How do you translate a DAISY 3 digital talking book or DTBook XML file without losing the navigation, reading order, audio timing and accessibility that make the publication useful? A well-made accessible book is not just translated paragraphs placed into a ZIP file. Its chapters, headings, page markers, notes, tables, language changes, package references and synchronized narration work together so readers can move through the material by listening, magnifying text or using assistive technology. Translation must protect those relationships while making the words genuinely understandable.
People searching for DAISY 3 translation, DTBook XML localization, digital talking book multilingual conversion, Z39.86 accessibility or text-audio synchronization need a method that handles language and structure together. The package can contain a DTBook textual content file, audio resources, navigation information and synchronization instructions. Replacing the text alone might preserve a valid XML tree while causing a narrated sentence to play at the wrong moment or a chapter heading to disappear from a reader’s navigation controls.
This world-facing guide focuses on the DAISY 3 standard ANSI/NISO Z39.86-2005 (R2012), DTBook as its text vocabulary, and the surrounding digital talking book delivery system. It is intended for accessible publishers, libraries, educational services, translation teams, disability organizations, audiobook producers and software developers. The goal is an honest language edition that readers can navigate and understand independently, rather than an attractive file whose original accessibility has been sacrificed.
Preserving an accessible reading experience while translating the book
1. What DAISY 3 means
DAISY is an approach to making publications navigable and accessible to people who cannot readily use standard print. DAISY 3 refers to the ANSI/NISO Z39.86 digital talking book specification, reaffirmed in 2012. A delivery may include structured text, audio, and synchronized text-audio features, depending on the type of book. It is not the same thing as an MP3 audiobook or a plain web page. Translators must understand which functions a particular source provides and preserve equivalent functions where the target edition claims them. A book that plays audio but no longer supports useful navigation may represent a serious regression.
2. Why DTBook matters
DTBook is an XML vocabulary used to mark up textual content for a DAISY digital talking book. Its elements describe book structure rather than merely how paragraphs look visually. This semantic structure helps software locate headings, page positions, notes and other important features. Translating DTBook therefore requires more than copying visible words. The XML hierarchy is a contract with the reading system, and the language edition should retain that contract. When the content is designed for text-to-speech, the chosen vocabulary and correct language identification also influence whether speech output sounds intelligible.
3. Understand the package as a whole
A digital talking book is assembled from multiple related resources. These may include the package file, structured text files, navigation files, synchronization documents and audio files. Their identifiers and references make the publication work as one experience. Begin by inventorying every resource and its role. A translator must not rename a file casually if other documents point to it, nor assume that translating an XML content file automatically produces a complete synchronized target edition. The package is the unit of delivery and validation. Each component should be reviewed against the specific source configuration.
4. Use the actual standard version
The DAISY 3 standard is ANSI/NISO Z39.86-2005 (R2012), not a generic label for every accessible publishing format. DTBook and the later NISO Z39.98 Authoring and Interchange Framework are related but distinct. A modern accessible EPUB may also use different structures. Before translating, document the source format, XML DTD, application profile, reading systems and intended output. If conversion is necessary, describe it as conversion rather than treating the output as identical to the source. Version awareness prevents a tool from silently discarding features that are legal in one format but unsupported by another.
5. Establish the authoritative book edition
A library may hold a corrected print edition, a previous digital edition, an accessible master and several narrated recordings. The newest filename is not necessarily the authoritative source. Record title, edition, publisher, identifier, revision and rights holder, then choose the text snapshot that translation will follow. A target-language book should not accidentally inherit outdated chapter order or withdrawn corrections from a stale master. Differences between the accessible and print editions may be purposeful, such as added explanations of figures. Preserve those decisions until the responsible editor authorizes new language-specific equivalents.
6. Create a protected structural inventory
Before text extraction, count headings, levels, paragraphs, page markers, lists, notes, tables, images, references and other key elements. Record IDs, file paths and links that depend on them. Those properties should remain unchanged unless the project explicitly authorizes a structural adaptation. The inventory provides a measurable baseline after text is translated. A valid XML document might still omit a note or merge two headings, so schema validation alone is not enough. A source-to-target comparison should identify both missing content and relationships that have shifted to the wrong location.
