thanhng8/omnivoice-tool

OmniVoice TTS Tool is an add-on that wraps an existing open-source text-to-speech engine called OmniVoice (made by another team, k2-fsa) and turns it into something easier to use on your own computer. Everything in this repository lives in a folder called tool/ and is not part of the original upstream project, so it is best to think of it as a friendly shell around someone else's voice generation model. The core piece is a local server, written in Python, that loads OmniVoice once when it starts and then exposes two things on a single port (8765): a normal web page you can open in your browser, and a WebSocket connection that streams generated audio. There are launcher scripts for Windows and for macOS or Linux that pop up a small numbered menu, letting you pick between running on a GPU with speech recognition, GPU only, CPU, a custom port, or advanced settings. After the first run, an offline flag keeps it from checking the network, so it starts in three to five seconds. The browser interface follows a three-step flow: pick a voice, write the text, then generate. You can choose Auto Voice (a random voice each time), Voice Clone (pick a saved voice or upload three to ten seconds of your own reference audio), or Voice Design (describe gender, age, pitch, accent, and dialect). The author says 646 languages are supported, with a prebuilt gallery of 45 curated voices covering English, Chinese, and Vietnamese. Every generation knob the underlying model offers is exposed as a slider or input, and you can insert non-verbal markers like laughter or sighs. Bulk import from spreadsheets and text files is supported, and output comes out as per-line wav files plus a zip bundle. Example client code is provided for browser, Chrome extension, Node.js, and Python.