Kam Mueang voice for Chrome (prototype, 15 October 2026)
========================================================

A Northern Thai (Kam Mueang) voice for every web page that speaks, from krueng.ai.
Once installed, a page's "Speak Kam Mueang" button, or any page using the Web Speech
API, can pick the voice "Kam Mueang (prototype)". Nothing is sent anywhere: the voice
runs inside the browser.

Privacy: it collects nothing and makes no network requests. The text a page asks it to
say becomes sound inside your browser. A short log of what it spoke is kept in the
browser's memory for the session, to check it by, and is gone when the browser closes.

Install (Chrome or Edge on Windows, Mac or a Chromebook)
--------------------------------------------------------
1. Unzip this file. Keep the folder kammuang_voice_chrome somewhere it can stay:
   the browser reads it in place, and the voice stops working if the folder moves.
2. Open  chrome://extensions  (in Edge:  edge://extensions ).
3. Turn on Developer mode (a switch at the top right; in Edge, at the left).
4. Click "Load unpacked" and choose the folder kammuang_voice_chrome.
5. Open https://krueng.ai/kammuang_recorder.html and press "Speak Kam Mueang".
   The first sentence takes a few seconds while the model loads; after that a
   sentence starts about half its length after you ask for it.

What is inside, and the terms
-----------------------------
- The voice: Meta's MMS-TTS Northern Thai model (facebook/mms-tts-nod), converted to
  ONNX. Licence CC BY-NC 4.0: free for non-commercial use, with attribution to Meta.
  https://huggingface.co/facebook/mms-tts-nod
- transformers.js (Hugging Face), Apache License 2.0. https://github.com/huggingface/transformers.js
- ONNX Runtime Web (Microsoft), MIT License. https://github.com/microsoft/onnxruntime
- The extension's own files: krueng.ai, free for anyone, for non-commercial use only.

This voice is for learning, listening and keeping Kam Mueang alive. Never use it to
impersonate anyone or to deceive. It is a prototype: the voice and accent are Meta's
model's, not a chosen speaker's; a voice recorded with a native speaker is to follow.
