# VoiceStudio

> Clone voices, dub video and dictate entirely on your own hardware

- Category: Multimodal Generation / Audio, Music & Voice
- Language: Python
- Licence: AGPL-3.0
- Repository: https://github.com/debpalash/VoiceStudio
- Website: https://voicestudio.sh
- Canonical page: https://ai-tldr.dev/tools/voicestudio/

## In short

VoiceStudio (previously OmniVoice-Studio) is a desktop application that bundles voice cloning, voice design, video dubbing, dictation and long-form audio production into one local workflow. It sits in the Audio, Music & Voice category of the Multimodal Generation tools and is open-source, written in Python and licensed AGPL-3.0. To get started: install a packaged build.

## Overview

VoiceStudio (previously OmniVoice-Studio) is a desktop application that bundles voice cloning, voice design, video dubbing, dictation and long-form audio production into one local workflow. Rather than shipping a single model, it acts as a front end over an engine catalogue — 16 text-to-speech engines and 11 speech-recognition engines, with a 646-language TTS catalogue whose real coverage and quality depend on which engine you pick. Engines are swapped from the Model Catalogue or with a keyboard shortcut, so a project can move between a fast small model and a heavier expressive one without leaving the app.

The project is local-first: voices, projects, settings and outputs stay on the machine by default, and the core creation path needs no account, API key or usage meter. Network-backed features are explicit opt-ins. Compute routing is automatic across CUDA, Apple Silicon MPS/MLX, ROCm on Linux and plain CPU, with optional remote workers for machines that cannot run the backend locally. Packages exist for macOS 13.3+ on Apple Silicon, Windows 10/11 x64, Linux x86_64 with glibc 2.39+, and Docker (CUDA, ROCm, CPU and worker-only profiles).

Beyond the desktop UI, VoiceStudio exposes a local REST/SSE/WebSocket API, an OpenAI-compatible audio API and an MCP server for synthesis and transcription, so it can be driven by scripts and agents as well as by hand. The application is AGPL-3.0; models you download keep their own upstream licence terms. The project describes itself as an active beta and points users at tagged releases for stable work.

## What it does

- Zero-shot voice cloning from a short reference clip, plus voice design from age, accent, pitch, style and delivery instructions
- Video dubbing that transcribes, translates, preserves speakers, synthesizes and exports the finished video
- Stories and audiobooks: multi-voice scripts, EPUB/PDF import, chapter rendering and .m4b export
- A dictation widget bound to a system-wide shortcut, with live transcription and optional local-LLM cleanup
- Supporting audio tools — Demucs vocal isolation, Pyannote/WhisperX speaker diarization, a batch job queue and AudioSeal watermark embedding and detection
- Automation surfaces: a local REST/SSE/WebSocket API, an OpenAI-compatible audio API and an MCP server
- GPU auto-detect across CUDA, MPS, ROCm and CPU, with per-engine checks and optional remote workers

## Getting started

The normal path is a packaged download rather than a source build. The first launch creates a managed Python environment and downloads the default model; later launches reuse both. On macOS the first launch needs a one-time right-click, then Open.

1. Install a packaged build

   Grab the package for your platform from the latest release: an Apple Silicon DMG for macOS 13.3+, an x64 MSI for Windows 10/11 (pick the current-user build to install without admin rights), or an AppImage for Linux x86_64 with glibc 2.39+.

```text
https://github.com/debpalash/VoiceStudio/releases/latest
```

2. Clone your first voice

   Open Voice Cloning, add a clean voice sample — three seconds works, 5 to 15 seconds usually gives a better prompt — then enter text, choose a language and select Generate.

3. Switch engines when the default doesn't fit

   Engines are installed, removed, selected and routed from the Model Catalogue, which shows engine, device and install state. Ctrl/Cmd+E switches engines straight from the status bar.

4. Run from source instead

   After installing the development prerequisites from CONTRIBUTING.md, the repo builds with bun. Use `bun run dev` for the browser UI.

```bash
git clone https://github.com/debpalash/VoiceStudio.git
cd VoiceStudio
bun install
bun run desktop
```

5. Diagnose a failed setup

   The app ships self-checks; run them from Settings → About → Run self-check, or from the command line. A scrubbed diagnostic bundle can be saved from the app when opening an issue.

```bash
uv run python backend/main.py --diagnose --deep
```

## When to use it

- Produce cloned-voice narration for audiobooks or long-form video without sending scripts and reference audio to a hosted provider
- Dub an existing video into another language while keeping the original speaker layout
- Add system-wide local dictation with transcription that never leaves the machine
- Serve synthesis and transcription to your own scripts or agents through the local REST, OpenAI-compatible or MCP endpoints
