TGRadioBooks
Check a book’s text after OCR before narration
When checking a book’s text after OCR, start with reading order, then review characters, broken words and footnotes. Compare the recognised text with the page images. TGRadio does not perform OCR: recognition and corrections happen outside the app, before you import a prepared file containing usable text.
Choose pages to compare with the scan
Keep the page images and a separate copy of the recognised text. For an initial review, choose passages near the beginning, middle and end, plus pages with complicated layouts. Include small print, tables or visibly marked pages where these occur in the source.
For example, accurate recognition of a clean page with large type says little about tightly printed notes at a chapter’s end. Compare selected passages line by line. Sampling can reveal recurring problems, but it cannot establish that every unreviewed page is accurate.
Restore the intended reading order
Check whether the end of one page joins the next in a coherent sentence. With two-column layouts, recognition can mix lines from neighbouring columns. Warning signs include abrupt topic changes, interrupted dialogue and sentences that resume several paragraphs after they began.
Compare the page layout with the order in the text file. Separate repeated running headers and page numbers from the narrative. Do not delete a line merely because it is short: it could be dialogue, a heading or an authorial aside needed to understand the passage.
Review similar characters and broken words
Look for words containing unexpected digits, mixed alphabets or internal breaks. For example, the Russian word “г0род” contains a zero instead of a letter. Visually similar Latin and Cyrillic characters can also appear inside one word. Verify corrections against the scan, especially in names and numbers.
Review line-end hyphenation separately. Rejoin a word split across printed lines, but preserve a genuine compound’s hyphen. Removing every hyphen would introduce new errors. Check quotation marks and sentence endings too, since punctuation helps determine where the narration pauses.
Inspect notes, captions and missing material
Find where footnotes have landed in the recognised text. A note may interrupt a sentence, while its reference number may become an isolated item read aloud. In a permitted copy, place necessary explanations where the main passage remains coherent, preserving the information they contain.
For example, an illustration caption may appear between two halves of a paragraph. Check its position against the page. TGRadio does not recognise images or formulas, so do not invent replacements for missing material. If the passage depends on it, seek a more suitable text source.
Listen to a corrected passage
Reopen the saved file and check that edits have preserved paragraphs and chapter order. A PDF with a text layer can be imported, but pages within one PDF may differ. Selectable text on the first page does not establish that every later page contains usable text.
Import the prepared book into TGRadio, choose a voice and press Play. Follow the source while listening to an available test passage. If you encounter a recurring error, return to the external editor and look for similar occurrences. A successful listening sample does not replace checking difficult pages.