VoiceStudio Guide (2026): Install the Open-Source Local ElevenLabs Alternative, Generate Speech and Dub Video

TL;DR: VoiceStudio is an open-source app, hosted at github.com/debpalash/VoiceStudio, that runs voice cloning, voice design, video dubbing, dictation, transcription and audiobook creation on your own computer. This guide covers what it is, how to install it on Windows, macOS and Linux, how to generate speech and start a dubbing project, and how to use it responsibly. The steps below follow the project's README and install docs as of 16 Oct 2026. The project changes quickly, so check the README before you install.

What VoiceStudio is

The repository describes VoiceStudio as "the open-source, fully-local ElevenLabs alternative." According to the README, it offers:

• Voice cloning and voice design
• Video dubbing with timed speech
• Dictation with a floating widget, plus transcription
• Audiobooks, stories and batch jobs
• A local API and MCP support for agents

The default engine is described as VoiceStudio, powered by k2-fsa/OmniVoice, and other engines can be chosen. The README says local workflows run on your hardware, remote services are optional, and usage analytics requires consent. The app is licensed under AGPL-3.0, but model weights and voices have separate conditions (see the responsible-use section).

What you need

Per the README:

Hardware	Support
NVIDIA GPU (Windows/Linux)	CUDA acceleration
Apple Silicon	Metal (MPS) acceleration
No dedicated GPU	Works on the CPU, slower; the setup installs a small CPU build of PyTorch (about 5 GB free disk)
Intel Mac	App UI only; connect to a remote backend
Windows on ARM	Experimental

The Windows guide says to expect about 10 GB of free disk for the app, its Python environment and model weights. On a CPU-only PC, it suggests choosing lighter engines (it names KittenTTS, Supertonic-3 and PocketTTS) and a small Whisper model.

Step 1: Install

Option A, one command (macOS and Linux) from the README:

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

Windows, in a regular PowerShell window:

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

Option B, download an installer from the Releases page: .dmg for macOS, .exe for Windows, and .AppImage or .deb for Linux. The Windows guide advises comparing your download with the release's SHA256SUMS.txt before running it. Platform guides for macOS, Windows, Linux and Docker are in the repo.

Piping a script into your shell runs whatever the server returns. If you prefer, download the script first and read it, or use the release installers instead.

Option C, run from source. The README lists Git, Node.js 22+, Bun and Rust/Cargo as requirements, then:

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

On first launch of the packaged Windows app, the docs say Python, FFmpeg and model weights are set up by the app itself. Don't close the app while the setup runs.

Step 2: Generate speech (text to speech)

The README's quick start is short: open Voice cloning, choose a voice or add a clean reference recording, enter your text, and generate. Install the required model when the app prompts you. Model downloads can be large, so use a wired connection if you can. If a model fails to download, the project's troubleshooting and model downloads pages are the place to start.

Practical tips (general good practice, not from the README):

• Use a clean reference clip: one speaker, no background music, no echo.
• Start with a short sentence to test the voice before you render a long script.
• On a CPU, pick a light engine first and move to a heavier one only if the speed is acceptable.

Step 3: Dub a video

The README lists video dubbing with timed speech as a core workspace. The app has a dubbing workspace alongside the cloning, voice design and model screens. Repository pages for features and engines are linked from the README's documentation table. A sensible workflow, based on those descriptions, is:

1. Open the dubbing workspace and import your video.
2. Let the app transcribe the original speech, then review the text.
3. Translate or edit the script for the target language.
4. Choose a voice you have the right to use and generate the dub.
5. Check the timing and export.

The exact button names can change between versions, so follow the in-app prompts and the project's docs.

Optional: use the local API

The speech-platform docs say the Python backend serves audio over HTTP on port 3900 by default, including an OpenAI-compatible transcription endpoint, POST :3900/v1/audio/transcriptions, and an MCP endpoint. Loopback clients need no credential. If you expose it beyond your own machine, the docs recommend a trusted network and HTTPS/WSS, and warn that an API key does not give network isolation.

Voice cloning: consent only

Voice cloning can be misused for impersonation and fraud. The project's own licence section says to clone voices only with permission. Please follow these rules:

• Clone only your own voice, or the voice of someone who has clearly agreed, ideally in writing.
• Do not imitate celebrities, public figures, colleagues or family members without consent.
• Do not use cloned voices to deceive anyone, for example in phone scams or fake endorsements.
• Label synthetic audio where it could be mistaken for a real person.
• Check licence terms. The README states that model weights, tokenizers, voices and generated-output conditions are separate from the app licence, and that a paid app licence does not grant those rights.

Troubleshooting

• Slow generation: you are likely on the CPU. Use a lighter engine.
• Setup stuck or failing: see the project's troubleshooting page and its GitHub issues.
• Intel Mac: the README says only the app UI runs locally, and you need a remote backend.

Sources: the VoiceStudio README, the Windows install guide and the speech-platform doc in the debpalash/VoiceStudio repository (read 16 Oct 2026). We have not benchmarked the software. Commands and options can change, so confirm them against the current README.

---

← Back to all posts