VoiceStudio Tops GitHub Trending With a Licensing Gap
A.I. / news
VoiceStudio Tops GitHub Trending With a Licensing Gap
The open-source ElevenLabs alternative gained 30,722 stars in five months, but its default OmniVoice model is licensed non-commercial even though the app itself is not.
VoiceStudio, an open-source desktop app for cloning and generating voices, reached No. 1 on GitHub's daily trending list Tuesday, according to the analytics site Trendshift. The project has gathered 30,722 stars since developer Palash Debnath opened its repository April 9.
Debnath describes VoiceStudio as a "fully-local ElevenLabs alternative" in the project's GitHub repository, covering voice cloning, voice design, video dubbing, dictation, transcription and audiobook creation across a 646-language catalogue. Everything runs on the user's own machine: a Tauri desktop shell wraps a React frontend and a FastAPI backend, with no account, API key or subscription required for the local workflow.
Sixteen engines under one interface
The app bundles 16 text-to-speech engines, defaulting to one called OmniVoice, alongside options including CosyVoice 3, GPT-SoVITS and VoxCPM2. For transcription, it ships 11 speech-recognition engines, defaulting to WhisperX, with Faster-Whisper and MLX Whisper as alternatives. The same interface handles video dubbing, replacing a clip's spoken track with a cloned voice in another language, and long-form audiobook narration from a manuscript file. Debnath's repository description lists all six capabilities together: voice cloning, voice design, dubbing, dictation, transcription and audiobook creation.
| Requirement | Minimum | Recommended |
|---|---|---|
| RAM | 8GB | 16GB+ |
| Disk | 10GB free | 20GB+ SSD |
| GPU | Optional | NVIDIA CUDA or Apple Silicon |
The app runs CPU-only if needed, though Debnath's documentation recommends 4GB to 8GB of video memory for GPU acceleration. Intel-based Macs are shut out of the local backend entirely, a limitation the README attributes to PyTorch's wheel availability for that architecture rather than any deliberate choice.
A commercial gap inside an open license
VoiceStudio itself is licensed under AGPL-3.0, which the README says does not restrict selling generated audio. The default OmniVoice weights it downloads on first run are licensed separately, under CC-BY-NC, which the README says means "downloaded model and tokenizer terms may" block commercial use even though the app's own license does not. Debnath's team offers separate commercial licensing for embedding VoiceStudio in a paid product, a workaround for the app but not for OmniVoice's own restriction. It is the same tension that shaped m-a-p's YuE2-3B music model, whose weights also carry a non-commercial license despite Apache-licensed inference code.
VoiceStudio ships AudioSeal, an imperceptible watermark, active by default on generated audio so it can later be detected as synthetic. The README asks users to "only clone or synthesize voices with explicit permission from the speaker," a policy the software itself has no way to enforce.
How fast it climbed
Trendshift, which tracks GitHub's trending pages, recorded VoiceStudio as the No. 1 Python repository of the day on Sept. 1, the No. 1 Python repository of the week in the site's Week 36 ranking, and the No. 1 repository overall, across every language, on Tuesday. The repository lists 21 contributors and 3,685 forks, and its most recent commit landed the same day it topped the trending page, a pace of activity that has followed other suddenly popular open-source AI projects, including OpenClaw's jump to 388,000 stars.
Sources
More in A.I.
- 01How a Heap Overflow and an SSO Bug Reached OpenAI's MonorepoHacktron chained a libheif image bug through OpenAI's own forum to hijack an engineer's Codex session and open a pull request in the internal openai/openai repository.
- 02Agility's Digit 5 Drops the Safety Cage, Not the SkepticismThe humanoid robot lifts 50 pounds and charges in 9 minutes, backed by $300 million in orders. An independent robotics writer says its business case still assumes a drop-in worker replacement.
- 03PrismML Shrinks a 27B Model to 5.9GB at 1.72 BitsTernary Bonsai 2 27B keeps 98.2% of its full-precision score by rebuilding Qwen3.8-27B's weights as three values instead of sixteen bits, and an independent tracker puts the retention slightly lower.
- 04OpenAI Discloses a Model That Wrote Its Own JailbreakAn unreleased Astra-family model added a fabricated persona to 27 training summaries this summer, and the successor model mostly ignored what it had written.