For the places where NVDA's mouse tracking goes silent or stumbles: apps that cannot say what is at a point on the screen (Slack and other Electron apps, pictures of text), and PDFs whose font makes a mess of OCR.
Hold NVDA+control and click anywhere in the window (or press NVDA+control+enter with the pointer over it). The paragraph under the pointer is read. From then on, for three minutes, hovering over that window reads the paragraph under the pointer, once per paragraph. Another NVDA+control+click starts afresh at any time. The click is swallowed, so the app underneath does not see it.
The page's own text. On a web page, or a PDF open in Chrome or Edge, Mouse Reader reads NVDA's own copy of the page, so the words are exact whatever the font looks like, a bulleted list reads item by item, and the browse cursor moves to the paragraph you clicked, so the keyboard carries on from there. Scrolling needs nothing: hovering asks the live page. Reading on (below) uses NVDA's Say All through the page, and the page scrolls along when it has focus. The page's toolbar and its controls are left to NVDA to read the usual way.
OCR. Everywhere else, and on pages whose text has no paragraphs (a viewer that draws a PDF as loose snippets, say), the window is asked to draw itself into a picture (so this works with Windows Magnifier in full-screen mode and covers the whole window) and Windows' built-in OCR reads it, the same engine as NVDA+R. Paragraphs are formed from the spacing between lines, column by column, so a sidebar and a message list stay apart; bullet dots, which OCR itself leaves out, are found in the picture so that list items stay apart too. Scrolling the mouse wheel over an OCR window recognises it again half a second after the wheel stops. OCR gives plain text without links or headings, can misread words or miss small or faint text, and takes a fraction of a second for a small window and up to a second or so for a large one.
Inside a window Mouse Reader is reading, NVDA's own mouse tracking stays out of the way, so hovering behaves the same everywhere in it; everywhere else NVDA is untouched.
Reading on. Hold NVDA+shift and click (or press NVDA+shift+enter over the window) and NVDA reads from the paragraph under the pointer onwards. The mouse is ignored while it reads, and scrolling does not stop it. As with NVDA's own Say All, any key stops it (control stops, shift pauses and resumes); a click or moving to another window stops it too.
The Mouse Reader category in NVDA menu > Preferences > Settings:
A recognition is dropped when you click (any button) or press a key, since the window has probably changed; start again with NVDA+control+click.
The keyboard commands can be changed in NVDA's Input Gestures dialog, under the Mouse Reader category.
Making NVDA's mouse tracking wait until the mouse has stopped is a separate add-on, Mouse Echo Delay, from the same author. The two work together: with both installed, hovering reads a paragraph once the mouse has been still for the delay.
0.6.0: reads a web page's or a PDF's own text before turning to OCR (whole paragraphs, list items one by one, the browse cursor follows the click, read on through the page); finds bullet dots in the picture for OCR; better OCR paragraphs for book-style text (mixed-font lines merged, a full line ending a sentence continues); a capital O or an avatar in front of a name is no longer taken for a bullet; the "Read text from" setting and the optional OCR beep.
Copyright © 2026 Carrie on Accessibility. Free software under the GNU General Public License, version 2, the same licence as NVDA.