7. Separate readable text from XML vocabulary
The DTBook DTD defines element names and allowed structures. Translators should edit the text that readers encounter, not technical element names, DTD references or controlled attributes. Even values that look like English can function as machine instructions. Use a schema-aware extraction process with explicit permitted fields. The resulting translation table should include enough context to tell whether a string is a heading, caption, note or metadata value. After reassembly, parse the XML again and compare protected data with the source. A general-purpose word processor is not a reliable substitute for this structured approach.
8. Language codes affect speech
A DAISY reader or text-to-speech system can use language information to select pronunciation rules and voices. Localizing text without updating the relevant language metadata may cause an entire target-language book to be pronounced with the wrong phonology. Conversely, marking an originally foreign quotation as if it were entirely in the target language can also reduce intelligibility. Use supported language tags accurately, including language changes within a document where needed. Confirm behavior in real reading systems because XML correctness does not guarantee that every speech engine handles multilingual passages well.
9. Headings are navigation landmarks
A structured book lets readers jump by heading level instead of listening through every page. Preserve the hierarchy when translating chapter and section titles. A long target-language heading may require a carefully worded equivalent, but its importance in the hierarchy should not change simply to make it fit a design. Test the navigation list in an actual DAISY reader. A heading that appears visually but is missing from navigation can trap a reader who depends on keyboard or audio controls. Conversely, a mislabeled heading can send them to the wrong subject.
10. Page numbers carry source correspondence
Accessible books may preserve print page markers so students and readers can cite or follow the same reference as someone using a physical edition. A translated title or reflowed display does not automatically justify renumbering those source markers. Keep the relation to the stated print edition, and distinguish a translated edition’s own pagination if it is genuinely different. Check repeated references in exercises, tables of contents and teacher instructions. A single incorrect page marker can disrupt a classroom session in which participants use different access formats but need to discuss the same passage.
11. Reading order is not decoration
A book can render attractively while presenting content in the wrong order to assistive technology. This risk is particularly important around sidebars, image descriptions, captions, marginal notes and tables. Translation tools must preserve the semantic sequence of the source and allow approved editorial changes only where essential for comprehension. Test linear reading from start to finish as well as navigation to specific sections. A side note placed before the paragraph it explains may be confusing when spoken aloud. Accessible publishing requires thinking about the reader’s experience over time, not merely on the screen.
12. Navigation resources need matching labels
A navigation file may list chapters, page positions or other landmarks separately from the DTBook body. Translating the main heading without translating its navigation label leaves mixed-language controls. Translating the label but breaking its target reference is worse. Maintain a mapping between source headings, target headings and stable navigation anchors. Verify that each menu item lands at the intended position. Also test abbreviated labels; a concise translation should still distinguish nearby chapters. A user must be able to navigate the translated book without relying on a sighted helper to interpret unexplained navigation entries.
13. Audio changes the translation brief
Where a DAISY book contains narration, changing the language of textual content often requires new audio or a clearly defined alternative arrangement. A textual translation accompanied by unchanged source-language audio is not a synchronized target-language talking book. Establish whether the project delivers translated text only, translated audio only, or both with synchronization. Describe the accessible functions honestly. If new recording is required, plan for speaker selection, pronunciation, pause length, editing and timing review. Content translation and audio production are separate disciplines whose results must be reconciled at delivery.
14. SMIL synchronization must be protected
Synchronized Multimedia Integration Language, or SMIL, may coordinate audio playback with segments of text. The references and timing are technical data, not prose. A longer translated sentence can require a different audio segment length and new synchronization points. Do not keep an old timing map while pretending it matches new narration. A proper revision retains the correspondence between spoken words and on-screen text, reauthoring timings where required and validating every link. Test jumping mid-chapter, pausing, restarting and moving by sentence or section where the reader supports those functions.
15. Sentence segmentation is consequential
A translation may divide one source sentence into two or combine clauses for naturalness. That is acceptable linguistically in many contexts, but can affect text-audio synchronization and fine-grained navigation. Decide how to preserve or deliberately remap segment identifiers. Maintain a trace between source content and target segments so editors can review what changed. Avoid splitting a sentence purely to satisfy a rigid source timing value if it damages comprehension. The correct unit of synchronization is one that works for the actual target narration and the reading system, not an arbitrary inherited string length.
