VoiceStudio

debpalash/VoiceStudioReviewed by AI

What it does

A desktop studio for AI speech that runs on your own machine. It clones a voice from a recording, designs new voices, dubs videos with timed speech, dictates with a floating widget, transcribes audio and builds audiobooks, in many languages. The models run locally; an NVIDIA GPU or Apple Silicon speeds it up, and the README says it also works on the CPU, just slower. There are installers for Windows, macOS and Linux, plus preview apps for Android and iPhone. The first time you use a feature the app asks before downloading the model it needs.

How easy it is to install

B · installer plus one simple step · Windows

Windows B
macOS B
Linux B

Each system has a ready installer in the latest release (VoiceStudio-0.5.7-win-x64.exe, the .dmg for Apple Silicon, .AppImage/.deb/snap on Linux); after installing you still download each speech model inside the app the first time you use it. The README marks Intel Macs as "App UI only, connects to a remote backend", so an older Mac is harder.

Numbers on GitHub

Stars: 96,872on Oct 31, 2026
Downloads: 1,968,463277,283 this week
Growth: +7,609stars in 5 days

Stars over time

Facts

Runs on: Windows, macOS, Linux
You need to install: Nothing extra
To build from source: Python, Node.js, NVIDIA graphics card, FFmpeg
Disk space: 212 MB–636 MB · from the installer
Accounts or keys: An OpenAI-compatible AI model, for cinematic dubbing and AI translation, Hugging Face token, for downloading some models
Cost: Free · AGPL-3.0 and free to use. pyproject.toml says a separate commercial licence is sold for closed-source use (contact hi@voicestudio.sh), and model weights, voices and generated output have their own terms.
License: AGPL-3.0 (Copyleft)
Can replace: ElevenLabs, Descript, Otter.ai
Last update: Nov 1, 2026

Risk signals (6)

Attentionvoice cloning can impersonate real people
The repository description offers "voice cloning, voice design, video dubbing" and the README's "License & responsible use" says "Clone a voice only with its owner's permission." A cloned voice can be used to imitate a real person without their consent, and some countries restrict that.
AI review
Infomacos gatekeeper blocks the app
README note: "On macOS, 'VoiceStudio is damaged and can't be opened' means Gatekeeper blocked the app, not that the download is broken", pointing to a "Gatekeeper fix" page.
AI review
Infomodel weights and output under separate terms
README: "Model weights, voices and generated output have their own licences, which the Model Manager shows for each model", and the app licence does not cover them.
AI review
Infooptional analytics off by default
pyproject.toml comment on the posthog dependency: "Opt-in product analytics (core/analytics.py). Inert unless the build ships a POSTHOG_PROJECT_TOKEN *and* the user opts in — default OFF."
AI review
Infodependency installed from a direct url
pyproject.toml: "kittentts @ https://github.com/KittenML/KittenTTS/releases/download/0.8.1/kittentts-0.8.1-py3-none-any.whl" — one speech engine is pulled straight from a GitHub release file instead of the usual package index.
AI review
Infobundles a video downloader
pyproject.toml lists "yt-dlp>=2026.7.4" as a dependency, so the app can pull videos from websites to dub or transcribe them. Whether that is allowed depends on the site and on local law.
AI review

Evidence is shown in its original language.

What Radar could not check

How much disk space the speech models add on top of the installer, since they download on first use
Whether the app runs at a usable speed on a PC with no dedicated graphics card
How good the voice cloning, dubbing and transcription quality is in practice
Whether the Windows-on-ARM, Intel Mac and phone builds work, since the README marks them experimental, limited or preview
Whether anything really leaves your computer, beyond the model downloads the app asks about