VoiceStudio v0.5.7 runs voice cloning on your phone, 89k stars

A video editor who pays ElevenLabs for voice cloning now has a free local option. The new option runs on a phone. VoiceStudio released version 0.5.7 on Sunday. The project calls itself the open-source, fully-local ElevenLabs alternative. GitHub lists 89,123 stars and 9,834 forks on the repository.

What the release adds

Maintainer debpalash writes that v0.5.7 is the first VoiceStudio to run on phones. Preview iOS and Android apps run the models on the device itself. Once the models are downloaded, the apps work offline. The release also adds a small CPU PyTorch build. That build is for computers without a dedicated GPU. Windows on ARM gets an experimental installer in the same release.

A new core maintainer, Shimul Debnath, joined the project with this release. The notes say the project no longer waits on one person for reviews and triage.

The desktop app switched from Tauri to Electron. v0.5.3 remains the final Tauri build. The notes walk users through the migration. macOS installers for Apple Silicon and Intel are signed and notarised. Windows x64 and ARM64 installers are not code-signed yet. SmartScreen shows a warning on first run.

What the tool does

VoiceStudio covers voice cloning, voice design, video dubbing, dictation, transcription and audiobook creation. The default engine is OmniVoice, from k2-fsa. The description says the models reach 646 languages. The app runs on CUDA for NVIDIA GPUs, and Metal for Apple Silicon. AMD Radeon uses ROCm or Vulkan. Everything else runs on plain CPU.

The release notes mark two moves. Remote compute, remote workers and GPU sharing left the open-source app for VoiceStudio Pro. Projects, voices and settings are kept. Models without a plainly commercial licence now ask the user to accept their licence once. That includes the OmniVoice default. Until then, API and MCP requests that use those models return a licence error.

The licence to read before using it

The app itself is AGPL-3.0. The release notes say model weights, tokenisers, voices and generated output each come under their own licence. A paid app licence does not cover those. A team planning to use VoiceStudio for client work has to read the app licence and the model licence.

For a creator who was paying for voice cloning by the minute, the question is simpler. A test run costs the download, the disk and the time. The repository has 89,123 stars. That number makes it one of the busiest open-source audio projects of the year. The repository was created in February 2026.

Source

VoiceStudio v0.5.7 release notes on GitHub.