Imperial War Museums (IWM) has used generative AI to transcribe, translate, and make searchable more than 20,000 hours of oral testimonies from its enormous archive, in partnership with Capgemini and Google Cloud.
The project uses Google’s Gemini models and was developed by Capgemini on the search giant’s cloud infrastructure.
The result is a resource that allows the public, researchers, and educators to search, explore, and interact with personal accounts of 20th-century conflict, many of which were previously locked away in audio-only formats.
IWM runs five branches across England, including the primary Imperial War Museum at the former Bethlem Royal Hospital in London and the Churchill War Rooms in Westminster.
IWM’s oral history collection spans more than 8,000 interviews with servicemen and women between 1945 and the 2000s.
While the recordings offer first-hand accounts of war and its aftermath, they have historically been difficult to access due to inconsistent audio quality, strong regional accents, military terminology and the lack of written transcripts.
“This project is a big step forward in our mission to broaden access to our vast collections,” said Nick Hodder, director of digital engagement and transformation at IWM.
“Our expert curators have been fully involved in this work, ensuring the technology delivers very high levels of accuracy.”
The partners claim the solution transcribe audio with 99% word accuracy and 94% speaker diarisation (partitioning audio according to the identity of the speaker).
It also extracts detailed metadata, such as names of individuals, places and military units, and generates written summaries to help users navigate the content.
A new “ask a question” function enables users to query any recording in natural language and receive accurate, citation-backed answers.
The partnership claims that this task would have taken more than two decades to complete manually, but this has been condensed into a matter of weeks.
“This project showcases the profound impact of generative AI in unlocking historical archives and making them accessible in new and engaging ways,” said Steven Webb, UK CTIO at Capgemini. “It’s a testament to how technology can connect us more deeply with our past to inform our future.”
John Abel, managing director, office of the CTO at Google Cloud, said: “The use of Gemini models to process and understand such a vast and nuanced audio collection demonstrates the sophisticated capabilities of generative AI to overcome complex challenges and deliver meaningful outcomes.”
“This initiative ensures that the voices of those who lived through conflict are preserved, understood, and more accessible than ever before,” added Hodder.