David Thiry is the founder of R2 Mechanics, a developer and systems architect focused on offline‑first, privacy‑driven solutions for archives, museums, and research institutions. With a background in building high‑performance, autonomous computing systems, he works at the intersection of technology, ethics, and cultural preservation. This article reflects his vision of transforming transcription from a static service into resilient cultural infrastructure.
Published on: August 03, 2025
This three‑part series explores how cultural institutions can turn fragile oral histories into living archives. Discover offline‑first transcription workflows, ethical standards, and practical tools like WhisperX, speaker diarization, and multilingual transcription to make voices accessible, searchable, and preserved for generations.
In a world that celebrates the digital preservation of culture, it is easy to assume that the voices of the past are being carefully transcribed and stored, ready for future generations to explore. The truth, however, is far more complex. Beyond the flagship projects that make headlines—those glossy initiatives of national libraries and well-funded institutions—lie thousands of archives and museums where memory remains trapped: on tapes, reels, and aging hard drives. These voices are not silent because they lack value. They are silent because no one can hear them.
The problem is not a lack of interest. In conversations with archivists from Eastern Europe to the Middle East, one theme emerges over and over again: they want to preserve these recordings, to transcribe them, to make them meaningful. But between ambition and reality yawns a wide gap. Most archives do not have the means—technical, financial, or infrastructural—to turn transcription into a living, usable part of their collections.
Imagine walking into the Oral History Archive of the Memorial to the Revolution in Timișoara, Romania. Decades of recorded testimonies from survivors of 1989’s political upheaval remain stacked in boxes—some labeled, others anonymous. Or consider the Polish Institute of National Remembrance (IPN), which holds thousands of hours of recorded witness accounts of political persecution and social transition, many of which have yet to be transcribed and catalogued.
These are not isolated cases. In Lebanon, the National Archives preserve extensive Ottoman-era registers and administrative records, yet transcription and access remain restricted by complex legal frameworks and resource limitations. In Ukraine, regional repositories such as the State Archives in Kharkiv (Державний архів у Харківській області) face severe infrastructural challenges—from outdated hardware to limited internet connectivity—making large-scale transcription projects nearly impossible.
In Syria, local organizations have documented oral testimonies of the civil conflict—fragile records of lived experience, often collected at great personal risk. Yet these recordings remain locked away on external drives or encrypted in personal devices, neither transcribed nor preserved in a way that makes them accessible to researchers or the communities they represent. In Oman, small cultural heritage projects working to preserve endangered dialects face similar obstacles. Without technical expertise or the ability to engage with expensive, cloud-based tools, their efforts remain trapped at the level of raw data.
Even in places like the United States or Germany, where resources exist, the problem is not solved. Instead, it takes a different form: populist prioritization. Funding often flows to projects with visibility or immediate cultural cachet—digitizing pop culture archives, celebrity interviews, or other media with high public appeal—while the deeper, quieter records of marginalized communities languish, underfunded and overlooked. The result is a global archival landscape where the loudest voices are preserved, and the most vulnerable remain unheard.
When asked why transcription rarely becomes part of their workflows, archivists often respond with a quiet litany of limitations. There are not enough hands to do the work. In many institutions, a single staff member is tasked with caring for an entire audio collection spanning decades. Budgets are already stretched thin; transcription projects, especially those requiring high-quality or multilingual processing, feel like an unattainable luxury.
The problem is compounded by technology. Most commercially available transcription tools are cloud-based, which poses immediate legal and ethical issues for institutions bound by strict data protection laws such as the GDPR—especially when dealing with special categories of personal data as defined in Article 9, including political opinions, religious beliefs, or health-related testimonies. Cloud-based transcription tools can be legally sensitive for highly sensitive oral testimonies—especially under GDPR Article 9 (e.g. political opinions, religious beliefs). While cloud can be GDPR-compliant under strict controls (e.g. Art. 28 data processing agreements), for many archives this remains a liability rather than an option.
Even when transcription does happen, it often arrives as little more than a Word document or a PDF. These files may fulfill the basic function of converting sound into text, but they do not constitute archives. They cannot be integrated into digital catalogues, cannot be searched or annotated in meaningful ways, and cannot stand as robust, future-proof records.
In other words, transcripts without structure are dead data. They exist, but they do not live.
This disconnect between transcription and archiving reveals a deeper, more troubling dynamic: a global inequality in cultural memory. When the British Library or the Library of Congress transcribes its recordings, they enter a well-funded ecosystem of digital preservation. They become searchable, citable, and part of the global scholarly conversation. When a grassroots project in Damascus or a regional museum in Romania does the same—without the means to integrate their work into archival standards—those voices remain invisible.
The implications are far-reaching. Whose histories become part of our shared record, and whose are lost to time? Who decides what is worthy of preservation? In the current landscape, the answers to these questions often have less to do with cultural significance than with access to infrastructure and funding. It is a quiet but profound form of cultural gatekeeping.
