VoiceStudio Hit 22,000 Stars in Five Months and One Person Wrote 82% of the Code
VoiceStudio bundles 16 TTS engines and 11 ASR engines into a local-first desktop app. It gained 7,300 stars in a single week. One maintainer accounts for 1,981 of 2,413 commits.
The WJS Desk
Sep 10, 2026 · 4 min read
The Number That Made Us Look
VoiceStudio (formerly OmniVoice-Studio) crossed 22,096 stars on GitHub as of September 10, up from roughly 14,800 a week earlier. That is 7,300 stars in seven days, according to star-history.com's tracking. The repository was created on April 9, 2026, which means it went from zero to 22,000 in five months.
The pitch: an open-source, fully local ElevenLabs alternative that handles voice cloning, voice design, video dubbing, dictation, transcription, and audiobook creation across 646 languages. No account, no API key, no subscription for local workflows. You download it, install model weights, and everything runs on your machine.
What It Actually Is
VoiceStudio is a Tauri v2 desktop app with a React + Vite frontend and a FastAPI backend. It runs a local server on port 3900 and exposes an OpenAI-compatible API, meaning any tool that speaks /v1/audio/speech or /v1/audio/transcriptions can use it as a drop-in replacement for a hosted service.
The scope is ambitious. It integrates 16 text-to-speech engines, from OmniVoice (the default, covering 600+ languages) to CosyVoice 3, GPT-SoVITS, and MLX-Audio for Apple Silicon. On the speech-to-text side, it bundles 11 ASR engines including WhisperX, Faster-Whisper, Parakeet TDT, and Moonshine for low-power devices.
# Quick start with Docker
docker run -d -p 127.0.0.1:3900:3900 \
-v omnivoice-data:/app/omnivoice_data \
--name voicestudio palashdeb/omnivoice-studio:stable
# Then use the OpenAI-compatible API
curl http://localhost:3900/v1/audio/speech \
-H "Content-Type: application/json" \
-d '{"model": "omnivoice", "input": "Hello world", "voice": "default"}' \
--output speech.mp3What We Could Not Test, and What the README Claims
We did not install VoiceStudio for this piece. The full setup requires downloading model weights that range from 2 GB to 15 GB depending on which engines you enable, and evaluating voice quality across 16 engines fairly would take more time than a spotlight allows. Everything below comes from the repository's documentation and the project's own comparison table.
According to the README, VoiceStudio's default engine (OmniVoice) supports zero-shot voice cloning from short reference clips. The project claims feature parity with hosted services on core tasks: text-to-speech, voice cloning, and transcription. It adds features that ElevenLabs does not offer at its standard tier, including video dubbing with speaker preservation, system-wide dictation, and batch processing with folder monitoring.
| Feature | VoiceStudio | ElevenLabs (Pro, $99/mo) |
|---|---|---|
| Voice cloning | Local, no upload | Cloud, audio uploaded |
| Languages | 646 (claimed) | 32 |
| Video dubbing | Built in | Separate product |
| Latency | Hardware-dependent | Optimized cloud inference |
| Setup time | Install app + download models (2-15 GB) | Create account, start |
| Ongoing cost | $0 (you provide hardware) | $99/month |
| Voice quality | We did not compare | Industry benchmark |
We left voice quality blank on purpose. The README does not publish listening tests or MOS scores, and we did not run our own. Claiming parity without measurement would be dishonest. If you are evaluating VoiceStudio for production, run a blind comparison on your own content first.
What It Does Not Do
- No streaming TTS in the browser. The API returns completed audio files. Real-time voice for live applications (game characters, phone bots) is not supported out of the box.
- No hosted option. If you want voice synthesis without managing hardware, VoiceStudio is not the answer. That is by design, but it limits the audience.
- AGPL license. This is the biggest hidden cost. AGPL requires that any network service built on VoiceStudio must release its source code. If you are building a commercial product that uses VoiceStudio as a backend, you either open-source your product or negotiate a commercial license. The README lists a contact email for proprietary embedding.
- Model license fragmentation. The 16 TTS engines carry different licenses: Apache-2.0, MIT, CC-BY-NC, and others. The application is AGPL, but a model licensed CC-BY-NC (like parts of OmniVoice) restricts commercial use of the generated audio. You need to check each engine's license before shipping.
Who Made It, and That Matters Here
Palash Debnath (debpalash on GitHub) has authored 1,981 of the repository's 2,413 commits, or 82%. The next contributor, psiberfunk, has 42. Nine other contributors account for the rest, most under 40 commits each.
That is a bus factor of one on a project with 22,000 stars, 16 TTS engines, 11 ASR engines, a desktop shell, a web UI, Docker images, and an MCP server. The code is substantial: 2,413 commits in five months is a pace of roughly 16 commits per day. Whether that pace is sustainable for one person is the question the star count does not answer.
The repository has only 8 open issues, which is unusually low for a project this size. Either the maintainer is extremely responsive, or most users are starring without filing bugs, or both.
Verdict
VoiceStudio is the most complete local-first voice toolkit we have seen on GitHub. If you need voice synthesis or transcription and your constraint is privacy (medical, legal, enterprise data that cannot leave the building), this is the project to evaluate. The OpenAI-compatible API makes integration straightforward for teams already using hosted voice services.
If your constraint is quality and you are building a consumer product, hold off until someone publishes independent listening tests. The feature list is impressive, but voice quality is the metric that matters for end users, and nobody outside the project has measured it yet.
If you adopt it, audit the model licenses before shipping anything commercial. The AGPL application license and the CC-BY-NC model licenses are two different restrictions, and mixing them up could be expensive.
22,000 stars in five months says the demand for local voice AI is real. Whether one maintainer can carry a project this ambitious is the bet you are making by depending on it.