16. Timing numbers are not words
Clock values, clip positions, playback order and audio-file references should never be passed through ordinary translation transforms. Locale-specific punctuation or decimal conventions can corrupt timing data. Preserve these machine values until the audio production team deliberately recalculates synchronization for the new recording. Run a consistency check to ensure that every timing window points to an existing audio resource and a valid text segment. Then listen through the output; numerical consistency cannot prove that a recording actually speaks the intended sentence in the interval assigned to it.
17. Text-to-speech and recorded speech differ
Text-to-speech systems synthesize words from structured text, while recorded audio supplies a human performance. Translation quality affects both, but their testing requirements differ. For synthetic speech, language tags, abbreviations, numbers and punctuation can change pronunciation. For recorded speech, pacing, intonation and exact script agreement matter. Do not claim equivalence simply because both outputs sound fluent. A target-language book should be tested with its supported speech pathways, and its release notes should say whether audio is synthetic, recorded, absent or partially available.
18. Pronunciation of names and technical terms
Accessible nonfiction often contains place names, people, scientific vocabulary, acronyms and foreign expressions. A translated spelling can be understandable in print but badly pronounced by a screen reader. Build a pronunciation register for important terms, distinguish translated names from transliterations and test the actual voice engines used in distribution. Where a product supports pronunciation hints, apply them without damaging the visible source identity. Do not silently simplify a proper name into a different entity to make it easier to pronounce. Good access keeps both recognition and accuracy.
19. Footnotes must remain reachable
Notes in academic books may carry definitions, references and important qualifications. A sighted reader might notice a superscript on a page, while a nonvisual reader depends on the note’s structural link and a usable return path. Translate note content and its navigational label while keeping the correct anchor. Do not move a footnote’s meaning into unrelated prose merely to avoid technical work. Test jumping to a note and returning to the original sentence. A complete text is not functionally equivalent if essential qualifications become hard to discover or interrupt reading at inappropriate moments.
20. Endnotes and references
Some publications gather notes at chapter ends or the back of the book. The target edition should distinguish an endnote from a bibliography entry or exercise answer. Maintain reference IDs, numbering policy and links, translating connective phrases when appropriate. Citations to a source-language book may retain their published title alongside an explanatory translation, rather than replacing the citation’s identity. If chapter numbering changes in a legitimate new edition, adjust cross-references under editorial control and document the mapping. Testing links in the finished reader is more reliable than checking isolated strings.
21. Tables require a reading strategy
A complex table may look clear in print but become difficult when linearized into speech. Translation can lengthen column headings and make abbreviated units ambiguous. Preserve the relationships between row labels, column headers and cell data, and use the target-language conventions approved for units and numbers. Test how the reading system announces the table. A translator must not convert a numeric range into a descriptive phrase that changes the values or make a header imply that percentages are absolute counts. Where a source table is inaccessible, remediation is an editorial accessibility task, not a license to invent data.
22. Lists, nested lists and instructions
Lists organize steps, questions, definitions and examples. A language edition should preserve logical grouping and order, even when grammar changes within each item. This matters for directions in educational books, safety instructions and procedural documents. Keep nested relationships explicit so the reader can understand which examples belong to which rule. A visual bullet may not reveal a missing list level, while audio navigation can make the error obvious. Translate lead-in sentences carefully: a negative instruction or conditional clause can change how every subsequent item is interpreted.
23. Images and alternative descriptions
Images may need alt text, longer descriptions or captions for readers who cannot see them. Translating the caption alone does not guarantee that its accessible description is meaningful. If the source has a genuine description, translate it with reference to the actual image. If it lacks one, commission an accurate original description rather than presenting an invented summary as the author’s wording. Distinguish objective visual information from interpretation. In an educational book, spatial details in a diagram may be essential for understanding a proof or process, so the description must be reviewed by someone knowledgeable about the subject.
24. Equations and mathematical content
Mathematical notation can be encoded or represented through specialized descriptions. A translation may alter surrounding explanations, but symbols, operations, numeric values and conditions must retain their meaning. Beware of locale-specific decimal separators, minus signs, variable names and scientific units. A valid DTBook file can still contain an equation described incorrectly in speech. Test the accessible math pathway supported by the project, whether textual description, a math-specific extension, or another authorized rendering. The goal is for a reader to understand the same mathematical relationship without the translator unintentionally changing the problem.