There is an uncomfortable truth at the heart of this discussion: transcription technology is not the issue. The tools exist, and they are powerful. What is missing is the ability—and in some cases, the will—to use them effectively. Archives and museums are left stranded between the need to preserve memory and the absence of the workflows, standards, and legal assurances that would allow them to do so.
As long as transcription remains a disconnected service, purchased piecemeal and delivered without thought to long-term usability, the problem will persist. The result is a world where countless hours of human testimony, cultural expression, and historical record remain locked away—technically preserved, but effectively lost.
If this seems bleak, it is because the stakes are high. Yet this is only the beginning of the conversation. In the next part of this series, we will look closely at what it takes to transform transcription from mere text into archive-ready knowledge—leveraging established archival frameworks such as METS (Metadata Encoding and Transmission Standard), which captures complex structural and descriptive metadata; EAD (Encoded Archival Description), which provides context and provenance for archival materials; and IIIF (International Image Interoperability Framework), which enables interoperability and access across platforms. Combined with robust metadata and timestamping, these standards turn loose words into lasting, interoperable records.
And in Part III, we will explore how R2 Mechanics bridges this gap—offering offline-first, modular solutions that allow institutions around the world, from major libraries to grassroots initiatives, to reclaim their voices and make them part of the global record.
For now, one truth must guide us: preservation without access is not preservation at all. Until transcripts are transformed into living, usable archives, the voices they contain will remain unheard.
If Part I revealed why so many archives cannot use transcription services, this part asks the next critical question: what does it actually take to make a transcript “archive-ready”?
The difference between a raw transcript and an archival asset is the difference between a stack of unbound pages and a meticulously catalogued volume in a library. One is text. The other is knowledge.
Turning spoken words into enduring records is a process—one that requires more than transcription software. It demands standards, metadata, workflows, and long-term thinking. Without them, the best-intentioned transcription projects collapse into unusable fragments, disconnected from the cultural and historical ecosystems they are meant to enrich.
Consider a grassroots NGO in Damascus documenting the stories of civilians displaced by war. Volunteers, often working under dangerous conditions, record hours of testimonies. They run them through free transcription tools. What they end up with are dozens of Word documents—each a fragile, static piece of text.
Can a researcher in Berlin access them? Can a digital heritage project in Paris integrate them? Can they be linked to the original audio for verification or annotation? The answer, in most cases, is no.
Raw transcripts fail because they lack structure. They do not carry the metadata that archives need—no timecodes to synchronize with audio, no information about speakers, no consistent encoding for multilingual content, and no context about the conditions under which the data was collected. They are frozen snapshots of speech, incapable of becoming part of a dynamic, searchable archive.
What turns unstructured text into cultural infrastructure is not just software, but standards. These frameworks provide the common language that allows archives, libraries, and researchers to share, search, and sustain their data over decades.
Some regional archives are already doing this. The Central State Historical Archives of Ukraine has begun implementing structured metadata frameworks like METS and EAD for parts of its oral history collections, while the Armenian Genocide Museum-Institute in Yerevan applies structured archival description to preserve survivor testimonies. These examples show that adopting robust frameworks is possible even for mid-sized institutions with limited resources.
METS (Metadata Encoding & Transmission Standard) allows complex digital objects—transcripts, audio, images—to be described, organized, and linked as one coherent archival package. Practically, this means that a single METS document can outline the structure of an interview, its transcript, the associated audio file, and contextual metadata such as speaker information or thematic keywords.
EAD (Encoded Archival Description) provides a structured way to represent finding aids, helping researchers understand where a transcript sits within a larger collection. Key EAD elements like <accessrestrict> make it possible to record access limitations for sensitive content, while <note> fields can provide context on redactions or sensitive topics.
IIIF (International Image Interoperability Framework), though often used for images, also offers powerful options for delivering synchronized multimedia content. The British Library’s Endangered Archives Programme and the Bodleian Libraries at Oxford are among the institutions that have integrated IIIF into their digital platforms, enabling cross-institutional access to heritage materials in interoperable viewers.
These are not buzzwords; they are the architecture of trust. They ensure that a transcript created in Oman today can still be opened, understood, and cross-referenced by a researcher in Toronto twenty years from now.
Standards are meaningless without metadata—the rich, descriptive information that gives a transcript context.
At minimum, an archive-ready transcript should contain:
For materials falling under Article 9 of the GDPR, archives should embed explicit metadata about consent, anonymization levels, and access restrictions. Best practices include obtaining informed consent in advance, securing related documentation, restricting access to sensitive content, and conducting Data Protection Impact Assessments (DPIAs) to assess potential risks.
The archival world cannot afford to build its future on commercial clouds. For many institutions—especially in regions with restrictive data laws or fragile political contexts—offline-first solutions are the only viable path.
Tools inspired by field platforms like the KoBo Toolbox exemplify this approach: data is collected and stored locally, under institutional control, before any optional syncing with central repositories. This protects sensitive testimony, keeps archives compliant with national sovereignty requirements, and shields collections from the volatility of commercial platforms.
Offline-first does not mean technologically backward. It means self-contained, secure, and sovereign. It means that a regional museum in Eastern Europe or a cultural center in Oman can process, enrich, and store its transcripts without depending on infrastructure that may not align with its legal or ethical obligations.
