Your voice studio, minus the cloud bill and the privacy policy

What it does

Voicebox is a desktop voice studio built in Tauri that keeps the entire speech pipeline on your machine. It wraps seven TTS engines — from the tiny Kokoro model to Qwen3-TTS — letting you clone a voice from a few seconds of audio, generate speech in 23 languages, and chain post-processing effects like reverb and compression. It also handles input: a global hotkey triggers Whisper-based dictation that pastes straight into macOS text fields, with an optional local LLM pass to scrub out false starts and ums.

The interesting bit

Rather than treating voice input and output as separate products, Voicebox treats them as one loop. It ships a built-in MCP server and REST API, so any MCP-aware agent can invoke voicebox.speak and talk back to you through a cloned voice — effectively turning your laptop into a private speech endpoint for Claude Code, Cursor, or Cline.

Key highlights

• Seven switchable TTS engines in one UI, including Chatterbox Turbo for paralinguistic tags like [laugh] and Qwen CustomVoice for natural-language delivery control
• Auto-chunking with crossfade for long-form generation up to 50,000 characters
• Multi-track Stories editor for assembling multi-voice podcasts or dialogues
• Native Tauri app with platform-specific acceleration for MLX/Metal, CUDA, ROCm, and DirectML
• Built-in MCP server and REST API for agent integration

Caveats

• Linux pre-built binaries are not yet available; you will need to build from source
• The deep dictation integration — accessibility-verified auto-paste, atomic clipboard handling, and first-run permission gating — is currently documented and most polished on macOS

Verdict

Content creators and agent builders who need offline voice cloning and dictation should try this; if you are looking for a simple, hosted API with no local GPU wrangling, stick with the cloud incumbents.