25. Poetry and dramatic dialogue
Line breaks, stanza structure, speaker turns and stage directions can be central to the meaning of a literary text. A language edition may need different rhythm or syntax, but should preserve the structural distinction between verses and spoken dialogue. Audio narration also needs appropriate voice and pause decisions. Mark editorial adaptations clearly, especially where a poem is paraphrased to fit accessibility constraints. Do not convert a poetic line into a chapter heading or a character name into an unmarked stage instruction. Structural accuracy makes expressive translation accessible to readers navigating by voice.
26. Foreign quotations and mixed languages
A nonfiction book may quote Latin, French, Chinese or other languages within a translated narrative. Decide whether to retain the quotation, translate it, or present both. Mark changes of language accurately so speech engines can pronounce passages appropriately where possible. Keep attribution and quotation boundaries visible in the structure. A source quotation should not silently become the translator’s paraphrase. For language learners and scholars, access to the original wording may be part of the purpose of the book. The best implementation supports comparison without forcing everyone to listen through duplicated material unnecessarily.
27. Right-to-left and complex scripts
Arabic, Hebrew and several other writing systems present challenges when mixed with Latin acronyms, numbers, tables and source page markers. The logical order of the XML content and the visual order in a reader must both be tested. Do not assume that mirroring navigation icons changes the direction of embedded mathematical equations or source citations. Test headings, footnotes, mixed-script lists and accessibility announcements with target-language users. Where a reader has limited support for complex shaping, document the limitation and avoid declaring the edition fully accessible merely because the content file parses correctly.
28. Metadata needs translated discoverability
The package and content may contain titles, creators, language statements, identifiers and publisher information. Translate display titles or descriptions according to a policy, but preserve book identity and rights statements. A translated title should not cause a library catalogue to treat the language edition as an unrelated work. Record the relationship between original and translated expressions, with edition and language clearly stated. Accurate metadata helps a user select a book in the intended language before downloading it. It also allows assistive devices to choose language-specific settings before starting playback.
29. The unique identifier is not a translated title
The DAISY specification uses identifiers to bind the package and content resources together. An identifier such as a book UID has a technical identity function even when it is stored beside readable metadata. Do not replace it with a translated title or reset it casually when editing text. Decide whether the target edition needs a new publication identifier while retaining explicit relation to the source, according to the publishing system’s policy. Then ensure that package and content references agree. A broken identifier relationship can prevent readers and libraries from reliably recognizing the publication.
30. Permissions and rights
Accessible-format production may operate under a license, publisher agreement or country-specific legal exception. Translation is not automatically permitted simply because an accessible format was lawfully produced. Establish rights for source text, translated expression, narration and distribution territories. Preserve copyright notices and explain any restrictions accurately in the target language without changing their legal meaning. Do not suggest that DAISY conformance grants redistribution rights. The technical format and the authority to make or distribute a language edition are independent questions and should both be settled before publication.
31. Extract strings in their book context
A translator presented with isolated XML fragments cannot tell whether a phrase is a heading, image description, question, note or spoken instruction. Supply context, section path, neighboring sentences, links to media and a source preview. Protect structural identifiers explicitly. A short word like table may refer to furniture in a narrative or a data structure in a textbook. Context also helps translators choose suitable sentence lengths for narration. Quality improves when the team knows how the text will be heard, not merely how it looks as a row in a translation memory.
32. Make a reader-focused terminology register
Accessible books should use consistent terms for navigation, chapters, notes, units, labels and discipline-specific vocabulary. Record preferred target expressions, forbidden ambiguities, pronunciation guidance and examples. Sometimes consistency must yield to the author’s quoted language or a technical standard; mark these exceptions rather than forcing all occurrences into one translation. Review terms that are critical when heard aloud, such as homophones and abbreviations. The glossary is useful when editors can see both semantic meaning and listening context, and it helps future editions remain coherent after corrections.
33. Validate against the DTBook DTD
DTD validation checks whether the XML uses the allowed structure for the declared DTBook version. It is essential, but it does not prove that a translation is complete or understandable. Validate each text file and the wider package using tools appropriate to the DAISY specification. Check that source-to-target structure remains plausible, IDs are unique, and references resolve. Then inspect actual language content. A missing negative, a wrong translated technical term or an unpronounceable acronym will not necessarily trigger a schema error. Accessibility requires tests beyond formal conformance.
