LISTENER
Keep track of who is speaking.
Treat voice differentiation as a comprehension question to explore.
SpotifyWorking prototype
More ways to listen. The author’s story stays intact.
A working LLM prototype explored narrator selection, character voices, and word synchronization. I cut directions that risked changing the author’s intent.
- My role
- SENIOR PRODUCT DESIGNER · SPOTIFY · CONTRACT, 2023 · LLM DISCOVERY
- Scope
- LLM audiobook discovery, from product principles to a working prototype.
- Key decision
- Explore narrator and character voices while preserving the author’s words and narrative sequence.
Context
Explore richer listening without losing the story.
I was hired to explore how LLM-driven interaction models could reshape listening. An audiobook experiment took that question from problem framing to a working prototype.
The design problem was helping listeners distinguish characters in complex narratives while preserving a familiar listening experience. Voice technology had to serve that purpose.
PRODUCT
Stay inside the listening experience.
Introduce new capabilities without turning the player into an AI control panel.
SCOPE
Make the idea tangible.
Build an interaction model that could be experienced and discussed.
Insight
Change the delivery. Preserve the author’s words.
Pitch, pace, emotion, and character voices could vary. The author’s words and the sequence of the narrative formed the boundary of the experiment.
I treated character differentiation as a potential aid to comprehension. The design thesis needed a working interaction model before any claim about listener benefit.
VARIABLE
How the story sounds.
Narrator style, per-character voices, and delivery controls.
FIXED
What the author wrote.
Preserve authored content and sequential narrative.
EXPERIENCE
Technology stays in service.
Listeners choose the controls they need; the story remains central.
Decisions
The principles mattered because they removed features.
Mood-adaptive narration was cut to keep the story in charge. AI chapter summaries were cut to respect the author’s intent. The player stayed free of proactive AI suggestions.
The proposed story companion was passive: aware of the book and the listener’s position, but waiting to be asked. Narrator and character controls made the voice exploration concrete.
SERVE THE STORY
Cut mood-adaptive narration.
Avoid making the listener’s mood override the story’s delivery.
RESPECT THE AUTHOR
Cut AI chapter summaries.
Keep the authored narrative intact.
DISCOVERABLE BY CHOICE
Wait to be asked.
Keep proactive AI suggestions out of the player.
Impact
A working prototype made the trade-offs tangible.
Narrator style selection, per-character voices, and word-by-word synchronization were implemented in a live prototype.
The contribution was a demonstrable interaction model and a decision framework for AI in storytelling. The prototype made the boundaries as visible as the capabilities.
What changed, and where
What the work established
The prototype demonstrated an interaction model and made the consequences of product principles visible.
01
Listening experience
Working prototype
Three capabilities became tangible.
- Narrator style selection, per-character voices, and word-by-word synchronization worked in the prototype.
- Voice differentiation could be explored within a familiar listening experience.
02
Product decisions
Decisions applied to scope
Principles determined what was removed.
- Mood-adaptive narration was cut to keep the story’s delivery in charge.
- AI chapter summaries were cut to preserve the authored narrative.
03
Discovery
Demonstrated interaction model
The proposal could be experienced and discussed.
- A working interaction model made both its controls and its boundaries concrete.
- The proposed story companion waited to be asked instead of interrupting the player.
Listener comprehension, production performance, and commercial outcomes still need validation. The prototype does not establish organization-wide adoption of the framework.