VoiceStudio is an open-source desktop (and preview mobile) speech studio that runs models on your own machine. The project README frames it as a fully local ElevenLabs-style alternative: voice cloning from a clean reference clip, text-prompted voice design, timed video dubbing, dictation, transcription, and audiobook-style narration. It lists support for 646 languages in the README product copy. The canonical voicestudio github repository is debpalash/VoiceStudio.
The default generation path is the built-in VoiceStudio engine powered by k2-fsa/OmniVoice. The Model Manager can install other engines (CosyVoice, IndexTTS, whisper-family ASR backends, and more) listed in the project's feature catalog.
What VoiceStudio is
According to the repository description and README:
- Local-first Electron app with a Python backend for speech models
- Cloning, design, dubbing, dictate/transcribe, narrate/batch jobs
- Optional local API, MCP hooks, recipes, and watch folders
- Homepage voicestudio.sh; source github.com/debpalash/VoiceStudio
- Application license AGPL-3.0; model weights and generated audio keep their own terms
Creator credit in the README team section goes to Palash Debnath (debpalash), with Shimul Debnath (velixio) listed on the core team in the v0.5.7 notes.
Feature and engine scope from the docs
The feature catalog groups product surfaces (cloning, design, dubbing, dictation widget, vocal isolation, diarization, batch queue, MCP, watermarking, local-first operation) separately from concrete engines. Speech generation defaults to OmniVoice; transcription defaults to WhisperX, with Faster-Whisper, MLX Whisper, PyTorch Whisper, Parakeet variants, and others available when installed.
Engine availability depends on which models you download, your GPU/CPU path, and any third-party license gates (for example gated Hugging Face repos). The catalog does not publish a single quality ranking across engines.
→ Project ranking: debpalash/VoiceStudio on AI Rank — daily GitHub star growth
Desktop install paths
Official one-liners from the README and docs/install/script.md:
macOS / Linux
curl -fsSL https://voicestudio.sh/install | sh
Windows (PowerShell)
irm https://voicestudio.sh/install.ps1 | iex
If voicestudio.sh is unreachable, the docs say to run the same script from the repository:
curl -fsSL https://raw.githubusercontent.com/debpalash/VoiceStudio/main/scripts/install.sh | sh
The shell installer targets macOS (Apple Silicon and Intel UI packages) and Linux x64. On Windows, the PowerShell script or the VoiceStudio-…-win-x64.exe / …-win-arm64.exe assets from Releases are the documented paths. Release packages are checksummed against SHA256SUMS.txt. The app asks before downloading models.
From source (development), the README shows git clone then bun run dev, which installs JavaScript dependencies and, for supported local backends, Python dependencies.
Release packages for v0.5.7
GitHub release v0.5.7 (published 2026-10-26, stable) ships Electron installers including:
| Platform (README status) | Example asset name |
|---|---|
| macOS Apple Silicon — Stable | VoiceStudio-0.5.7-mac-arm64.dmg |
| macOS Intel — Limited (UI / remote backend) | VoiceStudio-0.5.7-mac-x64.dmg |
| Windows 10/11 x64 — Stable | VoiceStudio-0.5.7-win-x64.exe |
| Windows on ARM — Experimental | VoiceStudio-0.5.7-win-arm64.exe |
| Linux x64 — Stable | .AppImage and .deb |
| Android 10+ — Preview | .apk / .aab |
| iPhone/iPad iOS 17+ — Preview | unsigned .ipa (sideload) |
Docker images are documented under GHCR (ghcr.io/debpalash/voicestudio). macOS docs require macOS 13.3+ and recommend about 10 GB free disk for the app, Python environment, and weights. Windows docs use the same ~10 GB guidance; machines without NVIDIA get a smaller CPU PyTorch install (docs: about 5 GiB free versus about 9 GiB for the CUDA path).
Install pitfalls documented upstream
Upstream install and troubleshooting docs call out these recurring failures.
- macOS Gatekeeper “damaged” dialog. Docs say the wording is misleading: quarantine plus missing Apple notarization/Developer ID signing can trigger it. Clearing
com.apple.quarantine(or using the shell installer, which clears it) is the documented fix. Notarization gaps are tracked in upstream issue discussions (#1779). - Intel Mac local backend. Packaged and source setup stop before creating a local Python environment because PyTorch no longer ships Intel-Mac wheels (#889). The UI can still talk to a remote backend.
- Wrong Mac DMG architecture. Opening the x64 DMG on Apple Silicon can lead setup into Intel PyTorch paths; install the
mac-arm64build instead. - Gated Hugging Face models. Speaker diarization via
pyannote/speaker-diarization-3.1needs an HF token and accepting the model license on the Hub page; a token alone is not enough. - Linux blank window / GPU compositor issues. Troubleshooting suggests launching once with
--disable-gpu(and optionally--ozone-platform=x11on Wayland). - Windows Smart App Control / application control. Unsigned runtime DLLs inside the Python environment can be blocked even when the installer itself is trusted; docs outline turning SAC off, using WSL2/Docker, or a remote backend—not inventing per-DLL workarounds here.
- First-run downloads. Large CUDA wheels and model shards can stall on restricted networks; docs cover retries, mirrors (
HF_ENDPOINT), andOMNIVOICE_TORCH_VARIANT=cpuwhen no NVIDIA GPU is present.
In-app Settings → About → Run self-check and the diagnostic bundle are the project's own first-response tools.
License notes and AI Rank placement
LICENSE-NOTICE.md states the application is AGPL-3.0 (including commercial use of the app under AGPL terms). Model weights, tokenizers, and generated output keep separate licenses shown in the Model Manager. Clone voices only with the owner's permission—that is the project's stated policy, not a legal opinion from AI Rank.
On the AI Rank open-source snapshot dated 2026-10-31 (last_crawled), debpalash/VoiceStudio sits at rank 5 (previous rank 21) with 94,982 stars, +1,402 stars over 1 day and +12,358 over 7 days, categorized under Applications / Content creation. Star growth measures attention on GitHub, not audio quality. A live GitHub API read during drafting showed a higher star total than the crawl; prefer the dated AI Rank numbers when citing the ranking page.
Keep exploring: see which open-source AI projects are rising today in the open-source rankings on AI Rank.
Or 📱 get the AI Rank app to follow them on your phone.
Cover image: GitHub Open Graph preview for debpalash/VoiceStudio; editorial composition by AI Rank.
Prepared with automated editorial assistance and checked against official sources and dated AI Rank data on October 30, 2026. We did not run our own benchmarks for this article.