Oral History Transcription: How to Preserve Community Voices Accurately

1. Full Verbatim vs. Clean Verbatim: Where Oral History Sits

The verbatim debate runs through all qualitative transcription, but oral history has its own particular version of it, and the stakes are higher than in most other contexts.

Full verbatim captures every utterance: filler words, false starts, repetitions, stutters, incomplete sentences, and non-verbal sounds. It's the right choice when the speech itself is what's being studied, in linguistic analysis, discourse research, or legal proceedings where the audio must mirror the text line by line.

Clean verbatim removes verbal clutter while preserving meaning and voice. Most oral history archives land here. The goal, as stated in the T. Harry Williams Center for Oral History's style guide, is that "transcribing verbatim does not mean that you must include every single sound." Fillers like "ah," "er," and "um" don't need to be included. Crutch phrases like "you know" and "you see" should stay unless they appear so constantly that they obscure the actual content. False starts that represent a genuine stumble rather than meaningful hesitation can be dropped.

The ethical line is clear: you can remove noise that doesn't change meaning, but you cannot change meaning, tone, or intent. Removing "um" from a sentence doesn't change what the narrator said. Correcting grammar does.

What never to do: Make a narrator sound more articulate or educated than they were. If someone says "I didn't know nothing about it then," changing it to "I didn't know anything about it then" is a rewrite. The double negative is part of how that person speaks, and in oral history, how a person speaks is data.

2. Dialect, Vernacular, and the Phonetic Spelling Problem

This is where well-intentioned transcriptionists make the most damaging errors, and where the standard is clearest.

Use standard spelling. Do not try to reproduce accents or dialects phonetically.

This is confirmed across multiple authoritative oral history style guides including the T. Harry Williams Center and Guilford College's oral history program. Writing "gonna" instead of "going to," or "wanna" instead of "want to," or rendering a Southern accent phonetically, does not preserve voice. It stereotypes it. A reader encountering exaggerated phonetic spelling reads the narrator differently than they would otherwise, and almost always reads them as less educated or less credible.

What does preserve voice is keeping the narrator's actual word choices, syntax, and sentence structures intact. If a narrator consistently uses a regional term, a historical phrase, or a non-standard grammatical construction, those stay exactly as spoken. They're rendered in standard spelling, but they're not corrected to match standard grammar.

The distinction: standard spelling, non-standard grammar and vocabulary. The transcriptionist doesn't clean up the grammar. They just don't reproduce the accent in writing.

3. Non-Verbal Cues and Annotation

Oral history recordings carry information that words alone don't convey. A five-second pause after a question about loss. A laugh that breaks the tension of a difficult memory. Tears that appear mid-sentence. A gesture toward a photograph on the wall. These belong in the transcript when they're meaningful.

The standard notation conventions, confirmed across oral history style guides, use bracketed descriptors in the body of the transcript:

• [pause] or [pause 5s] for significant silences

• [crying] or [voice breaks] for emotional moments

• [laughter] for spontaneous humor or lightness

• [points to photograph] or [gestures toward window] for physical references to objects or locations

• [inaudible] for sections that cannot be reliably heard, with a timestamp so researchers can return to the audio directly

• [unclear] for sections that are audible but uncertain, sometimes followed by a best-guess rendering in brackets: [unclear - sounds like "Martinez"]

The principle: annotate what's meaningful, not everything. A brief pause between sentences doesn't need notation. A long silence after a question about a deceased family member does.

4. Legal, Ethical, and Consent Requirements

Oral history carries legal requirements that most other qualitative research doesn't. The narrator's words in a transcript are their intellectual property. They don't automatically belong to the institution or researcher conducting the interview.

Deed of gift: The standard legal instrument in oral history is the deed of gift, a legal release that transfers rights from the narrator to the collecting institution. It must cover both the original audio recording and the written transcript. A deed that covers only the audio doesn't authorize publishing the transcript. Get this signed before the interview if possible, and certainly before any transcripts are made public or deposited in an archive.

