Wezesha means “to enable” in Swahili. It lets people who find a mouse, keyboard or screen hard to use control any website or app by voice. It runs on your own phone or computer, so what you say never leaves it.
It reads the page the way a screen reader does, so it knows every button and form field by name.
A site adds it with one line: <script src="voicenav.js">. Web apps can teach it their own actions.
Turn on “keep listening” once. After that, start with “computer” and say what you want: open a page, add to cart, fill in your phone number, hear the page read out.
Speech recognition and understanding run inside the browser. No audio is sent to a server, and after the first visit it works without a connection.
We are training the speech model on Kiswahili, and building it with organisations of persons with disabilities, as the UN Convention on the Rights of Persons with Disabilities asks.
This site is still being built. The demo works best in Chrome on a computer. The first visit downloads about 50 MB of voice models; after that it starts in a few seconds.