How to Install and Use VoiceStudio Locally
This guide is based on the open-source project VoiceStudio by debpalash, licensed under AGPL-3.0. StackGuide AI is not affiliated with VoiceStudio.
VoiceStudio is a desktop app for local speech work: voice cloning, voice design, video dubbing, dictation, transcription and audiobook creation, without sending your work to a hosted voice service. As of October 31, 2026, its GitHub repository has 96,513 stars. The app itself is free and, according to its README, has no paid cloud behind it; the speech models you download run on your own machine.
What VoiceStudio is, and who it is for
Think of it as a local studio, not a single voice model. You pick an engine and a model in the Model Manager, then use one workspace for cloning a voice, designing a voice from a description, dubbing a video, dictating, transcribing, or narrating a longer text. The README says the app asks before it downloads any model, so you choose only the engines and models you need rather than downloading everything at once.
It suits creators, educators and developers who want speech generated on their own hardware, and who are happy to manage model downloads and hardware limits themselves. It is not a hosted text-to-speech API, and it is not a shortcut to copying someone else's voice: the README's own licence note says to clone a voice only with its owner's permission, and this guide repeats that as a hard rule.
Prerequisites
- A supported system: macOS on Apple Silicon, Windows 10/11 (x64) or Linux (x64) are listed as stable in the README. Docker, Android 10+ and iOS 17+ are also listed, the last two as previews.
- Disk space for models: speech models are large downloads. Start with one engine and one model, and add more only when you need them.
- A GPU helps, but is not required: the README says the app is fully usable on CPU, just slower, and lists Metal on Apple Silicon and NVIDIA CUDA on Windows and Linux, with AMD ROCm as an opt-in on Windows.
- A clean reference recording, if you plan to clone a voice: use a voice you own or have explicit permission to use, recorded cleanly and without background noise or other speakers.
Step 1: Install VoiceStudio
macOS or Linux. Run the official installer script:
curl -fsSL https://voicestudio.sh/install | shWindows. Run this in PowerShell:
irm https://voicestudio.sh/install.ps1 | iexAlternatively, download an installer from the latest release on GitHub: a .dmg for macOS, a .exe for Windows, or a .AppImage or .deb for Linux. Docker images are published to GHCR; the README points to its own Docker install guide for that path, so check the official docs rather than guessing a container command.
If voicestudio.sh is unreachable, the README gives the script straight from the repository:
curl -fsSL https://raw.githubusercontent.com/debpalash/VoiceStudio/main/scripts/install.sh | shAs with any install script piped to a shell, you can download it first, read it, and only then run it. That caution costs a minute and is worth it.
Step 2: Pick a model, then generate your first speech
Open the Model Manager, start with the default VoiceStudio engine, and download one model. The app asks before downloading, so confirm only the model you intend to use first.
Then open Voice cloning, pick a voice or add a clean reference recording, type your text and generate. For a first run, keep the text short: a short sentence tells you whether the voice, the model and the audio output all work, before you queue a long narration job.
From there, the other workspaces follow the same pattern: Voice design describes a voice in words and generates it; Video dubbing translates and dubs a video with speech timed to the original; dictation and transcription work locally; and narration jobs can be queued in batches for stories and audiobooks. The README's documentation table links a separate guide for each of these, including multi-speaker dubbing and exporting dubs, so follow the official guide for the job you are actually doing.
Step 3: Keep models and output organised
Download only the models you need, and note which engine and model produced each output file. Model weights, voices and generated output have their own licences, which the Model Manager shows for each model, so a file's licence is not automatically the same as the app's AGPL-3.0 licence. If you later share an audio file, check the model and voice licences first, and keep the reference recording's permission in mind.
Common problems and fixes
| Problem | Fix |
|---|---|
| macOS says VoiceStudio is damaged and cannot be opened | According to the README, this is Gatekeeper blocking the app, not a broken download. Follow the official macOS install guide's Gatekeeper fix. |
| Generation is very slow | You are probably running on CPU, or with a model larger than your hardware handles comfortably. Try a smaller model or a GPU-backed engine if your machine has one, and keep the first test short. |
| A model download is huge or slow | Download one model at a time from the Model Manager, and check the official model-download guide for that engine before starting a second large download. |
| The cloned voice sounds wrong | Use a cleaner reference: one speaker, little background noise, and permission from the voice's owner. A noisy or mixed recording cannot be fixed reliably after the fact. |
| You are upgrading from version 0.5.3 or older | The README says to install the new app separately, because the older Tauri app is a different build. Follow its Electron migration guide. |
FAQ
Do I need an account or a cloud subscription? No, according to the README: VoiceStudio is free and has no paid cloud behind it. Sponsors fund maintainer time and infrastructure, but the app itself does not require a paid cloud account.
Can I use it offline? Once the app and the model you need are downloaded, generation runs locally. The first install and each model download need the internet.
Can I clone any voice I find online? No. Clone a voice only with its owner's permission. That is the README's rule, and it is also the only safe basis for sharing or publishing the result.
What if the installer script site is down? Use the repository script or a release installer from GitHub, as shown in Step 1, and check the official installer guide for prerequisites.
Licence and attribution
VoiceStudio is licensed under AGPL-3.0. In plain terms, and not as legal advice: you may use it commercially under that licence's terms, but if you modify it and offer it over a network, AGPL-3.0 requires you to make the corresponding source available to users of that network service. That is the key difference from a permissive licence such as MIT. Model weights, voices and generated output are covered by their own licences, shown in the Model Manager, so check those separately. Keep the app's licence notice with any copy you redistribute, and check the official LICENSE-NOTICE file for the full details.
What to try next
Once a short clip sounds right, try one longer job end to end: narrate a short story, transcribe a recording you made yourself, or dub a short video clip. That sequence tests the three things that matter for local speech work: model choice, output quality, and how long a real job takes on your hardware.
This guide was checked against the official README and repository as of October 31, 2026. StackGuide AI did not run this tool for this guide, so this guide makes no claim based on running the tool; the commands below are the ones the official README publishes.