Voicebox: 36K Stars Open-Source AI Voice Studio — Clone, Dictate, Speak Locally

Created

TL;DR: Voicebox is the free open-source AI voice studio that replaces both ElevenLabs (speech output) and WisprFlow (speech input) in one app — plus MCP-powered voice for your AI agents. 7 TTS engines, 23 languages, zero-shot voice cloning, global dictation hotkey. Every model runs locally. 36,000+ GitHub Stars, MIT license.

Why Voicebox Matters

The AI voice world has two incumbents: ElevenLabs owns the output side (TTS generation), WisprFlow owns the input side (dictation). Both are commercial cloud services — your voice data goes to their servers, and you pay monthly.

Voicebox does both in one open-source project — and adds AI Agent voice output via MCP, so Claude, Cursor, and Cline can talk to you using a voice you’ve cloned.

Key stats:

  • 📦 GitHub Stars: **36,000+**（Repository）
  • 🍴 Forks: 4,300+
  • 📝 License: MIT (fully free and open-source)
  • 🏠 Website: voicebox.sh
  • ⚡ Tech Stack: Tauri (Rust) + React + FastAPI (Python)

Core Features

🎤 Voice Cloning & TTS

Clone any voice from just seconds of audio with zero-shot learning. 7 TTS engines included:

  • Qwen3-TTS / Qwen CustomVoice — Alibaba’s Qwen team, highly natural
  • LuxTTS — Lightweight high-quality synthesis
  • Chatterbox / Chatterbox Turbo — Paralinguistic tags like [laugh], [sigh] for expressive speech
  • HumeAI TADA — Emotion-aware voice synthesis
  • Kokoro — 50+ curated preset voices

Covers 23 languages from English, Chinese, Japanese, Arabic to Swahili.

🎧 Post-Processing Effects

Pitch shift, reverb, delay, chorus, compression, filters — studio-grade control over every generated audio.

⌨️ Global Dictation (Voice Input)

Press a global hotkey, speak, and text appears in any text field. Powered by Whisper STT with push-to-talk and toggle modes.

🤖 AI Agent Voice Output

The killer feature: Voicebox ships with a built-in MCP server. AI coding assistants (Claude Code, Cursor, Cline) call a single oicebox.speak tool, and your AI agent reads out code reviews, errors, or messages in a voice you’ve cloned. It’s like giving your AI assistant a mouth.

🎭 Voice Personalities

Bind a free-form persona description to any voice profile. Combined with the bundled local LLM (Qwen3 0.6B/1.7B/4B), you get Compose, Rewrite, and Respond modes. Agents can also invoke these via MCP.

📖 Stories Editor

Multi-track timeline editor for conversations, podcasts, and narratives. Auto-chunking with crossfade for unlimited-length content.

Technical Highlights

🦀 Native Performance (No Electron)

Voicebox uses Tauri (Rust) instead of Electron, giving you native desktop performance with lower memory usage.

🔒 Complete Local Privacy

Every model runs locally — voice data, cloned voices, dictation history never leave your machine. Three inference backends:

  • MLX (Apple Silicon Mac)
  • CUDA (NVIDIA GPU)
  • CPU fallback (works without GPU)

Also supports AMD ROCm and Intel Arc GPUs.

🌐 REST API + MCP

Full REST API docs at http://127.0.0.1:17493/docs, plus a built-in MCP server for seamless AI tool integration.

Platform Support

Platform Download
macOS (Apple Silicon) DMG installer
macOS (Intel) DMG installer
Windows MSI installer
Linux Build from source
Docker docker compose up

Quick Start

  1. Download the installer from GitHub Releases
  2. First launch downloads required AI models (~2-3GB, varies by engine selection)
  3. Import audio or pick a preset voice to start cloning
  4. Press the global hotkey in any app to start dictating
  5. Configure Voicebox MCP in your MCP-aware AI tool and your agent can speak

💡 Windows users: Check the Troubleshooting Guide if you run into installation issues.

Summary

Voicebox is one of the most exciting open-source projects in AI voice for 2026. It’s not just a TTS tool or a dictation tool — it’s a complete voice I/O stack: input (dictation), processing (LLM refinement), output (TTS), all wired into the MCP ecosystem. For the first time, AI agents have a real voice of their own.

Whether you’re a content creator, an AI developer, or just want a private voice assistant, Voicebox is worth your time.

References: