EDUKATE SINGAPORE · GOLD STANDARD PRACTICAL CASE STUDIES · CASE 160
Permission to preserve someone’s recorded voice does not automatically permit creating new speech in that voice.
A fictional museum holds valid listening permissions for an oral-history interview but commissions a synthetic voice that reads words the narrator never said. This Gold Standard case separates archival access, rights, ethical participation, authentic history and new AI-generated performance.
Every narrator, museum, contract and voice demonstration is invented. The story does not determine the legal enforceability of any real voice or archival agreement, nor does it allege fraud by a named person or institution.
Chapter index
- 1. A respected teacher’s old interview begins speaking new words
- 2. An archive copy, a public recording and a synthetic performance are different
- 3. Why forty seconds of generated audio is not forty seconds of oral history
- 4. The Oral History Association places permission throughout the recording’s life
- 5. Voice cloning adds a further risk of apparent personal endorsement
- 6. The immediate correction concerns the preview and the public claim
- 7. Technological ability is not a grant of authority
- 8. A faithful alternative can still make history lively
- 9. The release system needs separate use categories
- 10. Practice: twelve authentic interviews and one unapproved digital voice
1. A respected teacher’s old interview begins speaking new words
A fictional neighbourhood museum preserves twelve recorded interviews about community education, including a conversation with retired teacher Lina. Years earlier Lina had agreed under the project documents that her original recording could be preserved, transcribed and made available to visitors for listening and historical study. The museum has honoured that arrangement for the original interview collection.
A design contractor proposes an interactive exhibition guide. It creates a forty-second demonstration using Lina’s recorded voice as input to a synthetic speech system, then makes the digital likeness read new explanatory sentences that she never actually spoke. Visitors viewing an internal preview hear a voice that resembles Lina and may reasonably assume that she approved the words.
The museum’s archive agreement does not clearly address creating a new artificial performance in Lina’s voice. The presence of a lawful archival audio file, or permission to let people listen to it, is not proof of authorisation for synthetic speech. The Gold Standard question is how to preserve the distinction between historical evidence and new imitation, and who may approve each proposed use.
2. An archive copy, a public recording and a synthetic performance are different
The museum holds O1, the original interview recording; O2, the signed release and relevant correspondence; O3, its approved preservation and access policy; O4, the contractor’s synthetic-voice proposal; O5, the generated preview; and O6, an authorised decision about further use. These records need separate interpretations.
O2 can establish permission for certain preservation, research and public-listening purposes within its actual wording. O1 provides the authentic source. O4 and O5 concern creating new speech from a model designed to imitate a distinctive personal voice. That action may raise additional consent, contractual, intellectual-property, privacy and ethical issues. The exact rights depend on the jurisdiction and documents; they cannot be decided from the word ‘donated’.
A recording is historical evidence of what Lina said on one occasion. A synthetic version may convey words that were never uttered by her, even when its sound is similar. Treating the generated audio as part of the original interview without clear explanation would misstate the source and meaning.
GS-R01 identifies the educational purpose; GS-R02 defines legitimate use conditions; GS-R08 requires a provenance trail that distinguishes original words from new content.
3. Why forty seconds of generated audio is not forty seconds of oral history
The fictional archive has twelve original interviews with relevant agreements. The contractor adds one forty-second synthetic preview. The number of original interviews remains twelve. The new preview is a separate creative artefact and should not automatically be listed as a thirteenth historical testimony from the same speakers.
The generated demonstration may read words written by the exhibition team today. For example, Lina’s authentic recording discusses how students learned science outdoors; the preview speaks a contemporary welcome and a recommendation about a new museum programme. Those new statements may be benign, but they were not part of her recorded recollection.
A transcript that faithfully represents Lina’s genuine interview would be one derivative use of the original content under the agreed terms. A new first-person speech performance goes further: visitors could infer that Lina currently endorses an initiative or remembers events described in phrases she never chose. The risk is one of attribution and autonomy as well as audio authenticity.
An editorial label or technical watermark can help identify synthetic media but cannot by itself create legal permission or resolve whether the use respects the narrator’s intentions.
4. The Oral History Association places permission throughout the recording’s life
The Oral History Association Best Practices describe preparation, informed consent, preservation and access as connected responsibilities. Projects should tell narrators about intended uses, preserve agreements and honour relevant restrictions rather than assume a completed interview gives permission for every future production.
The OHA’s manual on archiving oral history identifies rights and access metadata as part of proper stewardship. It explains that agreements, restrictions and narrator intent need to travel with recordings as new custodians and users handle them. The OHA Core Principles emphasise the collaborative relationship with the person whose life story was recorded.
These sources are professional ethics guidance, not a declaration that this specific synthetic preview is unlawful. They support the narrower operating conclusion: new uses should be assessed against the actual scope of original permission and the narrator’s relationship to the work, rather than authorised by file availability alone.
The Gold Standard treats the agreement as a live part of the collection’s metadata, not a piece of paper hidden away after recording day.
5. Voice cloning adds a further risk of apparent personal endorsement
The US Federal Trade Commission’s Voice Cloning Challenge materials describe both potential benefits of synthetic speech and the possibility of impersonation, financial harm and deceptive personal imitation. Its technical discussion of voice cloning safeguards explains that authentication and provenance measures each have limitations.
The museum is not alleged to have conducted a fraud or used the recording to target anyone financially. The risk in the fictional scene is that visitors hear a familiar voice reciting unapproved words and may attribute them to the retired teacher. A harmless-sounding sentence can still misrepresent who spoke, who approved it and what historical record exists.
