VoiceStudio Passes 47,000 Stars on a Xiaomi-Built Voice Engine

VoiceStudio, an open-source desktop app for voice cloning, dubbing and dictation, had 47,152 GitHub stars and 5,317 forks on Sept. 29, less than six months after its repository was created on Apr. 9, according to GitHub's own record. GitHub's trending list showed it gaining 4,712 stars that day.

The repository describes it as "the open-source, fully-local ElevenLabs alternative" covering voice cloning, voice design, video dubbing, dictation, transcription and audiobooks in 646 languages. Stars measure attention, not use, and the README carries no comparison against ElevenLabs.

What sits underneath: OmniVoice from k2-fsa

The default engine is OmniVoice, published by k2-fsa under Apache-2.0 and described on its page as from Xiaomi's speech and language lab. Its README says it covers more than 600 languages and clones a voice from a reference clip, with a suggested length of 3 to 10 seconds. The speed figure it gives, a real-time factor of 0.025 or about 40 times faster than real time, is the developers' own and The Terminal did not run it.

The README also lists a limit the app's headline count hides: the voice design feature is "primarily trained on Chinese and English data only" and may be unstable in low-resource languages. It prohibits unauthorized voice cloning, impersonation and fraud.

Item	Value	Where it comes from
Stars, Sept. 29	47,152	GitHub repository record
Forks	5,317	GitHub repository record
Commits on main	3,471	Repository page
App licence	AGPL-3.0	Repository page
Default engine licence	Apache-2.0	OmniVoice repository

What version 0.5.6 added on Sept. 23

The release notes for v0.5.6 say the app can now answer Twilio phone calls in a saved voice, and that Claude Code, Cursor, Codex CLI and the OpenAI Agents SDK can connect to it with copyable setup. The notes say MCP, the protocol agents use to call outside tools, previously returned HTTP 405 in Docker. They also say long audiobook chapters now finish on 8 GB GPUs.

At 3,471 commits over 173 days, the project averages about 20 a day, a pace that suits a fast-moving product and makes any single build a moving target.

Licences and open problems to check first

The AGPL-3.0 applies to the app. The README says models "have their own licenses" and tells users to review them before commercial use, so a commercial dubbing service needs to read the licence for each engine it installs, not only the app's.

The issues page listed 12 open items dated Sept. 28 and 29. They include two backend crashes, a missing-ffmpeg failure and a report that reasoning text from local thinking models leaks into dub translations. GitHub's record counted 48 open issues and pull requests in total. The README says usage analytics require consent and that the app is Electron-only after Tauri was deprecated at v0.5.3.

The install line is curl -fsSL https://voicestudio.sh/install | sh. The Terminal has not read or run that script, and readers should do so before piping it to a shell.

Compare the pattern with the Paperclip agent orchestrator, which had ten security advisories and Hindsight's 41,000 stars, where popularity ran ahead of scrutiny. The next thing to watch is the 48-item queue, which held two backend crashes on Sept. 28 and 29.