Ivory Dog

From Conversation to Book

How to Turn Voice Recordings and Transcripts Into a Manuscript

Article 010 · About 11 minutes to read

In this article
  1. 1. Understand the Four Layers of Recorded Material
  2. 2. Preserve the Original Recording First
  3. 3. Produce a Transcript
  4. 4. Decide What Type of Transcript You Need
  5. 5. Establish a Consistent Transcription Convention
  6. 6. Verify the Transcript Against the Audio
  7. 7. Build a Searchable Index
  8. 8. Extract Material Without Losing Provenance
  9. 9. Separate Speaker Statements From Editorial Interpretations
  10. 10. Transform Spoken Material Into Readable Prose
  11. 11. Know When to Preserve Direct Quotation
  12. 12. Use Claude for Editorial Organization, Not Invented Testimony
  13. 13. Return to the Author When the Source Is Incomplete
  14. 14. Protect Sensitive Source Material
  15. 15. Make Recorded Material Accessible
  16. 16. Conduct a Final Source-to-Manuscript Audit
  17. Conclusion

A recording captures speech. A transcript represents that speech in written form. A manuscript transforms selected material into a coherent work intended for readers.

These are three different artifacts.

Confusing them can lead to serious editorial problems.

An accurate transcript may be repetitive, fragmented, and difficult to follow as continuous prose. A polished manuscript may be beautifully written yet misrepresent what the speaker actually said.

The challenge is to develop readable prose while preserving the integrity of the original information.

For authors creating books through conversations with Claude, this distinction is especially important.

The conversation supplies ideas, experiences, explanations, and decisions. The manuscript presents the selected material through an intentional structure.

A reliable process must connect the two.

1. Understand the Four Layers of Recorded Material

The transformation from recording to manuscript is easier to manage when the material is divided into four layers.

The original recording preserves the captured audio or video.

The verified transcript represents the spoken words, with documented corrections and indications of uncertainty.

The editorial source file organizes relevant passages, themes, facts, and decisions.

The manuscript presents the material in polished prose for readers.

Each layer serves a distinct purpose.

The recording is the primary source for what was spoken.

The transcript enables searching and reference.

The editorial source file supports selection and organization.

The manuscript is the final expressive work.

Changes made to one layer should not silently overwrite the others.

Visual 1. Recording-to-manuscript workflow
StageInputWork performedOutput
1. PreservationOriginal audio or videoSave protected master and working copyArchived source recording
2. TranscriptionWorking recordingConvert speech into textDraft transcript
3. VerificationRecording and draft transcriptCorrect recognition errors and speaker labelsVerified transcript
4. IndexingVerified transcriptIdentify topics, timestamps, and key passagesSearchable source index
5. Editorial extractionIndexed materialSelect relevant passages and document provenanceChapter source files
6. CompositionApproved chapter source filesDevelop readable prose without unsupported additionsManuscript draft
7. Author reviewDraft and supporting sourcesCorrect meaning, tone, interpretation, and omissionsRevised manuscript
8. Final verificationRevised manuscript and original sourcesTrace important claims and quotations to evidenceApproved manuscript
Original Ivory Dog editorial workflow. Each stage should preserve the ability to recover earlier versions and trace material to its origin.

2. Preserve the Original Recording First

Before beginning transcription, create a protected copy of the original recording.

Do not use the only existing copy for destructive editing.

A practical archive should identify the recording date, speaker or speakers, general subject, duration, recording method, and any restrictions on access or use.

Consistent filenames are helpful.

For example:

family-history_2026-11-16_session-01_original.wav

A corresponding working copy might include the designation working.

If the audio contains confidential or sensitive material, storage and access should be controlled appropriately.

The National Park Service's oral-history guidance recommends protecting original recordings, applying consistent filenames, and coordinating archival arrangements with an appropriate repository when long-term preservation is intended.[1]

For an ordinary private book project, a formal institutional repository may not be necessary.

