How Text to Speech Handles Footnotes and Tables

A reference marker in the middle of a sentence can interrupt the reading. LoudReader filters common bracketed or numbered markers from the spoken text. When it recognises a structured footnote and can resolve the reference, the reading view provides a control to open the note beside that reference.

The app also collects recognised footnote text and appends a Notes section as it reaches the end of the book’s reading flow. That does not establish that every note in every file was found: nonstandard markup, unresolved references and a PDF’s flattened layout can change what is available.

For study, open one early note and compare it with the source. If a citation matters to the argument, follow it when you encounter it rather than waiting until the end of a long book.

In structured content, ordinary data tables are preserved for visual inspection and marked as not read aloud. That avoids turning a grid of measurements into an unexplained stream of numbers while keeping the table available in the reading view.

Some books use table markup to lay out continuous prose. The app uses heuristics to identify those cases and unwrap them into paragraphs; dialogue layouts have their own handling. These checks are useful, but they cannot infer every author’s intention.

A PDF can be different. If its table has already become separate text lines during extraction, the original row and column structure may not be available to the later filter. Review the source table whenever you need a precise value or comparison.

A heading can carry content or merely repeat navigation. The filter treats web articles and books differently: navigation and page furniture are removed from web content, while book headers can contain meaningful chapter titles and dates.

PDF import has separate extraction and repeated-header handling. If a running title or page number keeps interrupting speech, compare the imported text with the original PDF. A cleaner EPUB from the publisher may be more useful than repeatedly listening around a difficult page layout.

Speech normalisation can rephrase page pointers and remove certain numbered figure asides. For example, a recognised “on page” reference may become “as mentioned earlier”. These changes make a spoken sentence flow differently from its written source.

They do not explain the referenced figure. When following an argument closely, keep the original page and cross-reference available. A fluent replacement phrase is not evidence that the visual information has been included in the narration.

LoudReader now includes on-device OCR for image-based PDFs. Recognition can recover prose from a legible scan, but it does not guarantee semantic footnote links or the structure of a table. Blurred superscripts and tiny print are particularly worth checking.

The scanned-PDF guide explains recognition limits. For file choices, see supported import formats. LoudReader generates speech locally; an annotated source still deserves a visual check alongside listening.