harrrshall/syntts

syntts is a text-to-speech tool that runs entirely inside a web browser, with no server or backend involved. It converts typed text into spoken audio and is built specifically for Hinglish, the everyday style of mixing Hindi and English words that many speakers use naturally across India. The model runs using WebGPU, a newer browser technology that lets web pages use the computer's graphics card for computation. Because everything runs on your device, nothing you type is sent to any server. Your text and the resulting audio stay entirely on your machine. The first time you visit the page, the browser downloads about 396 megabytes of model weights from Hugging Face, a platform where researchers publish AI models. After that initial download, the files are cached and the tool loads immediately on future visits. You can type in either romanized Hinglish, writing Hindi words with English letters such as "yaar kal ka match dekha", or directly in the Devanagari script used for written Hindi. The tool converts romanized Hindi input to Devanagari automatically while keeping English words in the Latin alphabet, which shapes how words are pronounced by the model. Four fixed speaker voices are available. The model has 89 million parameters. According to the README, the browser version produces audio that exactly matches the reference PyTorch model it was derived from. A live demo runs at the project's GitHub Pages address. No installation is needed: you visit the page in a supported browser. The README is sparse and does not state a license.