Nevertheless, the underlying principles of preservation, identification, and recoverability remain valuable.

3. Produce a Transcript

Transcription converts speech into text.

It may be performed manually or with automated speech-recognition software.

Automated systems can accelerate the process, but they may produce errors.

The most significant problems often occur with proper names, uncommon terminology, dates, numbers, overlapping speech, and poor-quality audio.

For example, a recording discussing electrical equipment might contain the phrase 15-amp circuit.

A transcription error that changes this to 50-amp circuit could create a consequential technical mistake.

Similarly, an inaccurate surname could compromise a family history.

Therefore, automated transcription should be considered a preliminary conversion, not an independent guarantee of accuracy.

4. Decide What Type of Transcript You Need

The appropriate transcription style depends on the intended use.

A strict verbatim transcript may preserve hesitations, repeated words, sentence fragments, and other features of spontaneous speech.

An edited or clean-read transcript may omit some nonessential vocal fillers while preserving the speaker's wording and meaning.

A thematic summary identifies important subjects but does not attempt to reproduce the entire conversation.

The Library of Congress explains that transcripts and indexes provide different forms of access to recordings, and its transcription guidance emphasizes accurate representation of the speaker's language.[2]

For book development, a carefully checked transcript is generally preferable to relying solely on a summary.

A summary can omit qualifications, context, or apparently minor details that later prove important.

However, the transcript itself should not be quietly rewritten into polished book prose.

That transformation belongs in the editorial stage.

5. Establish a Consistent Transcription Convention

Before reviewing a large collection of recordings, establish rules for how transcripts will represent uncertain or nonverbal material.

Useful conventions might include:

[inaudible 00:12:43] — Speech cannot be understood.

[name uncertain] — A possible proper name requires confirmation.

[long pause] — Silence is relevant to the meaning or delivery.

[interruption] — The conversation was interrupted.

[speaker unidentified] — The speaker cannot be established confidently.

These conventions are examples rather than a universal standard.

The important principle is consistency.

A transcript should not supply guessed words without marking uncertainty.

Nor should it suppress meaningful hesitation or disagreement simply to make the speaker appear more polished.

For scholarly oral histories, the transcription conventions should be documented and consistent with the requirements of any receiving repository.

6. Verify the Transcript Against the Audio

Transcription verification is a distinct editorial task.

The reviewer should compare the draft transcript with the recording, concentrating particularly on passages where recognition errors are likely or consequential.

A systematic review should check:

Speaker identification.

Names of people and places.

Dates and chronological references.

Numerical values and units.

Technical terminology.

Negations and qualifications.

Interruptions that affect meaning.

Uncertain or inaudible passages.

Negation deserves particular attention.

The difference between I did approve the decision and I didn't approve the decision can fundamentally alter the meaning of a recollection.

After correction, the transcript should be identified as a reviewed version, with the reviewer's name or role and the date where appropriate.

If it remains incomplete, its status should clearly indicate that limitation.

7. Build a Searchable Index

A substantial interview collection may contain dozens of hours of recordings.

Even an accurate transcript can become difficult to navigate without an index.

The index should identify principal subjects, important people, dates, significant events, and relevant timestamps.

For example:

Recording timeSubjectEditorial relevance
00:04:15Childhood in a coastal townPossible introductory chapter
00:19:40First employmentCareer chronology
00:37:10Industrial accidentMajor narrative turning point
00:52:30Change in professional philosophyCentral theme
01:08:05Reflections on retirementPossible concluding chapter
Illustrative index using fictional entries. Timestamps do not refer to an actual recording.

The Library of Congress recommends detailed interview logs and indexes because they allow researchers to locate important material without repeatedly searching through entire recordings.[2]

For book development, indexing also reduces the risk of overlooking relevant passages when the manuscript's structure changes.

8. Extract Material Without Losing Provenance

Once the transcript has been verified and indexed, relevant material can be assigned to chapters.