Narrator review: Standard practice across oral history institutions is to return the draft transcript to the narrator for review and sign-off before it enters any archive or publication. The Margaret Walker Center for Oral History includes explicit narrator review and approval language in its transcript front matter: "the following transcript has been reviewed, edited, and approved by the narrator." Narrator review allows the narrator to correct proper names and factual details, clarify ambiguous statements, and flag personal or family information they didn't intend to make public. It does not mean the narrator can rewrite what they said to make themselves sound better. The right of correction is limited to accuracy, not to retrospective editing of opinions or memories.

Sensitive and traumatic content: Oral histories frequently involve testimony about events the narrator found painful, frightening, or shameful. GBV, conflict, displacement, racism, poverty, and family trauma all appear in oral histories. Transcriptionists working with this material need to handle it carefully and confidentially. NDA-bound transcriptionists matter here for the same reason they matter in clinical research: the people behind these recordings are identifiable, and what they shared was often shared in trust.

Redaction and restricted access: Some narrators will ask, during review, to restrict certain portions of a transcript from public access for a defined period. A narrator might agree to have a transcript deposited but not made publicly accessible until after their death, or until a certain year. These restrictions should be documented in the deed of gift and honored by the archive.

5. Technical Standards for Archival Deposit

A transcript that can't be found in an archive in fifty years serves no one. Archival standards for oral history transcripts exist specifically to ensure longevity and searchability.

Front matter: Every transcript should include a standardized header before the interview text begins. The T. Harry Williams Center and the Margaret Walker Center both specify this. Standard fields include:

• Narrator name

• Interviewer name

• Date of interview

• Location

• Project title

• Name of the collecting institution

• A note on the transcription style used (verbatim, clean verbatim, or edited)

• A statement that the transcript has been reviewed and approved by the narrator

Timestamping and audio synchronization: Timestamps embedded in the transcript at regular intervals (every two to five minutes is standard) allow researchers in digital archives to jump directly from a point in the text to the corresponding moment in the audio. The Oral History Metadata Synchronizer (OHMS), developed by the University of Kentucky and widely used in digital oral history archives, indexes transcripts and audio together so researchers can search the text and land directly on the audio segment. This only works if timestamps are present and consistently formatted.

File formats for longevity: Different formats serve different purposes in an oral history collection:

• .docx for working files, revisions, and narrator review copies

• Unformatted .txt or PDF/A for archival preservation masters, since plain text and PDF/A are stable formats that don't depend on software that may not exist in twenty years

• XML for digital archives that require structured data for search and discovery

Don't deposit only a .docx file in an archive that expects to preserve materials for decades. It is a working format, not a preservation format.

6. The Step-by-Step Production Workflow

Step 1: Pre-Interview Preparation

Before transcription begins, create a proper noun list: local landmarks, family names, community organizations, historical events, and place names specific to the narrator's context. A transcriptionist working cold on an interview about a specific neighborhood in 1960s New Orleans without knowing any of the local geography will miss things a prepared one catches. Share this list with the transcriptionist before they start.

Step 2: First-Pass Transcription

Capture raw speech faithfully according to the style guidelines established for the project. At this stage, the goal is accuracy, not polish. Flag all uncertain words or passages with [unclear] and a timestamp rather than guessing.

Step 3: Audit and Edit (Second Pass)

Listen to the audio against the draft transcript. This is where misheard terms get caught, consistent formatting gets applied, timecodes get inserted, and bracketed annotations get added or refined. The second pass is where a rough draft becomes a research-grade document.

Step 4: Narrator Review and Sign-Off

Send the transcript to the narrator (or their designated family contact if the narrator is deceased or unable to review). Allow adequate time, typically two to four weeks. Receive corrections on proper names, factual details, and any sections the narrator wants to restrict or redact. Document all changes made during narrator review.

Step 5: Archival Deposit

Deposit the final, narrator-approved transcript alongside the original audio recording, the signed deed of gift, and the consent documentation. Ensure the transcript exists in both a working format and a preservation-grade format. Submit metadata that meets the archive's requirements for searchability.

