VoiceStudio: Open-Source Local Voice Cloning (ElevenLabs Alternative) — Install Guide

What is VoiceStudio?

VoiceStudio (debpalash/VoiceStudio, ★96k+) is the open-source, fully-local alternative to ElevenLabs: an AI voice studio that runs entirely on your own device. Clone, design, dub, dictate, transcribe and narrate in 646 languages — with no paid cloud behind it (AGPL-3.0).

It has been a fixture of GitHub trending (trendshift badge on the repo), and the promise is simple: the studio ships as a desktop app for macOS, Windows and Linux — plus Docker images and preview mobile apps — and every model runs locally. Your recordings never leave the machine.

Key capabilities

Feature	What it does
Voice cloning	Turn a clean reference recording into speech in a familiar voice.
Voice design	Describe a voice in words and generate it.
Video dubbing	Translate and dub videos with speech timed to the original.
Dictate & transcribe	A floating dictation widget plus local transcription.
Narrate & batch	Stories, audiobooks and queued jobs.
Connect & automate	Local API, MCP server, reusable recipes and watch folders.

The default engine is OmniVoice (from k2-fsa), and the built-in Model Manager lets you pick from dozens of other engines — you download only the models you need, and the app asks before downloading anything.

How to install

macOS / Linux

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

Windows (PowerShell)

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

Or grab an installer from the latest release: .dmg (Apple Silicon), .exe (Windows 10/11 x64), .AppImage / .deb / Snap / Nix (Linux), Docker images on GHCR, and preview .apk / .ipa mobile builds that run fully offline.

On macOS, "VoiceStudio is damaged and can't be opened" means Gatekeeper blocked the unsigned app — the project documents the fix in its install guide.

With a coding agent (Claude Code, Codex, Cursor, …), paste this into your agent session:

Install the VoiceStudio app on this device and verify it works, following
https://github.com/debpalash/VoiceStudio/blob/main/docs/install/agent.md

Agent skills: npx skills add debpalash/VoiceStudio installs two skills — voicestudio for audio work and voicestudio-maintainer for working on the repo itself.

System requirements

• No GPU is required — the app is fully usable on CPU, just slower. NVIDIA CUDA (and opt-in AMD ROCm on Windows) accelerate it where available; native GGUF engines can also use Vulkan on AMD, Intel and NVIDIA GPUs.
• macOS builds target Apple Silicon (Metal); Intel Macs get a limited build that connects to a remote backend.
• Android 10+ and iOS 17+ builds are preview and run on-device, offline.
• Models are downloaded on demand through the Model Manager; each model's license is shown there.

Voice work with DeepSeek Harness

VoiceStudio is not a DeepSeek Harness plugin — it is a standalone app — but it hooks into agent ecosystems through its local API and MCP server, and DSH conversations can drive it the same way other agent harnesses do. If voice output is what you want inside DSH itself, these are the native options:

• Voice & Speech plugins — text-to-speech and voice plugins built for DSH profiles.
• MCP & protocol plugins — MCP servers you can mount into a DSH profile.
• Agents, Automation & Workflows — scheduled narration and batch jobs driven by DSH agents.

FAQ

Is VoiceStudio really free? Yes — the app is AGPL-3.0 (including commercial use under its terms) and there is no paid cloud behind it. Model weights, voices and generated output carry their own licenses, which the Model Manager shows per model.

Does it work offline? Yes. Everything runs locally — cloning, dubbing, transcription — and the mobile builds run fully offline on-device. The only network use is downloading models, and the app asks first.

How is it different from ElevenLabs? ElevenLabs is a cloud service with usage pricing; VoiceStudio runs the same class of tasks (clone, design, dub, transcribe) on your own hardware with no per-character billing — you trade some convenience for privacy and zero marginal cost.

Can I clone any voice? Only with the owner's permission — the README says so explicitly. Cloning someone's voice without consent is misuse regardless of the tool.

Is there a Chinese/Japanese interface or docs? The README ships 简体中文 and 日本語 versions, and the app supports 646 languages for speech work.

Linux supported? Yes — .AppImage, .deb, Snap and Nix, with NVIDIA CUDA or CPU.