This process should preserve a connection to its origin.

For example:

Source: Interview 03, 00:24:10–00:29:35.

Topic: Reasons for leaving the family business.

Proposed chapter: Chapter 6.

Key material: Disagreement over investment decisions.

Verification status: Author's recollection; company records not yet reviewed.

Editorial restriction: Avoid attributing motives to other family members without evidence.

This source note makes the material easier to use responsibly.

If a later draft asserts that the business failed because of a specific investment decision, the editor can return to the source and examine whether the interview actually supports that conclusion.

Traceability is particularly important when AI assists with manuscript drafting.

A plausible sentence should not become authoritative merely because it has been written smoothly.

9. Separate Speaker Statements From Editorial Interpretations

Consider a fictional interview passage:

We expanded the factory in 1998. I thought it was too soon, but everyone else wanted to go ahead.

A manuscript might reasonably paraphrase:

The author recalls disagreeing with the decision to expand the factory in 1998.

That rendering remains close to the source.

But consider:

The reckless 1998 expansion reflected management's disregard for financial discipline.

The second sentence introduces judgments not established by the original statement.

It may or may not be supportable through additional evidence.

The distinction must be examined before the material is published.

A rigorous editorial process can classify proposed statements as direct testimony, verified documentary fact, attributed interpretation, or unresolved assertion.

This reduces the likelihood that unsupported interpretations will be incorporated into an apparently factual narrative.

10. Transform Spoken Material Into Readable Prose

The manuscript does not need to reproduce conversation literally.

Spoken explanations may move backward and forward in time, repeat subjects, and introduce relevant information after the discussion has shifted elsewhere.

An editor may reorganize the material for clarity.

The challenge is to preserve its meaning.

Consider this invented exchange:

Interviewer: What changed after the company moved?

Author: The customers were different. We had expected larger orders, but we didn't get them immediately. What really surprised us was the shipping cost. It hadn't looked so bad on paper.

A proposed manuscript passage might read:

The relocation did not produce the immediate increase in order size that we had anticipated. The new customer base behaved differently, and transportation costs proved more burdensome than our initial estimates suggested.

The passage is clearer and more concise.

However, the author should confirm that transportation costs accurately represents the specific shipping expenses discussed and that the manuscript has not exaggerated the financial impact.

The proper editorial question is not merely whether the new passage reads well.

It is whether it faithfully represents the intended meaning.

11. Know When to Preserve Direct Quotation

Some spoken passages possess literary or historical significance because of their precise language.

A memorable expression, distinctive turn of phrase, or emotionally important statement may lose its force if paraphrased.

Direct quotation can therefore be valuable.

However, quotations must be accurately attributed.

The author's own recorded words may be used according to the project's appropriate rights and editorial arrangements.

When quoting other interview participants, permission and contractual considerations may arise.

Consent to participate in an interview is not necessarily identical to permission for every later commercial use of the material.

The Oral History Association explicitly distinguishes informed consent from copyright.[3]

Authors should determine the rights and restrictions applicable to each interview before publication.

12. Use Claude for Editorial Organization, Not Invented Testimony

Claude can help classify transcript passages, summarize themes, suggest chapter assignments, and propose draft prose.

But the instructions governing the work should be explicit.

For example:

Using only the supplied interview material, draft a chronological account of the author's first year in business. Preserve uncertainty, do not invent quotations or events, and identify any questions that require further interviews.

Such instructions establish a useful boundary.

They do not guarantee perfect compliance.

The output still needs review.

If Claude supplies an unsupported explanation, misreads a sequence of events, or introduces a detail absent from the source, that issue must be corrected.

For important claims, reviewers should be able to identify the supporting transcript passage or independent documentary source.

13. Return to the Author When the Source Is Incomplete

The transcript may contain gaps that become apparent only during manuscript development.

A critical event may be described without its date.

A participant may be identified only by a first name.

A technical explanation may omit an important step.

