Below is a new article rewritten from the provided content, now structured with an FAQ section and a conclusion section.
—
## OmniVoice Studio: The Open-Source, Local Voice AI Workstation
You paste a paragraph of text into ElevenLabs, press Generate, and watch the character counter tick down. The free tier is gone before you finish testing. The Creator plan is $22 a month. The Pro plan is $99. And every audio file you generate leaves your machine and ends up on their servers—which matters the moment your content is sensitive, proprietary, or simply yours.
OmniVoice Studio is built on a different premise: everything runs on your hardware. Voice cloning, video dubbing, real-time dictation, and voice design are all local, all free for personal use, and require no API key or usage counter. The project describes itself as “the open-source ElevenLabs alternative,” and that’s accurate—though its language coverage alone makes the comparison interesting: ElevenLabs supports 32 languages; OmniVoice Studio supports 646.
With 7.1k GitHub stars and 1.1k forks, the latest release (v0.2.7, May 3, 2026) includes pre-built installers for macOS, Windows, and Linux. This article walks you through installing and running OmniVoice Studio end to end.
### What OmniVoice Studio Is and Why It Was Built
OmniVoice Studio gives you a professional voice AI desktop app that never phones home. No accounts, no subscriptions, no cloud calls during inference. Your reference audio, scripts, and generated files stay on your machine.
Here is how its feature set and pricing compare directly to ElevenLabs:
| Feature | ElevenLabs | OmniVoice Studio |
|———————|——————————–|——————————|
| Pricing | $5–$330/month, per-character | Free for personal use |
| Voice Cloning | 3-second clip | 3-second clip, zero-shot |
| Voice Design | Gender, age | Gender, age, accent, pitch, style, dialect |
| Languages | 32 | 646 |
| Video Dubbing | Cloud-only | Fully local |
| Data Privacy | Audio sent to the cloud | Nothing leaves your machine |
| API Keys | Required | Not needed |
| GPU Support | N/A (cloud) | CUDA, Apple Silicon MPS, AMD ROCm, CPU |
| Desktop App | No | macOS, Windows, Linux |
Under the hood, OmniVoice Studio is a Tauri desktop application—Rust on the outside, React on the inside, and FastAPI as the backend with 97 API endpoints. Persistent state lives in SQLite. The AI pipeline relies on four open-source components:
1. WhisperX for transcription, word-level speech recognition, and alignment
2. Demucs (Meta) for vocal isolation, separating speech from music and background noise
3. OmniVoice from k2-fsa, a zero-shot diffusion text-to-speech engine that enables cloning from a 3-second clip across 646 languages
4. Pyannote for speaker diarization, identifying who spoke when in multi-speaker recordings
GPU acceleration is auto-detected at launch. If you have under 8 GB VRAM, the TTS model offloads to the CPU automatically. The pipeline still runs, just slower. Apple Silicon Macs are the sweet spot for GPU-less users: the app automatically uses MLX-optimized Whisper and TTS backends that leverage the Apple Neural Engine and Metal Performance Shaders, roughly doubling throughput versus the CPU path.
### System Requirements
Before installing, check that your machine meets the minimum specs. The app will run below these, but you will notice the difference.
| Component | Minimum | Recommended |
|———|———|————-|
| OS | Windows 10 (21H2+), macOS 12+, Ubuntu 20.04+ | Any modern 64-bit OS |
| RAM | 8 GB | 16 GB+ |
| VRAM | 4 GB (TTS auto-offloads to CPU if less) | 8 GB+ (NVIDIA RTX 3060+) |
| Disk | 10 GB free (models + cache) | 20 GB+ SSD |
| Python | 3.10+ (managed by uv) | 3.11–3.12 |
| GPU | Optional (CPU works) | NVIDIA CUDA, Apple Silicon MPS, AMD ROCm |
You do not need a GPU to use OmniVoice Studio. On CPU, TTS synthesis is roughly 3× slower, and transcription of long videos takes longer, but short voice clones and dictation remain perfectly usable.
### Installing OmniVoice Studio
#### macOS
**Prerequisites:**
– macOS 12 (Monterey) or newer (Apple Silicon or Intel)
– Python 3.11+
– Bun
– Xcode Command Line Tools
– FFmpeg
Install them in order:
“`bash
# Install Python via Homebrew (or use pyenv)
brew install python@3.11
# Install Bun
curl -fsSL https://bun.sh/install | bash
# Install Xcode Command Line Tools
xcode-select –install
# Install FFmpeg
brew install ffmpeg
“`
Then clone and run:
“`bash
git clone https://github.com/omnivoice-ai/omnivoice-studio.git
cd OmniVoice-Studio
bun install
bun run desktop-prod
“`
The first launch is slower—it builds the Tauri shell, creates a Python virtual environment via uv, syncs all Python dependencies, and downloads model weights (~2.4 GB). The splash screen shows live progress for each step.
