD
  A[React‑TS Frontend] -->|REST| B[FastAPI Backend]
  B -->|PyTorch/MLX| C[Qwen3‑TTS Engine]
  B -->|Whisper| D[Transcription]
  B -->|SQLite| E[Database]
  subgraph Desktop
    F[tauri (Rust)] --> A
  end
  subgraph Web
    G[React‑TS app] --> A
  end

• Frontend: React with TypeScript, Tailwind CSS, Zustand & React Query for state and data fetching.
• Backend: FastAPI providing a typed API, automatic docs, and async performance.
• Models: Qwen3‑TTS and Whisper are available both as PyTorch and MLX backends, giving platform flexibility.
• Persistence: SQLite stores voice profiles, project metadata, and generation history.

How to Get Started

1. Install

# On macOS (Apple Silicon)
brew install qt@5  # for Tauri dependencies
bun install
cd backend && pip install -r requirements.txt
bun run dev   # Launch the desktop app

For Windows or Intel‑based macOS, download the MSI or ZIP from the releases page.

2. Clone a Voice

1. Open the app and click Create Profile.
2. Record or upload 5–10 seconds of clear speech.
3. The model will generate a profile called My Voice.
4. Export the profile if you want to share it.

3. Build a Story

1. Drag the new profile onto the timeline.
2. Type your script or paste from a document.
3. Use Batch Generation to synthesize the whole script.
4. Arrange clips, trim, and mix using the timeline tools.
5. Export the final mix.

Use Cases Where Voicebox Shines

Extending Voicebox

• Plugin System: Add new voice models (e.g., XTTS, Bark) or audio effects as separate Tauri packages.
• Mobile Companion: Future plans include a phone app to control a desktop Voicebox instance over a LAN.
• Real‑Time Synthesis: A coming feature will stream generated audio as it is produced, enabling live performances.

Community & Contribution

Voicebox is built to be welcoming and open:

• Contributing: Pull requests are encouraged; see CONTRIBUTING.md for setup.
• Security: Follow SECURITY.md to report issues responsibly.
• Releases: New stable builds are published on GitHub Releases.
• Docs: Comprehensive API docs are available at http://localhost:8000/docs when running.

Bottom Line

Voicebox turns a laptop into a professional, privacy‑preserving voice studio. Whether you’re prototyping a speech‑based game, drafting a podcast, or building a personal accessibility tool, you no longer have to rely on costly cloud APIs. Jump in today, fork the GitHub repo, and start building voice experiences that stay under your control.