Image credit: (a) Sophie in a radio station, 1990s Shenzhen: This image is a digital restoration of an original photograph provided by Sophie Xi, depicting her at work during her late-night radio show. Used with the permission of Sophie Xi. (b) Participant engaging with digital Sophie’s voice: Photographed by Tongge Yu, this image portrays a participant interacting with the recreated digital persona of Sophie. The glowing mobile phone symbolizes the connection between the past and the present.
At SIGGRAPH 2025, the Art Papers program highlighted projects that push the boundaries of creative research, including “Talking to the Midnight Broadcast: Reviving 1990s City Memories With AI.” Artists Fan Xiang and Tongge Yu combined historical radio recordings, personal diaries, and modern AI techniques to reconstruct the voice and presence of a late-night Shenzhen radio host, creating an interactive experience that bridges past and present while exploring the city’s cultural memory.
SIGGRAPH: Your project reconstructs a beloved 1990s Shenzhen radio program through AI. Can you share how you approached blending historical recordings with modern technology to bring this broadcast back to life?
Fan Xiang (FX): In spring 2024, Tongge and I were at Tsinghua University as advisor and student, and everyone around us was talking about ChatGPT and training models. Around the same time, a close friend posted a box of cassette tapes from the 1990s. She used to host a late-night call-in radio show in Shenzhen. I found myself wishing the show still existed so people could call again and talk to the same voice they once listened to. We started by digitizing and restoring the tapes, then treated them as a time capsule and built an experience where people today can have a conversation with a recreated version of the host. In a way, the project is also a collaboration across generations. I was born in the 1970s, and Tongge was born in the 2000s.
Tongge Yu (TY): I was immediately drawn to the tapes because Shenzhen has been meaningful to both of us. We each spent our early 20s there as newcomers and remember its particular energy. Listening to Sophie’s conversations, I heard people talking about careers and ideals, but also everyday topics like relationships, pop culture, food, and travel. Many of those concerns are still with us today. Our approach was to keep those original voices and themes as the backbone, then use current tools to make the archive conversational, not just something to be stored and forgotten.
SIGGRAPH: The digital twin of Sophie is central to your work. What were the creative and technical considerations in authentically capturing her voice, personality, and presence?
FX: Sophie also shared diaries and short essays from the same period, which helped fill in what the tapes could not capture. As Sophie’s friend, I was surprised by how closely the reconstructed voice matched her. At the same time, the personality is not the same as the person I know. The digital Sophie can be endlessly patient and will answer anything, without shyness, refusal, resistance, or avoidance. That gap pushed us to face a question we did not start with: What makes a person feel real beyond a voice?
TY: We focused on two core parts: voice and memory. For voice, we aimed to capture timbre, rhythm, and the small details that make a radio host sound present. For memory, we built a curated base from the tapes and the writing Sophie shared, so the digital Sophie could respond in a way that stays consistent with the 1990s context. We also designed constraints so it does not drift into experiences or facts that belong to today.
SIGGRAPH: How did working with ephemeral, local media from decades ago shape your understanding of the city’s cultural memory, and what did you learn in the process?
FX: Talking by voice with a host from her younger years feels similar to chatting in a studio or listening to a call in show, but it is also clearly different. The digital Sophie does not have a smartphone, and the Shenzhen she describes is still a young city just beginning to grow. That contrast becomes a way to sense the boundary between past and present. The boundary is geographic, cultural, and also historical.
In recent years, many city memory projects in China focus on a much longer timeline, often a century or more. But urban culture from roughly the 1940s to the 2000s is often outside formal preservation. Many buildings have already been erased by rapid urbanization, and the young people from that era are now older. If radio and television were major media of city memory for that period, then what becomes the medium of city memory in 2025?
TY: After we built the digital Sophie, we shared the demo with friends who also have personal ties to Shenzhen. Their reactions were unexpectedly emotional. One person said it reminded him of stories about his parents struggling and building a life there in the 1990s, and it made him wonder how people at the time imagined the future. It reinforced something for me: Personal viewpoints are limited, but they are also vivid. A history book can summarize the 1990s in a sentence, but an individual can paint a scene.
SIGGRAPH: What role do you see AI playing in preserving or reinterpreting intangible heritage, and how does your project challenge conventional ideas of archival and memory?
TY: AI can be useful for preserving and reinterpreting intangible heritage, especially by turning an archive into something people can interact with. At the same time, the risk is that generated content can sound authoritative even when it is incomplete or wrong.
For our project, we see it as a new way to approach a piece of history, but it is subjective, not truth. I do not see it as a challenge to traditional archives, more as another perspective. AI works by learning patterns from existing materials, it does not create history, it recombines what is already there.
SIGGRAPH: For audiences experiencing your work, what is the emotional or intellectual response you hope to evoke, and how does it connect past listeners with present-day technology?
FX: I hope the work invites people to talk rather than be taught. For younger audiences, especially Gen Z in Shenzhen, a conversation can be more engaging than being told what the city used to be. At the same time, many people who lived through the 1990s respond with real excitement. They are eager to speak with the past again, and to hear those midnight voices through the device in their hands.
SIGGRAPH: What guidance would you give to artists and researchers who are interested in submitting to SIGGRAPH Art Papers or attending an Art Papers session at SIGGRAPH 2026?
FX: I have published three papers in SIGGRAPH Art Papers, and each time I treated the writing process as telling a story to a close friend, explaining how the work came to be. I have also served as a reviewer. When reviewing, I care a lot about motivation: Why did the artist need to make this project? The technical execution matters, but in art, the personal reason behind it matters just as much.
TY: This was my first SIGGRAPH Art Papers submission. My background spans mechanical engineering and product design, and I hope more researchers from other fields will share their work as well. The community benefits when people bring in different methods, languages, and ways of making.
Inspired by innovative storytelling like this? Submit your work to SIGGRAPH 2026 Art Papers and bring your ideas to life.
Xiang Fan is a research-driven visual artist and educator whose practice intersects visual art and the humanities through storytelling with data. Confronted by the complexity of the world, she often finds herself in a state of confusion, seeking trustworthy answers through AI and data visualization—answers that, at times, turn out to be unspoken common knowledge for a few. Her work reflects a persistent curiosity about perception, truth, and the quiet narratives hidden within systems.
Tongge (Melody) Yu is a graduate student in Mechanical Engineering and a MAD Design Fellow at MIT. Her work bridges design and engineering through interactive prototypes and research on how emerging AI systems can shape communication, memory, and experience. She previously studied design at Tsinghua University, where she worked with Prof. Xiang Fan on data visualization and AI applications.