**Pre-built DMG users:** Download the latest DMG from the Releases page, mount it, and drag OmniVoice Studio into /Applications. If Gatekeeper blocks it, run:
“`bash
xattr -cr “/Applications/OmniVoice Studio.app”
“`
#### Windows
**Prerequisites:**
– Windows 10 (21H2+) or Windows 11, x64
– Python 3.11+
– Microsoft C++ Build Tools
– Bun
– FFmpeg
Install from a regular (non-admin) PowerShell:
“`powershell
winget install Python.Python.3.11
# Download and run “Desktop development with C++” workload
winget install Gyan.FFmpeg
irm bun.sh/install.ps1 | iex
“`
Then clone and run in PowerShell:
“`powershell
git clone https://github.com/omnivoice-ai/omnivoice-studio.git
cd OmniVoice-Studio
bun install
bun run desktop-prod
“`
**Note (Windows-specific):** On machines with under 16 GB VRAM, certain TTS engines can OOM during torch.compile kernel compilation. If you see `OutOfMemoryError: CUDA out of memory`, open Settings → Performance and toggle “Disable torch.compile (Windows)” on, or set:
“`powershell
$env:TORCH_COMPILE_DISABLE = “1”
bun run desktop-prod
“`
#### Linux (Debian/Ubuntu)
**Prerequisites:**
“`bash
sudo apt install python3.11 ffmpeg
curl -fsSL https://bun.sh/install | bash
sudo apt install libwebkit2gtk-4.1-dev libayatana-appindicator3-dev librsvg2-dev libssl-dev libxdo-dev build-essential
“`
**Clone and run:**
“`bash
git clone https://github.com/omnivoice-ai/omnivoice-studio.git
cd OmniVoice-Studio
bun install
bun run desktop-prod
“`
**AppImage users:** Make it executable and run:
“`bash
chmod +x OmniVoice.Studio_*.AppImage
./OmniVoice.Studio_*.AppImage
# If you see a white screen on Ubuntu 24.04+:
WEBKIT_DISABLE_COMPOSITING_MODE=1 ./OmniVoice.Studio_*.AppImage
“`
**Pre-built .deb:**
“`bash
sudo apt install ./OmniVoice.Studio_*.amd64.deb
omnivoice-studio
“`
#### Docker (backend only)
“`bash
git clone https://github.com/omnivoice-ai/omnivoice-studio.git
cd OmniVoice-Studio
docker compose -f deploy/docker-compose.yml up
“`
The backend API is then available at http://localhost:8000.
### Setting Up Your Hugging Face Token (Optional but Recommended)
This is required for speaker diarization and larger voice-design engines.
1. Create a free Hugging Face account and generate a read token.
2. Either:
– Use the app: Settings → API Keys → paste your `hf_…` token, or
– Set an environment variable:
– macOS/Linux: add `export HF_TOKEN=hf_your_token_here` to ~/.zshrc or ~/.bashrc, then source it.
– Windows PowerShell: `[Environment]::SetEnvironmentVariable(“HF_TOKEN”, “hf_your_token_here”, “User”)`
3. Accept the model terms on the Hugging Face page for pyannote/speaker-diarization-3.1 (one-time).
### Cloning a Voice
Voice cloning is the core feature. OmniVoice from k2-fsa is a zero-shot diffusion TTS model trained on 646 languages—no fine-tuning required. You provide a 3–10 second reference clip and synthesize in the chosen language.
**How to clone:**
1. Go to the Voice Clone tab.
2. Record in-app or upload a clean, quiet 3–10 second speech clip.
3. Select the target language (646 available).
4. Type or paste the text.
5. Click Generate.
For best results, use a quiet, natural reference clip without background music. Noisy phone recordings reduce quality; run noisy audio through the Vocal Isolation tab first if needed.
### Dubbing a Video
The dubbing pipeline transcribes a video, translates speech, clones speaker voices, synthesizes dubbed audio, and muxes everything back into an MP4—all locally.
**How to dub:**
1. In the Dub tab, paste a YouTube URL or upload a local file.
2. Choose the target language.
3. Click Start Dub.
4. Monitor progress: download → transcription → diarization → translation → synthesis → mux.
For a 5-minute video with GPU, expect 8–12 minutes; CPU-only takes longer. Outputs include the dubbed MP4, SRT subtitles, isolated stems, and the original transcription.
Use “Add to Queue” for batch jobs. The conductor processes them sequentially with live progress per job.
### Designing a Voice
In the Voice Design tab, adjust sliders for gender, age, accent, pitch, speed, emotion, and dialect. Press Preview to listen, and use A/B Comparison to toggle between configurations.
Save a design to your Voice Gallery to use it later in Voice Clone without a reference clip.
### Using the Dictation Widget
The dictation widget provides system-wide, real-time transcription. The global hotkey is:
– macOS: Cmd+Shift+Space
– Windows/Linux: Ctrl+Shift+Space
Press the hotkey anywhere, speak, and the transcription auto-pastes into the focused app. Configure or change the hotkey in Settings → Dictation. Note: each activation transcribes and discards the audio—use the Transcription tab for editable, session-long transcripts.
### Choosing a TTS Engine
OmniVoice Studio ships six TTS engines, configurable in Settings → TTS Engine or via the OMNIVOICE_TTS_BACKEND environment variable:
| Engine | Languages | Clone | Best For |
|——————|———–|——-|—————————————-|
| OmniVoice (default) | 600+ | Yes | General-purpose |
| CosyVoice 3 | 9 + 18 dialects | Yes | Instructed generation with style control |
| MLX-Audio | Multi | Varies| Apple Silicon, fastest on M-series |
| VoxCPM2 | 30 | Yes | Cross-platform cloning, strong accents |
| MOSS-TTS-Nano | 20 | Yes | Fast cloning on lower-powered machines |
| KittenTTS | English only | No | Lightweight CPU-only English TTS |
For English-only, CPU-only workflows, KittenTTS and MOSS-TTS-Nano offer near real-time synthesis. On Apple Silicon, MLX-Audio maximizes speed using the Apple Neural Engine.
You can plug in a custom TTS backend by subclassing TTSBackend in backend/services/tts_backend.py and registering it in _REGISTRY.
### Using OmniVoice Studio via the MCP Server
OmniVoice Studio includes an MCP server, enabling integration with Claude Desktop, Cursor, and other MCP-compatible clients.
The server runs on localhost:8765 when OmniVoice Studio is running. Add it to Claude Desktop:
“`json
{
“mcpServers”: {
“omnivoice”: {
“command”: “npx”,
“args”: [“-y”, “@omnivoice/mcp-server”],
“env”: {
“OMNIVOICE_API_URL”: “http://localhost:8765”
}
}
}
}
“`
Once connected, Claude can call OmniVoice tools to generate audio, clone voices, and start dubbing jobs locally.
### FAQ
**Do I need an API key or account?**
No. OmniVoice Studio runs entirely locally and does not require authentication, API keys, or recurring subscriptions.
**Can I use it without a GPU?**
Yes. The entire pipeline runs on CPU, though TTS synthesis is slower. Apple Silicon Macs get significant acceleration via the Apple Neural Engine and MLX backends even without a discrete GPU.
**How much disk space is needed?**
Plan for about 10 GB free for models and cache, or 20 GB+ SSD for best performance.
**Are pre-trained models downloaded automatically?**
Yes. During first launch, the app downloads the required model weights (approximately 2.4 GB) and Python dependencies via uv.
**Is my data private?**
Yes. Nothing leaves your machine. All processing—transcription, voice cloning, dubbing, and synthesis—is local.
**What if I encounter issues on Windows?**
Ensure Microsoft C++ Build Tools are installed and the “Disable torch.compile” setting is enabled if you encounter CUDA OOM errors on machines with under 16 GB VRAM.
**How do I switch TTS engines?**
Use Settings → TTS Engine in the app, or set the OMNIVOICE_TTS_BACKEND environment variable (e.g., export OMNIVOICE_TTS_BACKEND=cosyvoice on macOS/Linux).
### Conclusion
OmniVoice Studio makes a practical case for local-first voice AI. Not because cloud tools are bad, but because 646 languages, no usage meter, and audio that never leaves your machine add up to something genuinely different.
The setup—installing the app, downloading a 2.4 GB model, and optionally adding a Hugging Face token—is a one-time investment. After that, it’s simply using the tool. It’s in active beta, and while some edges are rough, the core pipeline—cloning, dubbing, dictation, and voice design—works reliably. The community is responsive, and releases have been consistent since launch.
For developers, content creators, and researchers who work with audio and care about where their data goes, OmniVoice Studio is worth having locally.
—
