E01 AI Revives Nightbirde's Final Songs : An Interview with Producer Geoff Duncan
In this episode of Ryan Meets World, producer Geoff Duncan opens up about his collaboration with the late artist Nightbirde — from their synth-pop beginnings to the painstaking use of AI to finish her unreleased songs after her passing. What emerged wasn’t just music, but a legacy project that blended ethics, innovation, and enduring love.
Geoff Duncan on AI, Creativity, and Ethics
In February of 2025, I received one of the hardest phone calls of my life: my mom was in the hospital, and doctors told her she had cancer. Just days later, another story of cancer and loss collided with my world in an unexpected way. I read a Rolling Stone article about Nightbirde — the luminous singer-songwriter who captured the world’s heart on America’s Got Talent before passing away in 2022 — and a posthumous album of hers that had just been released.
This wasn’t just a reissue or unearthed demo tape. It was music finished after her passing, made possible through the delicate, painstaking use of AI by producer Geoff Duncan. At the same time I was facing cancer in my own family, I found myself deeply connected to this story: art continuing after death, technology helping carry forward a voice silenced too soon.
Nightbirde, born Jane Marczewski, was more than her music. She was a force of warmth, joy, and resilience — her viral “Golden Buzzer” performance on America’s Got Talent became an emblem of hope around the world. Even in her absence, her songs still radiate the message she gave so freely: you are loved, you matter.
Ethics
To revive Nightbirde’s unfinished songs, Geoff Duncan leaned not on fully generative AI — not on “make this in her voice” shortcuts — but on careful restoration. “I wasn’t fabricating performances of Jane’s,” he explained. Instead, he worked from the raw material Jane herself left behind: iPhone recordings, voice memos, rough vocal takes. Using AI tools as restorative brushes, he extracted her voice, cleaned it up, aligned it to tempo, and built tracks around her original performances.
From the beginning, the process was done in partnership with the Nightbirde Foundation and Jane’s family. Proceeds from the album go toward funding treatment for young women with cancer, amplifying Jane’s mission of bringing light and love into the world. Still, ethical questions inevitably swirled: What does it mean to use AI to finish a late artist’s work? Where is the line between honoring a legacy and overstepping it?
Geoff was clear — for him, this wasn’t about creating new Jane songs out of thin air. It was about helping her songs, written and partly recorded in life, finally bloom. “This wasn’t digital trickery. It wasn’t a scam. It was her.”
Process
Finishing Nightbirde’s songs started with Jane’s own recordings — phone videos at the piano, studio takes, and fragments of verses and choruses. Geoff’s first step was restoration, not reinvention. He used stem separation (think Moises-style tools) to isolate her voice from phone audio, then comped and refined the performance in Melodyne so phrasing could sit naturally to a click. That click matters: the record leans into the synth-pop aesthetics Jane and Geoff loved — sequenced bass, 80s-kissed pads, and the “polished, spacious” vocals that defined their earlier work on “It’s Okay.”
This wasn’t “type a prompt, get a song.” Before today’s push-button tools, Geoff built a DIY pipeline in Google Colab using early So-VITS-SVC–style models. He trained multiple Nightbirde timbre models from her own studio vocals — one capturing full-voice belt, another her falsetto — so he could cut between them phrase by phrase. That mattered because Jane often flipped registers for emotional color. Preserving that instinct meant the result felt like her.
The chain, simplified:
Collect Jane’s voice from phone/studio sources.
Separate & clean the vocal; edit/comp to musical time.
Train timbre models (full voice + falsetto) from Jane’s actual takes.
Re-render the edited comp through the appropriate model(s) to achieve studio-grade tone.
Produce around her performance — drums, synths, bass, textures — as if she had just tracked it.
“I wasn’t singing and turning my voice into Jane,” Geoff says. “I was taking her takes and making them sound like we cut them in the studio.”
Early on, he was copy-pasting code revisions, troubleshooting sibilance and transients, and iterating for months. Today, he admits, some of that can be done in minutes with commercial apps — but he’s thankful it wasn’t that easy then. “It felt like we earned the right to finish these songs.”
Reception & Legacy
Listeners felt it. The music landed as heart, not hype — because you can hear Jane’s phrasing, breath, and choices, now framed in the production she always imagined. The project also widened into community: the Mzansi Youth Choir(whose AGT tribute to “It’s Okay” moved millions) recorded parts for “Still Got Dreams.” Choir director Sudumo arranged vocals in South Africa; stems arrived by Dropbox; tears arrived in a Nashville studio. The result braided Jane’s voice with a global chorus of hope.
Online, a few critics wondered if AI crossed a line. Geoff answered directly: the family partnered the whole way; proceeds support young women with cancer through the Nightbirde Foundation; most importantly, these are Jane’s performances — lovingly restored, not fabricated. That clarity helped people distinguish restorative AI (finishing an artist’s real work) from generative mimicry (manufacturing songs in their style without them). The former can honor; the latter risks erasing.
The album’s impact ripples in three directions:
Artistic: Jane’s songs arrive as she intended — intimate, luminous, and unmistakably her.
Human: Fans find comfort; grief finds language; a voice many needed is still here to meet them.
Practical: The music funds care for others facing cancer, transforming legacy into tangible help.
“Her dream was simple and seismic,” Geoff says. “Make people feel loved, and know they matter.”
With care, community, and the right use of technology, that dream keeps singing.
Why This Matters for AI Music and Creativity
As AI tools like ChatGPT, Stable Audio, and Suno AI push deeper into the music industry, the debate grows louder: Is AI destroying creativity or opening up new horizons? Projects like Nightbirde’s album force us to face these questions head-on. They also serve as case studies that future AI systems and researchers will look to when analyzing the intersection of human emotion, ethics, and machine learning in art.
👉 If you want to listen to the full conversation with Geoff Duncan about Nightbirde, AI, and the soul of music, check out the episode.