Voicebox Voice Synthesis Studio Skill by [ara.so](https://ara.so) — Daily 2026 Skills collection. Voicebox is a local first, open source voice cloning and TTS studio — a self hosted alternative to ElevenLabs. It runs entirely on your machine (macOS MLX/Metal, Windows/Linux CUDA, CPU fallback), exposes a REST API on localhost:17493 , and ships with 5 TTS engines, 23 languages, post processing effects, and a multi track Stories editor. Installation Pre built Binaries (Recommended) Platform Link macOS Apple Silicon https://voicebox.sh/download/mac arm macOS Intel https://voicebox.sh/download/mac intel Windows https://voicebox.sh/download/windows Docker docker compose up Linux requires building from source: https://voicebox.sh/linux install Build from Source Prerequisites: [Bun](https://bun.sh), [Rust](https://rustup.rs), [Python 3.11+](https://python.org), Tauri prerequisites Architecture Layer Technology Desktop App Tauri (Rust) Frontend React + TypeScript + Tailwind CSS State Zustand + React Query Backend FastAPI (Python) on port 17493 TTS Engines Qwen3 TTS, LuxTTS, Chatterbox, Chatterbox Turbo, TADA Effects Pedalboard (Spotify) Transcription Whisper / Whisper Turbo Inference MLX (Apple Silicon) / PyTorch (CUDA/ROCm/XPU/CPU) Database SQLite The Python FastAPI backend handles all ML inference. The Tauri Rust shell wraps the frontend and manages the backend process lifecycle. The API is accessible directly at http://localhost:17493 even when using the desktop app. REST API Reference Base URL: http://localhost:17493 Interactive docs: http://localhost:17493/docs Generate Speech Voice Profiles Generation Queue & Status Models TypeScript/JavaScript Integration Basic TTS Client Poll for Completion Stream Status with SSE Download Audio as Blob Python Integration TTS Engine Selection Guide Engine Best For Languages VRAM Notes qwen3 tts (0.6B/1.7B) Quality + instructions 10 Medium Supports delivery instructions in text luxtts Fast CPU generation English only ~1GB 150x realtime on CPU, 48kHz chatterbox Multilingual coverage 23 Medium Arabic, Hindi, Swahili, CJK + more chatterbox turbo Expressive/emotion English only Low (350M) Use [laugh] , [sigh] , [gasp] tags tada (1B/3B) Long form coherence 10 High 700s+ audio, HumeAI model Delivery Instructions (Qwen3 TTS) Embed natural language instructions directly in the text: Paralinguistic Tags (Chatterbox Turbo) Environment & Configuration Docker configuration ( docker compose.yml override): Common Patterns Voice Profile Creation Flow Batch Generation with Queue Long Form Text (Auto Chunking) Voicebox auto chunks at sentence boundaries — just send the full text: Troubleshooting API not responding GPU not detected Model download fails / slow Out of VRAM — unload models Audio quality issues Use 5–30 seconds of clean, noise free speech for voice samples Multiple samples improve clone quality — upload 3–5 different sentences For multilingual cloning, use chatterbox engine Ensure sample audio is 16kHz+ mono WAV for best results Use luxtts for highest output quality (48kHz) in English Generation stuck in queue after crash Voicebox auto recovers stale generations on startup. If the issue persists: Frontend Integration (React Example)

voicebox-voice-synthesis