VoiceStudio: Free Local Voice Cloning (Setup Guide)

[tldr] What it's: VoiceStudio is an open-source (AGPL-3.0) desktop app that does voice cloning, voice design, video dubbing, dictation, transcription and audiobook narration entirely on your own computer — no subscription, no upload, no per-character billing. It runs on Windows 10/11, macOS (Apple Silicon) and Linux, and claims support for 646 languages. Install is one command. [/tldr]

[howto_callout type="tip"] Bookmark this page (🔖): installers change often; the commands and the “first 3 minutes” steps below are the part you'll come back to. [/howto_callout]

Why people are switching from cloud voice tools

Cloud voice services charge per character, keep your audio on their servers, and stop working when your card expires or the API changes. A local studio flips all three: free after install, private by default, works offline — including on a plane.

	Cloud voice service	VoiceStudio (local)
Cost	Monthly plan or per character	Free (open source)
Your audio	Uploaded to their servers	Stays on your disk
Internet needed	Always	Only for the first model download
Limits	Quotas, rate limits	Your hardware
Voice rights	Vendor terms	You still need the speaker's permission

What you can actually do with it

Feature	Use it for
Voice cloning	Feed a clean reference recording, then generate new speech in that voice
Voice design	Describe a voice in words and generate it, no sample needed
Video dubbing	Translate a video and dub it with timing matched to the original
Dictation & transcription	Floating dictation widget, plus local speech-to-text for meetings and interviews
Narration & batch	Audiobooks, stories and queued jobs that run overnight
Automation	Local API, MCP integration and watch folders for pipelines

You only download the models you pick, the app asks before pulling anything large.

Requirements, the honest version

Windows: 10 or 11, 64-bit. NVIDIA CUDA works out of the box; AMD ROCm is opt-in; CPU works but is slower.
macOS: 13.3 or newer on Apple Silicon (Metal acceleration).
Linux: desktop build available, GPU acceleration through CUDA or Vulkan depending on the engine.
Disk: plan a few GB free, local speech models are the size of small apps, and you choose how many to keep.
RAM: 8 GB is workable for short clips; longer dubbing and audiobooks want 16 GB.

No GPU? You can still run it. Generation takes longer, so start with short sentences rather than a 40-minute audiobook.

Install in one command

Windows (PowerShell)

irm https://voicestudio.sh/install.ps1 | iex

Or download the EXE from the releases page and run it.

macOS (Homebrew)

brew tap debpalash/tap
brew trust --cask debpalash/tap/voicestudio
brew install --cask debpalash/tap/voicestudio

No Homebrew? Use the one-line installer:

curl -fsSL https://voicestudio.sh/install | sh

Linux / any Unix (script)

curl -fsSL https://voicestudio.sh/install | sh

Uninstall keeps your data:

curl -fsSL https://voicestudio.sh/install | sh -s -- --uninstall

From source (developers)

git clone https://github.com/debpalash/VoiceStudio.git
cd VoiceStudio
bun run dev

Your first three minutes

Open Voice cloning and pick the bundled demo voice, or add a clean recording of your own (10, 30 seconds of speech, no music, no background noise).
Type or paste the text you want spoken. Keep the first test short, one or two sentences.
Generate, listen, then adjust speed and expressiveness if the engine offers it.
Happy with the result? Export the WAV, or queue a longer batch job.

Real use cases that pay for themselves

YouTube & TikTok: narrate the same video in Bangla, Hindi and English without hiring three voice artists.
Courses & audiobooks: turn your written lessons into audio while you sleep.
Accessibility: dictate notes hands-free, transcribe interviews locally.
Localisation: dub a product demo for another market and keep the timing intact.

Ethics and licence, read this before you clone

The app is AGPL-3.0; commercial use is allowed under its terms, but if you modify and distribute it, the source must stay open.
Model weights and voices carry their own licences, the Model Manager shows each one.
Clone a voice only with the speaker's permission. Impersonating someone for money, fraud or fake statements is illegal in most countries and isn't a “grey area”.
Output of a cloned voice should be labelled as synthetic when it's published publicly.

FAQ

Is it really free?

Yes, the app is open source and runs locally. You pay nothing per character. Your only costs are your own electricity and disk space.

Can it run on a laptop without a GPU?

Yes, via CPU, just slower. Short sentences and dictation are fine; long audiobooks become an overnight job.

Does it work without internet?

After the models are downloaded, yes. The first run needs a connection to fetch the model you chose.

How is this different from buying an ElevenLabs subscription?

Same category of output, different trade-off: no recurring fee and no uploads, but you supply the hardware and install updates yourself.

Short, tested walkthroughs on the same topics: