VoiceBox
Your Voice, Running Locally: The Open-Source AI Voice Studio That Replaces ElevenLabs and WisprFlow
Cloud voice tools live on opposite sides of the same loop — ElevenLabs handles output, WisprFlow handles input — and both ship your audio to someone else's servers. Voicebox closes that loop on your own machine: a local-first, open-source AI voice studio that does cloning, generation, and dictation in one app.
What it does
- Clone any voice from a few seconds of audio (zero-shot), or pick from 50+ preset voices.
- Generate speech in 23 languages across 7 TTS engines — Qwen3-TTS, LuxTTS, Chatterbox (Multilingual + Turbo), HumeAI TADA, Kokoro, and more.
- Dictate into any app with a global hotkey (push-to-talk or toggle), powered by Whisper-based STT.
- Talk to your agents — a single
voicebox.speakMCP tool lets Hermes Agent, Cursor, or Cline answer in a voice you own. - Express & edit — paralinguistic tags like
[laugh]/[sigh], post-FX (reverb, pitch, chorus), auto-chunking for long scripts, and a multi-track Stories editor for podcasts.
Why it's worth a look
- Complete privacy — models, voice data, and captures never leave your machine.
- API-first — REST API plus a built-in MCP server, so you can wire voice I/O into your own apps and agents.
- Runs everywhere — macOS (MLX), Windows (CUDA), Linux, AMD ROCm, Intel Arc, and Docker. Native Tauri (Rust), not Electron, so it stays light.
Get it
Download for macOS (Apple/Intel), Windows (MSI), or run it via Docker. Prebuilt Linux binaries aren't out yet, but build-from-source instructions are provided.