UNESCO’s Recommendation on the Ethics of Artificial Intelligence encourages responsible uses of AI in cultural heritage while recognising the need to protect expression, cultural context and affected communities. Its advice is not a universal voice-rights statute. The proper decision still depends on competent governance and the narrator’s actual rights and wishes.
6. The immediate correction concerns the preview and the public claim
The museum’s responsible curator pauses unrestricted use of the synthetic voice until the relevant rights and ethical questions have been competently reviewed. Staff preserve the authorised original recording, release documents, contractor instructions, generated demonstration and any preview communications under appropriate controls.
An accurate status would say that the recording is cleared for certain archival listening purposes, while the proposed synthetic-voice performance has not been shown to be authorised under the same agreement. The curatorial team should not claim the original release included all present and future AI uses if the evidence does not establish that. It should also not automatically condemn every synthetic cultural exhibit, because some may be designed with meaningful participation and proper permission.
If Lina can be consulted appropriately, her preferences may be important under the actual arrangements; the museum should give her a clear explanation of the new purpose and the likely effect of hearing a generated voice. Any legal permission, public display or withdrawal decision belongs to the qualified responsible parties, not to a demonstration that a machine can mimic the original.
GS-R09 makes the uncertainty visible before a public exhibit could turn a new invented sentence into a purported historical statement.
7. Technological ability is not a grant of authority
The designer argues that the software can create highly realistic speech and that the archive already owns a good recording. Those are observations about technical capability and available input, not sufficient evidence of permitted use. The museum must distinguish ownership or custody of the audio file from applicable rights to create, train or distribute a new synthetic voice.
The actual release may grant a licence for recording, transcription and public access without addressing identity imitation. In another project, a narrator may agree expressly to particular synthetic uses subject to review. The Gold Standard does not declare every agreement equivalent. It asks who has authority to interpret the existing terms and approve a new proposal, and what limits should be documented.
The institution should also distinguish the right to use a person’s words from the right to present new words as spoken by them. A clear written script with a narrator’s historical facts is still newly authored. Approval, attribution and provenance need to match the form of the final product.
GS-R03 names the accountable decision maker, while GS-R07 requires that restrictions travel from archivist to designer to exhibition publisher.
8. A faithful alternative can still make history lively
The museum may have legitimate alternatives while the synthetic-voice question is unresolved. It could present approved excerpts of Lina’s actual recording with accurate transcription, use a clearly identified contemporary narrator for new explanatory text or build a visitor experience around the verified historical material and its context. Every option still needs appropriate rights and accessibility review.
The original interview remains a valuable piece of evidence without a model generating new dialogue. A visitor can hear how Lina actually described her science lessons and compare that recollection with other archival records. The distinction between quotation and interpretation can become part of the museum’s educational purpose.
If the museum later authorises a synthetic reconstruction under an appropriate process, clear labelling should still distinguish it from an authentic recording. It should not insert new invented first-person remarks into an archive catalogue as if they were part of the original conversation.
GS-R11 asks whether future staff will be able to honour the same distinction when exhibitions and technologies change again.
9. The release system needs separate use categories
The museum revises its records to identify permission for preservation, researcher access, public listening, transcription, adaptations and any proposed synthetic performance according to the actual documents. An authorised reviewer can then see which uses are confirmed and which need further assessment without publishing private terms to every contractor.
The digital asset system should flag when a new production requests use of a distinctive identifiable voice to generate new material. A routine archival preview is not automatically an acceptable proxy for that event. Contractor agreements should also carry relevant limits so technical teams do not infer broad consent from file delivery alone.
A programme with a valid archive can remain flexible about new formats when permissions and decisions are genuinely considered. The objective is not to ban innovation by default, but to prevent changing technology from erasing the narrator’s role and the meaning of the record.
GS-R12 governs controlled changes. The rights status belongs to the exact action proposed, not merely to a green checkmark beside the recording title.
10. Practice: twelve authentic interviews and one unapproved digital voice
Exercise. A museum stores twelve interviews under agreements covering preservation and public listening. A contractor creates a forty-second synthetic demonstration in one narrator’s recognisable voice, reading new sentences written today. There is no clear decision on that use in the records. What has the original release established, and what remains unresolved?
One reasoned finding is: ‘The records support the specified archival access to the authentic interviews, but do not demonstrate authority to create or distribute a new artificial speech performance in the narrator’s likeness. The preview is not an original testimony and should not be described as such. The use requires competent rights and ethical review before further release.’
The retest should challenge the asset system with an allowed listening request and a separate proposed synthetic-voice request, and require appropriate distinct decisions. The Gold Standard lesson is that keeping someone’s voice accessible for history is not the same as being authorised to make that voice speak again.
Oral-history ethics, voice cloning and further reading
- Oral History Association · Best Practices
- OHA · Archiving Oral History, Complete Manual
- OHA · Core Principles for Narrator Permission
- US FTC · Voice Cloning Challenge and Risks
- US FTC · Technical Approaches to Voice Cloning
- UNESCO · Recommendation on the Ethics of Artificial Intelligence
- Gold Standard · Information and Record Integrity
- Case 54 · Museum Photograph Licensed for Print, Not Online
- Case 113 · AI Reconstruction Labelled an Original Photograph
- Gold Standard Casebook
OHA guidelines, FTC materials and UNESCO recommendations provide professional ethical and technical context. Actual consent, copyright, privacy and use conditions depend on the governing documents and law. GS-R01–GS-R12 is eduKate’s educational management reference.