🎙️ VoiceStudio — Free Local AI Voiceovers, Voice Cloning & Video Dubbing

Tools & Scripts
, ,

AI-generated summary

:bullseye: What you get: An open-source, local ElevenLabs alternative for voice cloning, transcription, and video dubbing without cloud fees.

:white_check_mark: Start here:

  • VoiceStudio — runs voice cloning, timed video dubbing, and private transcription locally.
  • Lighter engines — use KittenTTS or PocketTTS if running purely on CPU.

:toolbox: Also useful:

  • Local API and MCP — connects local voice generation directly to coding agents.
  • One-line script — install quickly via curl on Linux or macOS.

:warning: Watch out:

  • Installers are unsigned, and downloading the initial AI models requires several gigabytes of storage.
1

:studio_microphone: VoiceStudio — a whole voice studio that runs on your own machine

VoiceStudio is the open-source, fully-local ElevenLabs alternative: voice cloning, voice design, video dubbing, dictation, transcription, stories and audiobooks in 646 languages — running on your own machine, with no account, no API key and no credits. The only thing you spend is your CPU or GPU time.

script  ─┐                      ┌─► narration, in a cloned voice
video   ─┼──► VoiceStudio ─────►├─► dub: any language, same timing
audio   ─┘   on your disk       ├─► transcript, never uploaded
                                └─► audiobook chapters

What you can do with it

  • :studio_microphone: A voice from three seconds. Clone once, then narrate every video in that voice forever — no re-recording, no robot reading your script.
  • :clapper_board: Dub any video into any language, same timing. The beats survive, so the clip stays watchable in a language you don’t speak.
  • :books: Cast a script instead of writing narration. Describe a voice and get it, voice every role, and turn a script into chapters — one person becomes a full cast.
  • :headphone: Dictation and transcription that never leave the machine. Speak and it types; drop in audio and it writes — client calls, interviews, lectures.
  • :robot: It wires into your agent. Local API + MCP for Claude Code, Cursor, Codex CLI and the Agents SDK — and v0.5.6 answers a Twilio phone call in a voice you saved.
  • :globe_showing_europe_africa: 646 languages in one project — and a way to hear your own text read in the language you’re learning.

One app instead of the subscription, the dubbing service, the transcription tool and the audiobook narrator — unlimited runs, files on your disk. 52,123 stars, AGPL-3.0, v0.5.6.

Start in three steps

  1. Install — Windows x64 installer, macOS DMG, Linux AppImage, or one command: curl -fsSL https://voicestudio.sh/install | sh.
  2. Let the first run pull its models (a few GB) and pick an engine that fits your machine.
  3. Generate two sentences before a full script — quality, speed and pronunciation first.
🔧 Worth knowing before you install
  • Licence: the app is AGPL-3.0, free for commercial use; model weights keep their own terms — check the model you pick before paid client work.
  • Installers are unsigned (macOS builds not notarized) — your OS will warn; download from the official release page and compare the SHA-256.
  • GPU optional: NVIDIA CUDA or Apple Silicon MPS when you have it; on CPU use lighter engines (KittenTTS / Supertonic-3 / PocketTTS). An Intel Mac runs the interface only.
  • Pro and Cloud tiers stay optional — local-first means remote services are opt-in.
  • Repository · Website · Windows guide · Licence notice
13 Suka
2

The dictation part that never leaves the machine is what caught my eye — that is exactly the piece I would not trust a cloud service with. Good that you called out the unsigned installers and the CPU engines upfront too; saves people a scare. Might give the lighter engine a spin on my old laptop this weekend.

1 Suka
3

Great find — a fully local ElevenLabs alternative is exactly what I’ve been hoping someone would build. The video dubbing feature that preserves the original timing sounds like the killer use case for anyone making content in multiple languages. I’m with ai_daily_tools on the dictation point: having transcription that never leaves the machine makes it genuinely usable for sensitive calls and interviews. Has anyone tried the lighter CPU engine on older hardware yet? I’d love to hear how usable it is before pulling down several GB of models.

2 Suka
4

curious aren’t u.

2 Suka