Editorial Practice · 9 min
A transcript can contain the right words and still represent the source poorly.
Spoken language carries more than vocabulary. It contains rhythm, hesitation, repetition, dialect, changes of language, relationships between speakers and clues about how something was said.
Turning that source into readable text inevitably involves decisions.
Where does a sentence end? Should a repeated phrase remain? Is a dialect form preserved or normalised? How should a language change be represented? What happens when a speaker cannot be identified with confidence?
For archival and oral-history material, these are not cosmetic details.
They determine how future readers encounter the source.
Cultural awareness in transcription therefore does not mean guessing what a speaker intended or claiming authority over the culture represented in a recording.
It means recognising where editorial decisions can change how a voice is represented — and keeping those decisions visible, controlled and separate from the source record.
Originally published July 2025 · Updated August 2026
Accuracy
Correct word recognition is essential, but it is only one dimension of transcript quality.
Context
Speaker structure, language changes, timing and relevant source context affect how the text can be understood.
Separation
Source-faithful transcription and editorial refinement should remain clearly distinguishable.
Traceability
A reader should be able to understand what comes from the recording and what was added during later editorial work.
Part I
Speech recognition answers an important question:
What words are most likely being spoken?
Archival transcription has to answer several more.
Who is speaking?
Where in the recording does the passage occur?
Does the speaker change language?
Is a phrase repeated?
Is an interruption meaningful to the structure of the conversation?
Is the wording actually clear enough to transcribe with confidence?
A recording and its transcript are therefore complementary records.
The Library of Congress notes in its oral-history guidance that even a thorough transcript cannot capture every element present in an audio recording, including tone of voice and emotional expression.
The transcript should make the recording easier to use.
It should not pretend to replace it.
A transcript is a structured representation of the recording — not a substitute for it.
Written text has structures that spontaneous speech does not always provide neatly.
Punctuation, paragraph boundaries, speaker labels and chapter breaks all require decisions.
Even a seemingly small change can alter how a passage feels.
Consider:
Source-faithful
I — I never said that.
Editorial
I never said that.
Both may communicate the same basic proposition.
But the repetition in the first version may matter to a researcher interested in speech behaviour, hesitation or the dynamics of the interview.
That does not mean every repetition must always be preserved.
It means removing it is an editorial choice, not merely a formatting operation.
The appropriate choice depends on the purpose of the transcript.
A research transcript, an archival record and a publication-ready document may legitimately require different levels of editorial intervention.
A source-faithful version may preserve:
repetitions;
false starts;
uncertain passages;
original word forms;
speaker changes;
source-aligned timing.
A reading version may improve:
punctuation;
paragraph structure;
chapter organisation;
repeated filler;
speaker naming;
presentation.
Neither approach is universally superior.
The important principle is that the reader should be able to distinguish the record of the source from the later editorial layer.
Part II
Speech does not become less meaningful because it differs from standard written language.
Dialect, regional vocabulary, informal grammar, idiosyncratic expressions and non-standard pronunciation can all be part of the historical record.
Automatically normalising them may improve readability.
It may also remove information.
For that reason, normalisation should be treated as an editorial decision rather than an invisible correction.
A source-faithful transcript can preserve the spoken form while a separate reading version provides a more conventional written presentation where the project requires it.
This allows usability to improve without silently rewriting the source.
Mixed-language recordings introduce another layer of editorial responsibility.
A speaker may change language between sentences, within a sentence or in response to another participant.
Those transitions can carry information about the conversation itself.
They should therefore not disappear simply because one language becomes the dominant language of the final document.
Transcription and translation are separate tasks.
Transcription records what was spoken.
Translation creates a derived representation in another language.
Where both are required, the original-language record should remain identifiable.
A translated or editorial version can then be linked to it as a separate layer.
This preserves access without obscuring what language was actually used in the source.
Speaker attribution can look like simple formatting:
Speaker 1
Interviewer
John Smith
But replacing an anonymous speaker label with a personal name is a factual claim.
If an identity has not been established with sufficient confidence, the transcript should not present it as certain merely because a cleaner document is easier to read.
The same applies to titles, roles and relationships.
Known information can be incorporated.
Uncertain information should remain uncertain.
A well-structured transcript therefore distinguishes between:
speaker boundaries established from the recording;
identities established from reliable project context;
provisional or uncertain attribution.
That distinction becomes especially important in historical recordings where the people who originally knew the participants may no longer be available.
Punctuation is useful because speech rarely arrives with written sentence boundaries.
But punctuation can also influence interpretation.
Compare:
You said he left.
with:
You said he left?
The words are identical.
The implied meaning is not.
For clear speech, context may resolve the difference.
For uncertain archival material, it may not.
When punctuation would require interpretation beyond what the source supports, conservative editorial treatment is preferable to creating false certainty.
