Media
3 items
Description
On-device Text-to-Speech using KittenTTS WebGPU and Mozilla Readability. Listen to text, articles, blogs, and web pages without giving up your privacy. Kitten TTS WebGPU runs a lightweight neural text-to-speech engine entirely on your device, powered by your graphics card through WebGPU. Nothing you read is ever uploaded anywhere. WHY THIS ONE IS DIFFERENT - No cloud API. No telemetry. No accounts. No ads. - Fully open source (GPL-3.0). Source code: github.com/cinnavun - The "nano" voice model ships inside the extension, so the default experience needs no download and works offline. - Optional larger models (micro and mini) download once from Hugging Face. Synthesis still runs 100% on your machine. WHAT YOU CAN DO - Select text, right-click, and play it in the background while you keep scrolling. - Open the side panel (Alt+Shift+K) to paste or edit text, pick a voice, and set the speed. - Listen to a full article with one shortcut (Alt+Shift+A). - Toggle low-latency streaming or full pre-render before playback. - Audio you generate is cached locally, so replaying the same text is instant. - Save any generation as a WAV file. GPU NOTES WebGPU is enabled by default in current Chrome, but support depends on your graphics drivers. If synthesis fails, check chrome://gpu to confirm it is active. - A dedicated GPU is strongly recommended. Integrated graphics works, but a short article can take 30+ seconds to synthesize. - On hybrid laptops (dedicated + integrated GPU), set your browser to High performance in Windows: Settings > System > Display > Graphics. Otherwise Chrome usually picks the power-saving integrated GPU. - Hardware acceleration must be enabled in your browser settings. This is a goodwill project, not a business. It is free, the code is public, and there is nothing to upsell. If something isn't working right on your system, I would appreciate your feedback and patience as I improve on it.