Transcribing Voice Chat from Past Virtual World Gatherings: Consent, Accuracy and Names

Voice recordings preserve a different kind of world history

Virtual world gatherings were not always conducted through typed messages. Some communities used voice communication for meetings, celebrations, collaborative activities and informal conversation. These recordings can preserve tone, pauses, interruptions, laughter and the dynamics of group participation. They may also reveal how communities organised events and communicated across different regions and time zones.

For a digital historian, the recording is valuable because it captures characteristics that a conventional chat export cannot represent. A written message may show the words exchanged, but audio can also preserve hesitation, emphasis and simultaneous speech. These details can be important when documenting the atmosphere of an event or examining how participants interacted.

The presence of an audio file does not establish that every participant expected permanent preservation. A gathering recorded for a private community archive may have been intended for a limited audience. The original circumstances of recording and any subsequent agreements should therefore accompany the file.

Consent begins with the circumstances of recording

Consent needs to be considered across the life of a recording. Permission to participate in a voice gathering is not necessarily permission for someone else to record it. Permission to record is not automatically permission to upload the audio to a public repository, create searchable transcripts or publish identifiable extracts.

Older virtual world recordings present particular difficulties. The person who created the file may no longer be contactable, and the original participants may have used pseudonyms. Existing announcements, community rules, recording notices and correspondence can help establish what participants were told. If no documentation survives, the archive should record that uncertainty.

The Oral History Association's guidance emphasises the importance of agreements, contextual documentation and preserving access restrictions alongside recordings. The Oral History Society also recommends risk assessment and transparency when AI tools are used with oral history material. These principles provide useful starting points for historical voice chat collections, although an old multiplayer gathering is not automatically an oral history interview. (Oral History Association, Archiving Oral History: Manual of Best Practices; Oral History Society, AI guidance, 2020s.)

• Establish who recorded the gathering and why.
• Identify any recording notices or participation agreements.
• Separate permission to preserve from permission to publish.
• Record restrictions and unresolved consent questions.
• Consider whether public access could expose participants or third parties.

Preparing audio without changing its historical content

Start by preserving the original audio file in its acquired form. Record the filename, format, duration, source, acquisition date and any available technical metadata. Where the recording has been transferred from another medium, retain information about that process. Create a working copy before adjusting levels, removing background noise or converting the file for transcription.

Listen to the complete recording before generating a transcript. This initial listening helps establish the number of apparent speakers, the type of event, the presence of music or game sounds and the points at which audio quality changes. It can also reveal whether the recording contains private conversation, personal information or material that may require restricted access.

If the recording is split into several files, preserve the original sequence and document any gaps. A missing section should not be filled by joining unrelated material or using an automatically generated summary as a substitute for the original conversation.

Using AI transcription as a first draft

Automated speech recognition can convert spoken language into searchable text. This can reduce the initial labour involved in cataloguing long recordings, especially where a collection contains many gatherings. Its usefulness depends on audio quality, microphone conditions, language, accents, vocabulary and the number of simultaneous speakers.

Virtual world recordings can be especially difficult. Participants may speak through inexpensive microphones, use usernames unfamiliar to general speech models, discuss invented place names or talk over one another. Background music and sound effects can also be mistaken for speech. A transcription system may produce a plausible sentence even when the underlying audio is unintelligible.

The Oral History Society warns that AI-generated transcripts can contain misleading material and recommends checking them against the original audio. The same principle applies to archived gaming and virtual world recordings. An automated transcript should be identified as a draft until a suitable human review has taken place. (Oral History Society, After the Interview and AI guidance.)

• Retain the original recording before transcription.
• Generate a separate machine-assisted transcript.
• Review the complete recording against the text.
• Mark uncertain speech and overlapping speakers.
• Keep a record of corrections and editorial interventions.

Speaker names, usernames and uncertain attribution

Names are among the most consequential parts of a transcript. A speech recognition system may interpret a username as an ordinary word, merge two speakers or assign a sentence to the wrong person. A participant might also have used different usernames at different points in the community's history.

Speaker attribution should rely on evidence from the recording and associated records. A participant introducing themselves, an accompanying text-chat log or a contemporaneous event programme may help establish an identity. An editor should not assign a real-world name to an anonymous voice merely because a later recollection suggests who it might have been.

Where identity is uncertain, use consistent neutral labels such as Speaker 1 and Speaker 2. If a participant's username is established but their real identity is not, retain the username according to the collection's privacy and access policy. If several voices overlap, indicate the overlap instead of forcing the exchange into a misleading sequence.

For example, a transcript might identify a speaker as Username_A where the audio and accompanying event records support that attribution. A second voice could remain Speaker 2 if no reliable identification exists. This preserves useful information without turning a plausible guess into a historical claim.

A transcription convention for difficult passages

A consistent transcription style helps readers understand how the transcript relates to the recording. Use timestamps at regular intervals and around important changes in speakers or subject. Preserve meaningful pauses, interruptions and non-verbal sounds where they contribute to the historical record. Do not silently rewrite informal speech into polished prose.

Uncertainty should be visible. A word that cannot be understood can be marked as [unclear], with an approximate timestamp. An uncertain interpretation may be marked [possibly: ...] if the collection's transcription policy permits it. The distinction matters because an editor should not present an interpretation as a directly heard word.

The International Council on Archives' Guide 8, published on 29 June 2022, treats transcription as an archival document and discusses the importance of transcription guidelines and automatic transcription technology. A project should adopt a consistent convention before preparing a large collection. (International Council on Archives, Guide 8: The Transcription as an Archival Document, 2022.)

• [00:02:14] Speaker 1: The meeting starts here.
• [00:02:19] Speaker 2: I think the eastern room is open.
• [00:02:24] [overlapping speech]
• [00:02:27] Speaker 1: [unclear]

Publication, privacy and long-term access

A transcript makes a recording easier to search, quote and circulate. That accessibility can increase the privacy consequences of material that was previously difficult to locate. A voice file buried in an old archive may receive little attention, while a searchable transcript can expose names, personal details and remarks to a much wider audience.

An archive should assess whether personal information, third-party references or sensitive exchanges require restricted access or redaction. The approach should consider the circumstances of the original conversation, any applicable agreements and the intended use. The Oral History Society and Oral History Association both emphasise documenting restrictions and preserving the relationship between audio, transcripts and supporting records.

AI processing adds another layer. Before uploading an archive to an external transcription service, examine its data retention, reuse and access conditions. Sensitive material may require local processing or an institutional service with appropriate safeguards. Record the service used, the processing date and whether the audio was transferred outside the archive's own systems.

The transcript is an access copy, not a replacement recording

A good transcript helps researchers find relevant passages without requiring them to listen to every minute of a recording. It should also make clear that speech recognition, editorial punctuation and speaker labels are interpretative stages. The audio remains the primary evidence for the spoken exchange.

The next practical step is to create a transcription record for each gathering containing the original audio identifier, permission documentation, transcription convention, timestamps, speaker attribution status and a list of unresolved passages. Keep the transcript and recording linked in the archive catalogue.

Questions readers ask