34. Inspect navigation and synchronization together
Use automated comparisons to confirm the existence of references, then perform interactive tests on a reader. Jump from the table of contents to a middle chapter, navigate by page, visit a footnote, return to the main text and replay a sentence. When audio is present, verify that spoken output and highlighted text correspond. Also test interrupted playback and resumption. A package that works only when played continuously from the beginning does not meet the practical needs of many readers. Navigation and synchronized access need end-to-end confirmation.
35. Include native users in accessibility QA
Developers and translators may understand the specification yet miss barriers encountered by people using screen readers, refreshable braille displays or audio-only navigation. Include reviewers who read in the target language using the actual access methods. Ask them to find a chapter, locate a cited page, follow a footnote and answer a content question. Record the obstacle, not just whether the user eventually succeeded. This reveals excessive announcements, unclear labels and navigation traps that a visual proof cannot show. Functional independence is an essential publication criterion.
36. Handle legacy-book conversion honestly
Many collections contain older DTBook and DAISY materials that are being moved into newer accessible formats. A conversion pipeline may preserve the prose but lose page correspondence, navigation depth, note relations or audio alignment. Document which features are preserved, transformed or unavailable, and test a representative sample against the source. Do not call a converted EPUB the original DAISY 3 package. The goal is continuity of reader capability, not merely a different file extension. If a feature cannot be preserved, explain the loss and plan a compensating accessibility approach.
37. Quality review across three disciplines
Language editors must verify terminology, tone and factual meaning. Accessibility reviewers must verify navigation, reading order and usability. Technical specialists must validate XML, file references and synchronization. Audio producers must review pronunciation and timing where recorded narration exists. These roles should exchange specific evidence rather than approve a single vague quality score. For instance, a sentence may be excellently translated but misaligned to its audio clip. A book should not be released until each applicable dimension has an owner and a demonstrated pass condition.
38. Preserve versions and repair history
Save the approved source package, translated texts, audio masters, mapping tables, validation logs and final distribution files. Record the language, edition, build date and known issues. When a source correction arrives, determine which target strings, audio clips and navigation references depend on it. Updating text without replacing the affected recording may create a new mismatch. A traceable revision trail makes it possible to repair the right components instead of rebuilding the entire book blindly. It also gives libraries confidence that the file they hold has a known relationship to the publisher’s source.
39. Publish a truthful capability statement
A target-language release should state whether it offers structured text, synthesized speech support, recorded audio, text-audio synchronization, page navigation, notes, image descriptions and other accessibility features. Do not call every DAISY file a fully synchronized talking book. Readers have different access needs, and precise descriptions help them choose usable resources. Include language and edition identifiers, format version, source relationship and support information. A credible publisher reports limitations honestly and improves them where possible. Better communication is part of access, not an admission that the publication has failed.
40. Make the international handoff reproducible
Deliver the validated package, declared DAISY standard, exact DTBook DTD, language tags, pronunciation policy, audio specifications, source mapping, QA report and reader-compatibility notes. Test the full workflow with an organization in the target market because device and assistive-technology support can vary. Provide a clear path for corrections to language, synchronization and structure. Keep the author’s intended content and the reader’s independent access as the two non-negotiable goals. A translation succeeds when someone can learn from the new-language book without losing the essential tools provided by the accessible source.
Worked translation and accessibility failures
Chapter navigation disappears
An accessible science textbook is translated correctly, but the production tool collapses several chapter headings into ordinary paragraphs. The book sounds reasonable when played from the beginning, yet students cannot jump to a required chapter. Compare the source and target heading hierarchy, restore the correct DTBook structure and regenerate navigation resources. Then have a reader locate the assigned chapter without a visual table of contents. This tests the feature that actually failed. A translation that preserves the words while removing independent navigation has not preserved the publication’s educational function.
Audio and highlighting drift
A target-language recording lasts longer than the source-language recording, but the old SMIL timing references remain. On-screen highlighting moves ahead while the narrator is still speaking. The document may continue to parse and the audio may sound excellent independently. Re-segment the target-language text where necessary, regenerate timing against the actual recording and test skipping, pausing and resuming from several positions. Do not stretch audio arbitrarily to fit inherited timings if it harms intelligibility. The reader needs accurate alignment, not the illusion of technical compatibility.
