Voicebox

// Overview

Voicebox is a local-first desktop voice studio that covers both halves of the voice loop — speech out and speech in. It clones a voice from a few seconds of reference audio, generates speech across seven TTS engines and up to 23 languages, transcribes with Whisper, and pastes dictation into whatever text field is focused. Models, voice data and captures stay on the machine; the project positions itself as a free and open-source alternative to ElevenLabs and WisprFlow in one app.

The engines are switchable per generation and have different strengths: Qwen3-TTS and Qwen CustomVoice for multilingual cloning and natural-language delivery control, LuxTTS for a lightweight English model at 48 kHz, Chatterbox Multilingual for the broadest language coverage, Chatterbox Turbo for fast English with paralinguistic tags like `[laugh]` and `[sigh]`, HumeAI TADA for long coherent audio, and Kokoro for 50 curated preset voices in an 82M model. Generations are versioned with provenance — original, effects versions and re-seeded takes each record their lineage — and long text is auto-chunked at sentence boundaries and crossfaded back together.

Everything is also reachable programmatically. A REST API on localhost exposes generation, agent speech, transcription and profile listing, and a built-in MCP server lets any MCP-aware agent — Claude Code, Cursor, Cline and others — speak in a voice you have cloned with a single `voicebox.speak` call. Voice profiles can carry a free-form persona that a bundled local LLM uses for Compose, Rewrite and Respond modes, which agents can invoke over the same MCP surface. The app is built with Tauri rather than Electron and ships for macOS, Windows and Docker, with Linux built from source.

// What it does

• Seven switchable TTS engines — Qwen3-TTS, Qwen CustomVoice, LuxTTS, Chatterbox Multilingual, Chatterbox Turbo, HumeAI TADA and Kokoro
• Zero-shot voice cloning from a reference sample, plus 50+ curated preset voices, across up to 23 languages
• Global dictation hotkey with push-to-talk and toggle chords, accessibility-verified paste into the focused field on macOS, and optional LLM cleanup of filler words
• Whisper speech-to-text on MLX, CUDA, ROCm, DirectML or CPU, with every dictation and upload kept in a Captures tab for replay and re-transcription
• Built-in MCP server and REST API so agents and scripts can speak, transcribe and browse profiles
• Eight post-processing effects via Spotify's pedalboard, with presets assignable per voice profile
• Stories editor — a multi-track timeline for conversations, podcasts and narratives, with per-clip version pinning
• Auto-chunked unlimited-length generation with crossfade, and an async queue that keeps the UI free while audio renders

// Getting started

Voicebox ships as a native desktop app built with Tauri. Installers are published for macOS and Windows; Docker and build-from-source cover the rest.

docker compose up

[laugh] [chuckle] [gasp] [cough] [sigh] [groan] [sniff] [shush] [clear throat]

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

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

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

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Reach for it to narrate scripts, articles or chapters in a cloned voice without sending the audio to a cloud service
• Reach for it as a system-wide dictation tool that pastes cleaned-up text into any application
• Reach for it when a long-running coding agent should announce completions and questions out loud
• Reach for it to compose multi-voice conversations, podcasts or narratives on a timeline

// How Voicebox compares

Voicebox alongside other open-source audio, music & voice tools AI/TLDR tracks, ranked by GitHub stars.