A family story may contain conflicting recollections.

These are reasons to ask additional questions.

They are not invitations to invent details.

A follow-up conversation with Claude can identify precisely what the author needs to clarify.

Where the author cannot supply the information, the manuscript should either acknowledge the uncertainty or omit the unsupported detail.

The appropriate choice depends on its importance to the book.

14. Protect Sensitive Source Material

Recorded conversations can contain personal information, private experiences, and references to people who never agreed to participate in the project.

Authors should consider the security of recordings, transcripts, and manuscript drafts.

Access should be limited appropriately, especially when sensitive information is involved.

Before uploading source material to third-party AI services, authors should review the applicable data-handling terms and determine whether they have the authority to share the material.

For interviews involving other people, informed consent and agreed publication restrictions should be documented.

The process should also accommodate requests to exclude sensitive passages where appropriate and legally permissible.

The strongest editorial safeguard is to consider these matters before collection begins, rather than attempting to resolve them after a book has been prepared for publication.

15. Make Recorded Material Accessible

If the finished project includes audio interviews or video materials, accessibility should be considered separately from the accessibility of the manuscript.

The World Wide Web Consortium distinguishes between transcripts, captions, and audio descriptions, explaining that they serve different accessibility needs.[4]

A text transcript can help people who cannot hear an audio recording.

Captions are important for recorded video containing meaningful audio.

Descriptive transcripts may also convey visual information required to understand video content.

For book-development archives, properly formatted text can assist searching, assistive technology, and later editorial review.

The exact accessibility requirements depend on the material and the means of publication.

16. Conduct a Final Source-to-Manuscript Audit

Before approving a manuscript developed from recordings, conduct a dedicated verification pass.

This should occur after ordinary developmental editing and copyediting.

Select the manuscript's important factual statements, direct quotations, dates, and substantive interpretations.

Trace them back to the original source material.

Check whether the meaning has been preserved.

Confirm that uncertain details remain appropriately qualified.

Verify that source material has not been combined in ways that create a misleading sequence.

Examine whether important passages were omitted in a manner that distorts the author's account.

Then review the corrections once more.

This final step is especially important because a well-written manuscript may conceal inaccuracies behind polished prose.

A successful editorial process must distinguish apparent quality from demonstrated reliability.

Conclusion

Voice recordings and transcripts can provide extraordinarily rich source material.

They preserve stories, explanations, professional knowledge, and personal perspectives that might otherwise remain unrecorded.

But they are not finished books.

The transformation into a manuscript requires accurate transcription, indexing, source tracking, careful selection, composition, author review, and verification.

Claude can assist with many of these operations, helping the author organize information and develop prose.

The editorial standard remains the same: no unsupported details should enter the manuscript merely to make it sound complete.

The objective is to produce a book that is both readable and faithful to the material from which it emerged.

A reliable manuscript should allow its author to answer two questions with confidence:

Is this what I intended to communicate?

Can I establish where its important factual claims came from?

A book that satisfies both requirements has a much stronger foundation than one that merely transforms a long recording into an equally long document.

References

  1. National Park Service. Introduction to Oral History. https://npgallery.nps.gov/GetAsset/06d646f7-f80d-4cda-926c-f643dcebc9ef/original
  2. Library of Congress. Transcribing Interviews — Veterans History Project. https://www.loc.gov/programs/veterans-history-project/how-to-participate/transcribing-interviews/
  3. Oral History Association. Informed Consent. https://oralhistory.org/informed-consent/
  4. World Wide Web Consortium, Web Accessibility Initiative. Transcripts. https://www.w3.org/WAI/media/av/transcripts/
  5. Oral History Association. Oral History Best Practices. https://oralhistory.org/best-practices/

Visual attribution

Visual 1, Recording-to-Manuscript Workflow: Original Ivory Dog editorial process table.

Additional instructional table, Illustrative Recording Index: Original fictional example created for this article.

No third-party visual assets are used.