The same principle applies to emphasis, irony and emotional states.
A transcript can document observable features.
It should be cautious about turning inferred meaning into fact.
Part III
Oral history is produced through a relationship between narrator, interviewer and project.
The Oral History Association describes oral history as both the interview process and the resulting recorded product, and emphasises the importance of narrator perspective, documented agreements, preservation and access.
This means the transcript exists within a larger documentary context.
Relevant information may include:
project purpose;
interviewer and narrator;
recording date;
access restrictions;
consent or release documentation;
collection context;
editorial history.
Not all of this information belongs inside the transcript itself.
But separating the transcript from that context can make future interpretation more difficult.
Making a transcript searchable or available online does not automatically answer the question of how the underlying material may be used.
Oral-history projects can include restrictions, agreements and expectations concerning access or publication.
The Oral History Association recommends that repositories respect applicable restrictions and preserve enough documentation to understand the terms under which interviews were created and made available.
Narrator review, informed consent and agreed conditions for public release are part of that same responsibility, as set out in the Oral History Association’s Statement on Ethics.
Technology can help implement access and organise documentation.
It should not decide those conditions on behalf of the institution, narrator or rights holder.
Human review adds the most value where interpretation, uncertainty or editorial judgement becomes important.
Not every minute of a long recording requires identical manual attention.
Diagnostic analysis can help identify passages involving:
weak or unstable recognition;
speaker-boundary uncertainty;
language transitions;
overlap;
difficult source conditions;
passages selected for editorial review.
This allows technical and human quality control to be concentrated where it has the greatest value.
Efficiency and editorial responsibility are therefore not opposites.
A structured workflow can reduce unnecessary manual effort while preserving deliberate human review for the decisions that genuinely require it.
Preserve
Do not silently normalise source language simply because another form reads more smoothly.
Separate
Keep source-faithful transcription, translation and editorial adaptation as distinguishable layers.
Attribute
Treat speaker names, roles and identities as claims that require evidence.
Mark
Keep uncertainty visible where the recording does not support a confident decision.
Document
Make significant editorial interventions and version relationships understandable.
Review
Direct human attention toward passages where interpretation or context materially affects the result.
Cultural awareness in transcription is not the ability to know every cultural meaning. It is the discipline to recognise where editorial intervention can change the record.
R2 Mechanics separates transcription evidence from later editorial presentation.
The source-faithful baseline remains connected to the original recording through timing, speaker structure and documented uncertainty.
Where a structured reading or publication version is required, it is produced as a separate project output.
Language changes and speaker attribution are treated as analytical dimensions rather than formatting details.
Diagnostic views can identify regions where recognition, speaker structure or language conditions require additional attention, allowing expert review to focus on the passages where editorial decisions matter most.
R2 Mechanics does not claim to determine the cultural meaning of a recording on behalf of an archive, researcher or community.
Our role is more concrete:
to preserve the relationship between source and transcript, make uncertainty visible and provide a structure in which informed editorial decisions can be made.
Before finalising a transcript, ask:
Source — Does the text remain connected to the original recording?
Language — Have dialect, non-standard language or code-switching been altered — and if so, deliberately?
Speaker — Are identities clearly established, or has uncertainty been turned into certainty?
Editing — Can the reader distinguish transcription from later editorial refinement?
Translation — Is translated text clearly distinguishable from the original-language record?
Context — Does the transcript remain connected to the project and collection information needed to understand it?
Uncertainty — Are unclear passages visible?
Purpose — Is the level of editorial intervention appropriate for how the transcript will actually be used?
Good transcription is not a competition to produce the smoothest possible text.
Nor is fidelity achieved simply by preserving every hesitation without considering the purpose of the document.
The stronger approach is to know which layer is doing what.
The source recording remains the evidence.
The source-faithful transcript makes that evidence searchable and reviewable.
Editorial work can improve readability and access.
Translation can extend the audience.
Context can help future users understand what they are reading.
Each adds value precisely because it remains identifiable.
Precision is not only getting the words right. It is knowing what came from the source, what was interpreted later and where uncertainty still remains.
From Source to Usable Record
If your collection includes complex speaker structures, mixed languages, oral-history interviews or material where editorial decisions affect how the source will be understood, we can assess the recording conditions and define an appropriate transcription and review workflow.
A short description of the material and its intended use is enough to begin.
David Thiry is the founder of R2 Mechanics and works on offline-first systems for difficult speech and archival audio. His work focuses on transcription, speaker analysis, source traceability and structured delivery for archival, research and institutional material.
Originally published July 2025 · Updated August 2026
Why structured transcription can improve access while preserving the relationship between written text and the original voice.
Read Article →How difficult recordings become structured, source-linked and reviewable archival records.
Read the Guide →How processing architecture affects external data paths, institutional control and sensitive recordings.
Read Article →