VoiceStudio

What it is

A free, AGPL-3.0 desktop app for voice work that runs on your own computer. Per its README it does voice cloning, voice design (describe a voice from attributes like gender, age, pitch and accent, no recording needed), video dubbing with timed speech, dictation with a floating widget, transcription and audiobook creation. It also exposes a local OpenAI-compatible API and an MCP server for agents. The README lists 646 languages for the default engine; other engines have their own language sets.

Setup

1. Check the basics. You need Windows 10 or 11 x64 (macOS 13.3+ on Apple Silicon and Linux x86_64 also have builds) and about 10 GB of free disk. An NVIDIA GPU is optional: without one it runs on the CPU, slower. The default engine recommends 6 GB of VRAM.

2. Download the installer from the latest release, VoiceStudio-Electron-X.Y.Z-win-x64.exe (0.5.6 at the time of writing, about 214 MB, 204 MiB as Windows shows it), and compare its hash with the SHA256SUMS.txt in the same release before running it.

   https://github.com/debpalash/VoiceStudio/releases/latestCopy

3. Skip the one-line script for now. The README mentions an install script at voicestudio.sh, but that address returned a 404 when Axo checked on 2026-10-07, so use the installer from the Releases page.

4. Run the setup wizard. It installs for the current user. The installer is not code-signed, so Windows may show a trust warning (the release notes say so).

5. First launch: the app bootstraps Python, FFmpeg and its environment by itself, and asks you to install a model when you need one, so the first run is not offline.

6. Open Voice cloning, choose a voice or add a short, clean reference recording (the docs recommend 5 to 15 seconds; the default engine's page says 3 to 10 is the sweet spot), type your text and press Generate. Use Voice design if you have no recording.

Cheat sheet

• Repository and README and the maker's site.
• Clone only your own voice or one you have permission for. The README says: "Clone voices only with permission."
• No GPU? It runs on the CPU, slower. On AMD with Windows the default engine runs on the CPU (one optional engine has a Vulkan build). Intel Macs get the interface only and need a remote backend.
• Per the license notice, the app license (AGPL-3.0) does not cover the models. The default weights are listed as CC-BY-NC (non-commercial), so check the model licenses before you use the audio commercially or in sponsored work.
• Analytics are off unless you consent. The first launch downloads Python, FFmpeg and model weights.
• There is a separate paid Pro build with a commercial license (the site says $99/yr or $299 one-time). Per the license notice, commercial use of the free app is allowed under the AGPL.
• Agents: the release notes say Claude Code, Cursor and Codex CLI can connect through the local API and MCP server.
• The README's PowerShell one-liner (the install script at voicestudio.sh) returned 404 on 2026-10-07; the Releases installer is the safe route. Any script you pipe into a shell should be read first.
• Not verified by Axo: generation speed, voice quality versus other tools, and how it behaves on your machine.

Axo's take

AXO

Haven't run it myself: this page comes from the maker's README and docs. Try it if you want a free voice studio on your own PC and you have a gaming-class NVIDIA card, or the patience to wait on the CPU. Clone only voices you own, and read the model licenses before you monetize anything.