spiral@out:~/blog$ cat transmissions/voicestudio.md

VoiceStudio: An ElevenLabs You Run Yourself

[2026-10-19] ::self-hosting, audio, tts, voice, local-first

VoiceStudio calls itself "the open-source, fully-local ElevenLabs alternative." It's a desktop app for voice cloning, voice design, video dubbing, dictation, transcription and audiobook production, and it claims support for 646 languages. The pitch is that all of it runs on your own hardware.

It's one of the biggest repos on GitHub Trending right now, with roughly 76,000 stars. The application is licensed under AGPL-3.0, with a big caveat about model weights that's covered below.

What it does

The README groups everything into three columns:

  • Create: clone a voice from a reference recording, or design a new one from a description. A floating dictation widget handles speech-to-text anywhere on your desktop.
  • Produce: dub videos with timed speech, and build stories, audiobooks and batch jobs.
  • Connect: a local API and an MCP server so AI agents can drive it, plus reusable "recipes" and watch folders for automation.

The feature catalog also lists vocal isolation, speaker diarization (working out who spoke when), a batch queue, GPU auto-detection and an AI watermark option for generated audio.

How it works

VoiceStudio isn't one model. It's a desktop shell around a large set of open-source speech engines that you install and switch between in a model manager.

Speech generation defaults to its own engine, which is powered by k2-fsa's OmniVoice. The catalog lists over a dozen alternatives, including CosyVoice 3, IndexTTS 2.5, KittenTTS, MLX-Audio, VoxCPM2, GPT-SoVITS, sherpa-onnx and MOSS-TTS, several with their own setup guides.

Transcription defaults to WhisperX, with Faster-Whisper, MLX Whisper, NVIDIA's Parakeet, Moonshine and FunASR as options. Live dictation uses sherpa-onnx. You can also point it at any OpenAI-compatible server if you'd rather not run speech-to-text locally.

The front end is an Electron app with a Python backend. (Version 0.5.3 was the last release built on Tauri. Electron is now the only desktop app, so Tauri users have to install it separately.) The project says local workflows run entirely on your machine, remote services are optional, and usage analytics only run if you consent.

Hardware

The README is straightforward about what runs where:

  • NVIDIA GPU on Windows or Linux gets CUDA acceleration.
  • Apple Silicon gets Metal acceleration.
  • No dedicated GPU: it runs on the CPU, more slowly, using a small CPU build of PyTorch that needs about 5 GB of free disk.
  • Windows on ARM is experimental and CPU-only, with the Python backend running under x64 emulation.
  • Intel Macs get the UI only and have to connect to a backend running on another machine.

Requirements vary a lot by engine. The project keeps separate performance and benchmark docs.

Trying it

On macOS or Linux it's a one-line install:

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

On Windows, in PowerShell:

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

The installer keeps your settings, projects and models across upgrades, and --uninstall removes the app while leaving your data. If you'd rather not pipe a script into your shell, there are .dmg, .exe, .AppImage and .deb downloads on the releases page, plus a Docker guide. To run from source:

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

Once it's running, open Voice cloning, pick a voice or add a clean reference recording, type your text and generate. The app prompts you to install the model it needs.

The fine print

  • The licence covers the app, not the models. AGPL-3.0 applies to VoiceStudio's own code. Model weights, tokenizers, voices and any conditions on generated output are separate, and the README says outright that even a paid app licence doesn't grant those rights. The model manager shows each model's source notices and flags gaps in the licence review, but the project says this doesn't certify commercial use. If you plan to sell anything you make with it, check each model's terms yourself.
  • Consent matters. "Clone voices only with permission," says the README. The optional AI watermark helps mark generated audio as synthetic.
  • It's a big, fast-moving app. A dozen-plus engines, an Electron shell, a Python backend, and a recent switch from Tauri make for a lot of moving parts. Expect some setup friction, especially on less common hardware.

Who it's for

VoiceStudio fits people who want ElevenLabs-style tools without sending their audio to someone else's servers: podcasters and video makers dubbing their own content, writers producing audiobooks, anyone who wants offline dictation, and tinkerers who want to compare speech models side by side. The local API and MCP server also make it a handy speech backend for automation and agent setups.

The GitHub repo and voicestudio.sh are the places to start, and it's in the resource index now.

End of transmission.

<- BACK TO LOG