Version 0.3.2
Published on Firefox Add-ons and the Chrome Web Store on
The language a document is read in.
- Each part of a document is marked in the language the document declares for it, so a screen reader can read it in that language's voice. Declarations the text cannot bear out — digits declared English in a Japanese document, for one — are not followed.
- Document language, in the toolbar, chooses the language for one document, for a PDF that declares the wrong one — as six of the seven tagged Romanian public PDFs measured do.
- A usual document language, in the settings, off by default, for PDFs that give no language beyond the one declared for the whole file.
- Document information says which language the Reader uses, and why.
- Copying no longer stops at a control character, and in Firefox the view switch's list is readable in the light theme.
Release notes for 0.3.2 on GitHub
Version 0.3.1
Published on Firefox Add-ons and the Chrome Web Store on
Wording only: text that put two sentences side by side now joins them the way each language does.
- In English, a reading's heading count no longer runs into its description: "…they wrote. 6 headings.", where it was "wrote.6 headings."
- In Japanese, no stray space after 「。」 in five places that joined two sentences, among them the settings' on-device AI section and the notice that pages need OCR.
- The Japanese interface names the views as you choose them, in the 「表示」 menu, instead of pointing to tabs that do not exist.
- The English status line reads "Finished analysing the PDF (3 pages)."
Release notes for 0.3.1 on GitHub
Version 0.3.0
Published on Firefox Add-ons and the Chrome Web Store on
Accessible PDF View is now on Firefox Add-ons, for Firefox 144 or later on desktop, and this version fixes what reading it in Firefox showed.
- In Firefox, the Japanese interface is no longer announced as English. With NVDA, 「英語」 came before the Reader's Japanese text. The page's language is now set before the page is built, and switching it in the settings rebuilds the interface in the new language. A rebuild stops OCR running at that moment, and may stop an image-description run; what either had finished is kept.
- Settings that do nothing in Firefox are left out: the language for AI-generated text, and the note about generated results under the structure choice. Firefox has no on-device model an extension can use.
- Each radio button is named by its label alone. The explanation under each choice used to be part of the name, so the whole paragraph was spoken every time. It is now the radio's description, still shown on screen.
- The document's own title keeps the document's language, so a Japanese title is still marked Japanese when the interface is English.
- PDF.js 6.5.438. A tagged PDF whose structure tree has no ParentTree now offers its tagged readings; until now it could only be read by inference.
Release notes for 0.3.0 on GitHub
Version 0.2.4
Published on the Chrome Web Store on
These apply to the tagged reading, and to the combined reading built on it.
- A small figure the author described is shown as a picture. A tagged figure under 24pt on a side used to get a notice instead of its picture. If the author wrote alternative text for it, it is now shown however small it is, unless it is a thin line under 2pt. It is drawn finer than other figures so it can be made out. For these figures the picture now takes the place of the on-screen 「画像: …」 notice, and the author's words are its alternative text, as for any figure whose picture is shown. A small figure without the author's own words keeps the notice.
-
An empty alternative text is no longer treated as decorative.
Since 0.2.2 the Reader took a figure with an empty
/Altto be one the author marked decorative. In what was measured, Word and PowerPoint do not mean that by it: both leave a picture marked Decorative out of the structure, and PowerPoint writes an empty/Altfor a shape given no alternative text. For such a figure the Reader now says the author gave no alternative text, as it does for one without/Alt. This reverses part of what 0.2.2 below describes. - PowerPoint's 「(グループ化されたオブジェクト)」 is removed from a group's alternative text, where it read as part of the author's words.
Release notes for 0.2.4 on GitHub
Version 0.2.3
Published on the Chrome Web Store on
- The title is read once. The Reader shows a PDF's title as the page heading and leaves out a first heading that repeats it. Word tags its Title-style paragraph as a heading, and when the title had a space between two Japanese characters the two did not match, so the title was read a second time.
- Heading counts are the headings you can move to. The structure picker and the notice that another reading has headings no longer count a title heading the Reader does not show, and the Markdown view no longer repeats the title.
- A tagged document whose only heading tag is its title can now get the combined reading, on the same terms as one with no heading tags.
-
Chrome 147 or later is required (Edge 147). The bundled PDF.js uses
newer JavaScript without checking for it, and earlier releases already depended on
it: without
Math.sumPrecise, which Chrome has from 147, embedded fonts fail to load and text is read from a substitute font. That was seen in testing, not in an old browser. The store no longer installs or updates the extension in an older browser; a copy already installed there stays on 0.2.2 until the browser is updated. - The libraries underneath are current: PDF.js 6.4 and pdf-inspector 1.25.2, still with this project's fix for Japanese and Chinese text in fonts without a ToUnicode map. On the documents it is tested against, the structure the Reader shows is unchanged. The extension is 0.45 MB smaller installed.
Release notes for 0.2.3 on GitHub
Version 0.2.2
Published on the Chrome Web Store on
These apply to the tagged reading, and to the combined reading built on it.
- Each figure shows its own part of the page. Many PDFs wrap each figure in a page-sized clip — InDesign and PowerPoint exports both do — and the Reader showed the whole page in the figure's place, once for every such figure. In two documents it was checked against, that was 6 of 14 figures and 6 of 29; it is none now.
- A figure no longer shows another figure's picture. A figure too small to crop could be matched to an image nearby. It now keeps its notice instead.
- A picture drawn larger than its frame is shown at the size of the frame, rather than with a wide empty margin.
- The notice under a figure without alternative text says what is known. Where the PDF could be read for it, the Reader now says the author gave none, rather than that none could be obtained; says when a figure is too small to show; and treats a figure the author marked decorative as decorative.
Release notes for 0.2.2 on GitHub
Version 0.2.1
Published on the Chrome Web Store on
- The analyser behind the inferred reading is a newer version of pdf-inspector. Text shown one particular way in a PDF is no longer missing, two lines of very small type set close together are no longer read as one, and a page drawn from many thin lines is analysed much faster. The fix from 0.2.0 for Japanese and Chinese text is still in it.
- ビューアー is spelled one way in the Japanese interface.
Release notes for 0.2.1 on GitHub
Version 0.2.0
Published on the Chrome Web Store on
-
Structure tags are read even when the PDF does not declare itself Tagged PDF.
A PDF with a structure tree but without the declaration the standard requires
(
/MarkInfowith/Marked true) used to be read by inferring its structure from the layout. The author's tags are now used, provided they carry at least half of the text on the pages. - The document information panel says when a PDF does not conform. For such a PDF, "Structure tags" reads "Yes, but not declared as Tagged PDF", with a note that other viewers and assistive technology may not use its tags.
- Japanese and Chinese text in fonts without a ToUnicode map is read. In the inferred reading and the Markdown view it came out as noise, and the page was flagged for OCR.
Release notes for 0.2.0 on GitHub
Version 0.1.0
Published on the Chrome Web Store on
The first public release.