Voicebox — Docker Edition

A Docker containerization fork of jamiepine/voicebox for NVIDIA GPU users on Windows.
Runs the full app (backend + web UI) in a single container with CUDA acceleration.

---

⚠️ This is not the original project.
This fork adds Docker support for NVIDIA GPU users. If you want the native desktop app for macOS or Windows, go to jamiepine/voicebox.

---

What this fork adds

	Original	This fork
Deployment	Tauri desktop app	Docker container
Platform	macOS / Windows native	Windows + Docker Desktop
GPU	MLX (Apple Silicon) / PyTorch	CUDA via NVIDIA GPU
Web UI	Embedded in desktop app	Served by FastAPI at port 17493
Setup	Install app + Python	docker-compose up

Changes made to the upstream source:

• Added Dockerfile — multi-stage build (Bun frontend → Python runtime)
• Added docker-compose.yml — base CPU service definition
• Added docker-compose.cuda.yml — NVIDIA GPU overlay
• Modified backend/main.py — serves the React web UI from the same FastAPI port
• Updated .gitignore — excludes model weights, local data, and generated audio

---

Requirements

• Docker Desktop with WSL2 backend enabled
• Windows 10/11
• NVIDIA GPU (tested on RTX 5070 Ti, 12 GB VRAM)
• Docker Desktop → Settings → Resources → Enable GPU

CPU-only mode works but TTS generation will be very slow.

---

Quick start

git clone https://github.com/sergio-caracas/voicebox-docker.git

cd voicebox-docker

# Start with NVIDIA GPU acceleration (recommended)
docker-compose -f docker-compose.yml -f docker-compose.cuda.yml up -d

Open http://localhost:17493 in your browser.

First run: The Qwen3-TTS model (~4 GB) downloads automatically on your first generation request.
It is cached in a Docker volume and will not re-download on subsequent starts.

---

Access points

URL	Description
http://localhost:17493	Web UI
http://localhost:17493/docs	FastAPI interactive API docs
http://localhost:17493/health	Health check

---

Stopping and restarting

# Stop (data is preserved in Docker volumes)
docker-compose down

# Restart with GPU
docker-compose -f docker-compose.yml -f docker-compose.cuda.yml up -d

# Rebuild after code changes
docker-compose -f docker-compose.yml -f docker-compose.cuda.yml up -d --build

---

Full documentation

See README_DOCKER.md for complete instructions:

• Volume management and data persistence
• Environment variables
• GPU verification
• Troubleshooting

---

Credits

All credit for the original application goes to Jamie Pine and contributors.
This fork only adds containerization. The core app, AI models, and all features are from the upstream project.

• Upstream repo: jamiepine/voicebox
• Original README: preserved in CONTRIBUTING.md and CHANGELOG.md
• License: MIT

---

voicebox.sh