forked from kglitchy/speak-it
No description
- JavaScript 93.8%
- Python 1.9%
- CSS 1.8%
- Shell 1.3%
- HTML 0.7%
- Other 0.5%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
The README now covers only installing and using the extension. AGENTS.md holds the layout, architecture, browser constraints and testing notes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
| src | ||
| tools | ||
| .gitignore | ||
| AGENTS.md | ||
| Makefile | ||
| README.md | ||
Speak It
A Chrome and Firefox extension that reads web pages aloud with Kokoro-82M. Everything runs in the browser: no server, and no text or audio leaves your computer.
Install
You need uv, npm and curl.
make # download and build the model, voices and libraries (a few minutes)
make install # Chrome
make install BROWSER=firefox # Firefox 140 or later
make install opens the browser's extensions page, copies the path to load to the clipboard and prints the
steps. Chrome keeps the extension installed. Firefox removes it when it quits, so run
make install BROWSER=firefox again after restarting Firefox. Then pin Speak It to the toolbar.
Reading needs WebGPU to be fast. Without it, Chrome is slower and Firefox can't keep up with playback. Firefox has WebGPU on Windows and Apple-silicon Macs only.
Using it
- Toolbar popup: read the page or the selection, and pick the voice and speed.
- Right-click: Read selection aloud, or Read page from here.
- Shortcuts:
Alt+Shift+Rreads the selection (or the page),Alt+Shift+Ppauses or resumes. Change them atchrome://extensions/shortcuts, or in Firefox under Manage Extension Shortcuts inabout:addons. - Player: previous/next, play/pause, speed and stop. The page follows the spoken text; scroll away and the arrow button brings you back.
- Click any sentence while reading to jump to it.
- Google Docs are read from their plain-text export, without highlighting.
There are 41 voices in American and British English, Spanish, French, Hindi, Italian and Brazilian Portuguese.