What it is
VoiceStudio is a desktop app by Palash Debnath, open source since April 2026. In its own words: “fully-local voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation in 646 languages.”
- Clone a voice from a short, clean recording, then type anything and hear it in that voice.
- Dub a video into another language, keeping the timing.
- Make an audiobook: a book in, chapters out.
- Transcribe audio, and dictate.
“No account, no cloud. It only goes online when you ask it to: checking for updates or downloading a model.”voicestudio.sh, FAQ
For comparison: the voice in our reel runs on ElevenLabs Creator, $22 a month for 121,000 credits.
Before you start
| Mac | Apple Silicon (M1 or newer), macOS 13.3 or newer. Intel Macs can run only the app’s interface, not the voice engine. |
| Windows | Windows 10 or 11, x64. Windows on ARM is experimental. |
| Linux | Supported (AppImage or .deb on the releases page). |
| Disk | About 10 GB free for the app, its Python environment and the model weights. On a Windows PC without an NVIDIA card, about 5 GB. |
| Memory | About 8 GB of RAM to start. |
| Graphics card | Not required. It just makes everything faster (Apple Silicon and NVIDIA are used automatically). |
Install it safely
Download only from the official releases page: github.com/debpalash/VoiceStudio/releases. Then check the file against the release’s own checksum list before you open it. It takes one minute.
Mac
- On the releases page, download
VoiceStudio-Electron-<version>-mac-arm64.dmgandSHA256SUMS.txt. - Open Terminal and run the command below. Compare the result with the line for the same file name in
SHA256SUMS.txt. They must match exactly.
shasum -a 256 ~/Downloads/VoiceStudio-Electron-*-mac-arm64.dmg
- Open the DMG and drag VoiceStudio into Applications.
- First launch: macOS may say “the developer cannot be verified”. The app isn’t notarized by Apple yet. If your checksum matched, right-click the app → Open → Open. On macOS 15 Sequoia: double-click once, then System Settings → Privacy & Security → Open Anyway.
- On first run it downloads its own Python, FFmpeg and voice model. Give it time and disk space.
Windows
- Download
VoiceStudio-Electron-<version>-win-x64.exeandSHA256SUMS.txtfrom the releases page. - In PowerShell, in your Downloads folder, run the command below and compare it with
SHA256SUMS.txt.
Get-FileHash .\VoiceStudio-Electron-*-win-x64.exe -Algorithm SHA256
- Run the installer. It installs for your user only, no admin rights needed.
About the one-line installers: the maker also offers curl … | sh (Mac/Linux) and irm … | iex (Windows) installers that verify the checksum for you. We show the manual way so you can see each step before anything runs on your computer.
Your first voice
- In the sidebar, open Voice → Clone.
- Add a clean recording: about 3 to 10 seconds of one person speaking, no music, little echo. Use your own voice, or the voice of someone who said yes.
- Type your script, pick the language, and press Synthesize audio.
- For a video in another language, open Dubbing and choose the file. For a long text, use the audiobook workflow: a book in, chapters out.
The license check (the catch)
The app and each voice model come with their own licenses. Before you sell or publish anything you make with it commercially, check both.
| What | License | What it means for you |
|---|---|---|
| The VoiceStudio app | AGPL-3.0 | Free, including commercial use of the app itself. If you modify it and offer it to others over a network, you must share your source. Building it into a closed product needs their paid license (Pro: $99 a year or $299 once). |
| The default voice model (OmniVoice) | code Apache-2.0 · weights CC-BY-NC | Non-commercial. Its model card: “The pre-trained model is licensed under the CC-BY-NC due to constraints from its training data.” Don’t sell audio made with it. |
| Other voice engines | each has its own | Open the model card before commercial use. Some also have a separate tokenizer license. |
| The voice you clone | permission | Only your own voice, or one you have permission to use. The maker: “Only with the owner’s say-so. Impersonation and fraud are off the table.” |
This is a plain-language summary of the projects’ own notices, not legal advice. When money is involved, read the license text itself.
Checklist
- My computer qualifies: Apple Silicon Mac, or Windows 10/11 x64, or Linux, with ~10 GB free.
- I downloaded from github.com/debpalash/VoiceStudio/releases, nowhere else.
- The SHA-256 of my file matches
SHA256SUMS.txt. - I clone only my own voice, or one I have a clear yes for.
- Before selling anything: the model I used allows commercial use (the default OmniVoice weights don’t).
Prompts for your AI agent
Paste these into Claude, ChatGPT or any agent that can read files and the web.
Check the download
I downloaded VoiceStudio from the official releases page: https://github.com/debpalash/VoiceStudio/releases/latest. The file is at <PATH TO THE FILE>. Download SHA256SUMS.txt from the same release, compute the SHA-256 of my file, and tell me whether it matches the line for that file name. Show me both values. Don't open, install or run anything else.
Check a model’s license
I want to use audio made with the voice model <MODEL NAME> (in VoiceStudio) in something I sell. Open its model card and license (Hugging Face or GitHub) and tell me in plain words: are the model weights allowed for commercial use, and are there conditions (attribution, non-commercial, a separate tokenizer license)? Quote the exact license lines with links. If anything is unclear, say so instead of guessing.
Sources
- github.com/debpalash/VoiceStudio: stars, license, latest release (GitHub API, Oct 4, 2026)
- Install on macOS and install on Windows (the maker’s docs)
- VoiceStudio license notice
- k2-fsa/OmniVoice model card (License section)
- voicestudio.sh: features, FAQ, Pro pricing
- elevenlabs.io/pricing: Creator $22/month, 121,000 credits