Omnivoice_Playground
A web-based audio workstation for AI-generated speech. Type a voice description or upload a voice sample, and it produces spoken audio using the OmniVoice model, all wrapped in a dark, studio-style interface with drag-and-drop timeline editing.
Studio Noir - OmniVoice Playground is a web interface for experimenting with AI-generated speech. It is built on top of a text-to-speech model called OmniVoice and lets you produce spoken audio from text in two different ways: by describing the voice you want in plain English (for example, "female, young adult, high pitch, excited"), or by uploading a short audio clip of someone speaking so the system can copy that voice's characteristics.
The interface is designed to look like a professional audio workstation. It shows a timeline with waveform grids where you can arrange and edit audio clips, adjust playback speed, and manage tracks with a drag-and-drop layout. The visual style is dark and cinematic, which the author calls Studio Noir. The front end is built with React and TypeScript, while a Python server running in the background handles the actual speech generation.
Speech synthesis runs significantly faster if your computer has an NVIDIA graphics card, because the underlying model uses the GPU for computation. The README includes troubleshooting steps for getting the correct software versions installed when GPU acceleration is not activating properly. If you have no compatible GPU, the tool falls back to running on the CPU, which is slower but still functional.
For people who just want to explore the interface without downloading the AI model, there is a Mock Mode. In this mode the application runs immediately with no model download required, so developers can test the visual layout and controls without waiting for a large file.
Setup on Windows is handled by a launcher script called run.bat. Double-clicking it checks for the right Python version, creates an isolated environment, installs all dependencies including the GPU-optimized packages if applicable, compiles the frontend, and opens the app in a browser automatically. A similar script is available for Linux and macOS.
Where it fits
- Generate realistic spoken audio from a text description of a voice style, like 'young female, excited tone'.
- Clone a voice by uploading a short audio clip and producing new speech that matches it.
- Arrange and edit AI-generated audio clips on a visual timeline, like a lightweight audio workstation.
- Test and explore the studio UI instantly using Mock Mode, with no large model download required.