Sonic Transaction

An emotional act, extracted. Sonic Transaction takes the most ordinary thing a visitor can hand over, a Spotify playlist link, and turns it into an advertising profile, in front of them. Genres are weighted against published personality research, rendered as a unique 3D artefact, run through a generative model for a set of advertising segments, and printed as a physical receipt for the data transaction the visitor just completed.

The Data Mirror

Any playlist link a visitor pastes in gets its genres extracted and weighted against published personality research to estimate a Big Five profile. The system renders that profile as a 3D artefact, a generative model turns it into a set of advertising segments, and a physical receipt prints, recording what the mirror showed and what data it used to show it.

Spotify Wrapped and the Flattering Mirror

Spotify Wrapped already runs this process every year: it infers behaviour from listening, packages it as personal insight, and presents it as a fun, shareable ritual. Wrapped stops before the part that matters most, the advertising segment that behavioural inference is built to feed. Sonic Transaction runs the same mechanism and keeps that part in view.

Keeping the Mirror Honest

Every weight in the scoring pipeline comes from a published table, Anderson et al. (2021)'s genre-personality correlations. Every track is logged as found, confirmed absent, or flagged as a failure. The system never fills a gap with a guess.

The Limits of the Reflection

The genre-personality correlations this piece scores against were built on Spotify's proprietary infrastructure, no longer accessible outside the company, so applying them to independently sourced data produces a research-grounded estimate rather than a validated prediction. Genre-based proxies misread certain personality types in a consistent, predictable direction. Whatever the underlying research gets wrong about a person, this mirror reflects that error back, which is close to the point: an advertiser's model of a person is built the same way, assumptions and errors included.

Data Sources & Pipeline

A playlist link is the only visitor input required. Track IDs are extracted from the Spotify embed page, audio features come from ReccoBeats, genre tags from Last.fm mapped against a published 66-genre correlation table (Anderson et al. 2021), scored 70% genre and 30% audio feature, aggregated into a Big Five estimate, rendered as the artefact, matched to advertising segments, and printed as a receipt.

The Reflection, Itemised

The 3D artefact is generated live from the visitor's own estimated profile, so every visitor's shape is unique. Each of the five traits can be pulled out into its own orbiting object with a plain-language descriptor and a score. A generative model matches the profile to marketing segments using the same logic a platform would use to target the visitor, and the receipt itemises tracks analysed, genre breakdown, and the specific songs that drove each trait score, printed on thermal paper.

The design language visually distills to text by the end of the experience, to drive home the point that an emotional act gets turned into a data transaction.