Elizabeth Chai Vasarhelyi, Jimmy Chin, National Geographic's Endurance, Respeecher · September 13, 2026
Endurance Reconstructed Shackleton's Voice Without Inventing the Performance
IBC 2026 · Technical Presentation
What happened
IBC's AI in Practice session highlights the Endurance workflow in which Respeecher reconstructed Ernest Shackleton and several crew members' voices from extremely limited century-old recordings, including wax-cylinder material. Voice actors performed the diary passages; AI transformed those performances toward the historical voices. Some source recordings were under a minute and the Shackleton model reportedly required roughly a month of training, followed by performance recording, editorial refinement and human review.
Why it matters
This is a useful boundary between synthetic voice and synthetic performance. Historical writings supplied the words, actors supplied timing and emotion, filmmakers directed the sequence, sound artists refined it, and ML performed the identity transformation.
Antique Robots take
This is a much more interesting use of AI than typing "make Shackleton talk." The performance is still performed. The filmmakers still choose the words, timing and emotional intent. The model solves an otherwise impossible transformation problem. That feels less like replacing a craft and more like creating a new post-production technique — provided the audience is clearly told what they're hearing.