7. AI vs. Human Transcription in Oral History

Honest assessment: AI transcription has significant limitations in oral history contexts, and peer-reviewed research supports this.

A 2024 arXiv study on speech technology for oral history research found that manual correction of AI transcription output for oral history recordings "often equals or exceeds the effort equivalent to starting with manual transcription from scratch," particularly when recordings feature overlapping speakers, dialectal variation, background noise, or poor recording quality. These are exactly the conditions that define many oral history recordings, especially those made in home or community settings decades ago.

Where AI transcription falls short specifically in oral history:

• Older voices, which are often lower in volume and clarity, produce higher error rates in AI models trained predominantly on younger speech

• Regional accents and historical vernacular are underrepresented in AI training data

• Local place names, family names, and community-specific terminology get substituted with the phonetically closest common word the model knows

• Poor recording quality from older equipment or difficult acoustic environments degrades AI accuracy sharply

• Emotional content, long pauses, and non-verbal sounds are either ignored or misread

The recommended approach is a hybrid with clear boundaries:

Use AI transcription for a rapid first draft when audio quality is good, speech is clear, and the narrator has a standard accent. Use human transcription as the primary method for all other cases, and always use human review as the final quality pass before narrator review and archival deposit. For oral histories involving dialect, older narrators, difficult audio conditions, or Indigenous languages, human-only transcription is the appropriate standard. The arXiv study's finding that AI correction can exceed the work of starting from scratch should caution against assuming AI provides efficiency gains in these contexts.

Qualtranscribe handles oral history transcription with human transcriptionists matched to the narrator's language, regional variety, and subject matter context. NDA-bound transcriptionists, encrypted file transfer, and de-identification on request are standard. For multilingual oral histories, human transcription is available in 25 languages including Swahili, Amharic, French, Spanish, and Arabic, with dialect matching rather than generic language assignment. Get started here.

FAQ

Should I use full verbatim or clean verbatim for an oral history project? Most oral history archives use clean verbatim, which removes verbal clutter (repeated sounds, excessive filler words) while preserving the narrator's voice, word choices, and sentence structures. Full verbatim is appropriate when the speech itself is being analyzed linguistically or when a legal-grade record is required. Define your style before transcription begins and apply it consistently across all sessions.

Can I correct a narrator's grammar in an oral history transcript? No. Grammar correction changes the narrator's voice and is considered a rewrite in oral history standards. Non-standard grammar, dialect-specific constructions, and regional vernacular are part of the historical record. Render them in standard spelling but do not correct them.

What is a deed of gift and why does it matter for transcription? A deed of gift is a legal instrument that transfers rights to the narrator's words from the narrator to the collecting institution. It must cover both the original audio and the written transcript. Without a signed deed of gift covering the transcript, the institution doesn't have legal authorization to make the transcript publicly accessible, even if the audio is covered.

How long should I give a narrator for transcript review? Two to four weeks is standard. If the narrator is elderly, unwell, or geographically difficult to reach, allow more time. The review process is one of the few moments in oral history where the narrator has direct agency over how their words will be preserved. Rushing it undermines the ethical foundation of the practice.

What file format should oral history transcripts be deposited in? Deposit both a working file (.docx) and a preservation master (unformatted .txt or PDF/A). Plain text and PDF/A are stable formats that will remain accessible regardless of changes in word processing software. .docx files depend on software that may not be available in the same form in fifty years.

Can AI transcription be used for oral history? For clear recordings with standard accents, AI can produce a useful first draft that a human editor then refines. For recordings with older narrators, strong regional accents, dialectal speech, poor audio quality, or Indigenous languages, human transcription is more appropriate. A 2024 arXiv study found that correcting AI output for difficult oral history recordings can require more work than transcribing from scratch.

Related Reading

• Transcribing and Translating African Languages: Challenges and Solutions

• Full vs. Clean Verbatim: Which Transcription Style Do You Need

• How Transcription Services Support Academic Research and Fieldwork

• De-identification, Anonymization, and Pseudonymization: What's the Difference

• IRB Transcription Mistakes Researchers Make