A translated table changes a unit
A chemistry textbook table labels a concentration in milligrams per litre. During translation, the abbreviation is expanded incorrectly as grams per litre, changing the stated measurement by a thousandfold. The markup remains valid and all cells are present. The failure is factual, not structural. Verify numeric values, scientific units and table-header associations with a subject expert, then listen to how the screen reader announces the corrected header. Accessible data must be accurate both as mathematics and as spoken language. Visual similarity offers no protection against a wrong unit.
A print-page reference is lost
A school uses print and accessible formats of the same novel. In the target-language DAISY edition, page markers are renumbered according to a new layout without saying that they no longer correspond to the cited print edition. When a teacher requests page 72, students cannot find the same passage. Preserve the stated source-edition page references or clearly document a separate pagination scheme and provide crosswalks. Test page navigation through the actual reading system. The feature exists to help readers coordinate, not to imitate the appearance of a physical book.
A footnote becomes unreachable
An academic reader hears a claim followed by a note marker, but selecting it produces no response because the target-language reassembly changed the note ID. The footnote text is still in the book, so a simple word-count comparison reports no loss. A link-resolution test identifies the missing relationship, and a reader confirms that the return path works after correction. The lesson is to validate connections, not just content presence. For scholarly material, inaccessible evidence can be as harmful as omitted evidence.
A multilingual quotation sounds wrong
A Spanish-language biology guide includes an English quotation and a Latin species name. The package marks every line as Spanish, so synthetic speech pronounces foreign words inconsistently. The translator considers rewriting the terms to match speech, but that would corrupt the scientific citation. Instead apply appropriate language handling where supported, preserve scientific names and test voices in the intended reading systems. Provide a pronunciation guide if necessary. The user’s access needs are served by accurate language metadata and speech support, not by changing the evidence.
An image description invents data
A translated geography book describes a graph that the source accessibility file never described. The new alt text confidently claims that rainfall doubled, although the chart shows a different trend. A native editor approves the wording, but no one checked the figure. Reassess the actual image, supply an accurate description with subject-matter review and label it as added accessibility content where editorial policy requires. Accessibility improvements are welcome, but they must not introduce unsupported facts. The same evidential standard applies to added descriptions as to the main translation.
A conversion hides its limitations
A library converts a legacy DAISY collection into EPUB and announces that all accessibility is unchanged. Some volumes lose text-audio synchronization and print-page navigation. Users discover the problem only after downloading the books. Compare functionality before and after conversion, label each output format accurately and publish a capability statement for each volume. If an essential feature is absent, disclose it and prioritize remediation. International access depends on honest technical claims, not the convenience of calling every digital edition equivalent.
DAISY 3 and DTBook frequently asked questions
Is DAISY 3 the same as an audiobook?
No. DAISY 3 can support structured navigation, text, audio and synchronization. An ordinary audio recording may lack the navigational and text features of a digital talking book.
What is DTBook XML?
It is a structured XML vocabulary for textual content within the DAISY/NISO digital talking book ecosystem. Its hierarchy and links help accessible reading tools interpret the book.
Should identifiers change for a language edition?
Follow the publishing system’s edition-identity policy, but never arbitrarily translate identifiers or break references between package resources.
Can translated text use the original audio timings?
Only if an actual alignment check proves they match. New-language narration normally requires corresponding segmentation and timing review.
Do I need to translate navigation labels separately?
Yes, where navigation resources carry display text. Maintain the links between those labels and their intended headings or page markers.
Is DTD validation sufficient?
No. It checks structural conformance, not factual translation quality, usability, spoken pronunciation or accurate text-audio synchronization.
What if the original image has no description?
Commission an accurate description and get it reviewed rather than inventing one from the caption. Keep added accessibility content distinguishable where appropriate.
Can DTBook be converted to EPUB?
Yes, conversion workflows exist, but feature preservation must be tested. A converted EPUB is not automatically an identical DAISY 3 delivery.
Who should review a target-language talking book?
Language editors, technical validators, audio specialists when applicable, and users who rely on accessible navigation and assistive technology.