How do these elements come together? A simple archive-ready pipeline looks like this:
This is how raw words become living archives.
The lesson is clear: transcription must move from being a project-based service—a one-off purchase from a vendor—to being a component of archival infrastructure. This requires investment not only in technology, but in workflows, staff training, and partnerships that bridge the gap between transcription providers and archival ecosystems.
For too long, transcripts have been treated as the end product. In reality, they are the beginning: raw material waiting to be structured, contextualized, and linked to the living networks of cultural memory.
In the final part of this series, we will explore how R2 Mechanics bridges this gap—bringing modular, offline-first workflows to institutions that cannot or will not rely on commercial cloud services. We will look at how small museums, NGOs, and large cultural institutions alike can turn their recordings into archive-ready, interoperable collections—without sacrificing security, sovereignty, or sustainability.
Because in the end, transcription is not about words on a page. It is about making voices live—safely, contextually, and for the long term.
In the first two parts of this series, we examined why so many archives and museums cannot use transcription effectively, and what it takes to make transcripts truly “archive-ready.” Yet one question remains: how do we close this gap in practice?
Bridging ambition and reality requires more than software. It requires partnerships, workflows, and a shift in how we think about transcription itself—from being an isolated service to becoming part of an integrated cultural infrastructure. This is where R2 Mechanics positions itself: as a technical enabler for institutions, providing offline-first, modular solutions that allow even the most resource-limited archives to transform their recordings into living, interoperable repositories of knowledge.
Let us begin with the hard facts. According to HURIDOCS, the Syrian Oral History Archive (SOHA), launched in 2016, has collected approximately 400 testimonies between 2016–2018 across Syria, Lebanon, and Jordan1. According to archivists involved in the project, the collection has grown significantly, yet transforming these testimonies into structured, accessible records remains an ongoing challenge2.
In May 2024, the National Archives of India, in collaboration with Oman’s National Records and Archives Authority (NRAA), digitized over 7,000 documents of Indian diaspora communities in Oman — including previously unrecorded oral histories — reflecting the voices of tribal and local communities3. Archivists emphasize that while these tribal narratives are preserved offline, the absence of standardized metadata and archival formats significantly limits their reusability.
Poland’s Institute of National Remembrance (IPN), established in 1998, is responsible for extensive collections of witness testimonies. Its public Chronicles of Terror database includes over 500 digitized testimonies from World War II survivors, with broader holdings extending significantly beyond the public database4. According to staff at the institute, the main challenge lies not in the sheer volume of material, but in transforming these recordings into searchable, interoperable collections.
Yet these collections share a common issue: much of the material remains raw, static, and structurally disconnected from wider archival ecosystems.
For archives dealing with sensitive material—oral histories of conflict, testimonies of persecution—cloud solutions are often legally and ethically untenable. Article 9 of the GDPR strictly regulates the processing of personal data revealing political opinions, religious beliefs, or health information. Using cloud-based transcription also carries other risks: losing control over where data is stored (and potential violations of data sovereignty laws), exposure to breaches due to insufficient provider security, and vendor lock-in that makes long-term migration difficult.
R2 Mechanics addresses these challenges head-on with an offline-first architecture. All transcription and metadata enrichment occur locally within the institution’s infrastructure. Modular workflows ensure that data never leaves controlled environments, maintaining full compliance with GDPR, local data laws, and institutional ethics guidelines. Outputs are delivered in standards-compliant formats—METS, EAD, and IIIF—ensuring long-term usability beyond any single vendor or platform. What begins as a legal constraint becomes a strategic advantage: archives retain sovereignty over their collections while adopting modern, AI-driven processing pipelines.
R2 Mechanics does not deliver “just transcripts.” It provides integrated archival assets through a scalable, modular pipeline:
To support this, open-source tools like Archivematica and Islandora (Fedora Commons) provide flexible, standards-aligned environments that can integrate METS, EAD, and IIIF in one workflow. This ensures that even smaller archives can adopt robust, future-proof infrastructures.
This approach transforms archives from passive repositories into living cultural infrastructures. A grassroots project in Damascus can safeguard testimonies in a sovereign, structured archive. An Omani heritage center can present its oral traditions in interoperable formats for global research. And the IPN can convert static documents into searchable, enriched collections without sacrificing sovereignty. Structured, standards-driven archives are more than collections; they are active participants in the global memory network.
Perhaps the most urgent lesson is this: transcription cannot remain a one-off project. It must become infrastructure—embedded in how archives think about collecting, processing, and sharing cultural memory. R2 Mechanics offers this not as an abstract vision, but as a practical, implementable model: modular, offline, standards-driven. A bridge between what archives aspire to and what they can actually achieve.
Because in the end, this work is about more than technology. It is about voices—making sure they are heard, respected, and preserved as part of our shared human record.
1 HURIDOCS – Syrian Oral History Archive
2 HURIDOCS Statement, 2023
3 National Records & Archives Authority – Oman
4 Institute of National Remembrance – Poland
