Spotify Carla Bruni

How AI led me to Carla Bruni

It had to be Edith Piaf, Madeleine Peyroux, or Joan Baez. It could also have been Eva Cassidy, Janis Ian, Sarah Jarosz or Joni Mitchell. More likely, it was my listening to all these artists that led Spotify to recommend Carla Bruni to me earlier this month. I have been listening to her songs since.

Before Spotify, I don’t think I would have listened to Bruni. All I knew before she ended up in my Discover Weekly recommendation was that she was a model married to the former French prime minister Nicolas Sarkozy.

Music discovery before internet

I grew up when musical taste was dictated by FM stations, songhits (magazines that print songs with accompanying guitar chords), and cassette tapes shared with friends.

At home, it was The Platters, Brothers Four, and Kenny Rogers. On Sundays it was the Beatles and Elvis Presley in most FM stations in General Santos City, the nearest metro from where we picked up broadcasts. In school, it was Guns N’ Roses, Aerosmith and Bon Jovi.

Spotify Carla Bruni

DISCOVERY. I discovered Carla Bruni through Spotify’s recommendation engine.

My greatest discovery happened while walking home among the pineapple fields in Dole Philippines. While passing by one of the company’s compounds I heard blaring on a trumpet speaker that served as PA system the singular voice of Bob Dylan asking “How many roads must a man walk down / Before you call him a man.” It was epiphany.

In UP Diliman I discovered Gary Granada, Grupong Pendong, Yano, Eraserheads, and Bayang Barrios.

When I started working for The Freeman as a City Hall reporter in 1996, I ended my day listening to that DJ with the raspy voice who closed the broadcast for the then Cebu station of RJ FM. With her, I discovered Spiral Starecase, Procol Harum, Billie Holiday, and Ella Fitzgerald.

Audio scrobbling

With the Internet came a wider source for music discoveries. Stories about artists would often mention other musicians in their genre. While reading about Billie Holiday, for example, I discovered Amy Winehouse.

I once used Last.FM’s audio scrobbling service and got matched with other users who had similar music tastes and I browsed through their catalogs to check out new artists and albums.

These discovery modes culminate in Spotify.

Apart from simplifying and making legal access to music cheaper, Spotify’s important innovation is its recommendation engine.

Through it I found Spanish guitar master Andres Segovia (because of Eliot Fisk), Renata Tabaldi (because of Maria Callas), and Pablo Casals (because of Yo Yo Ma).

And of course, Carla Bruni.

Bruni’s breathy take on The Winner Takes It All rivals the original vocals by Agnetha Fältskog. It is subdued and yet more emotional. It is my favorite of her English covers. She’s as enchanting in Moon River, channeling Audrey Hepburn in Breakfast at Tiffany’s.

Her other songs that I Iove include Dolce Francia, Little French Song, Mon Raymond and Darling.

The recommendation engine that accurately suggested Bruni to me tapped machine learning, big data, raw audio analysis, and natural language processing, wrote software engineer Sophia Ciocca.

Spotify Discover Weekly

In her post, she said Spotify uses 3 recommendation models to come up with individual Discover Weekly playlists. These are:

Collaborative filtering, which compares what people are playing and identifies users with similar taste based on the number of times they play music tracks. Spotify then recommends to people titles they haven’t played but were played by similar users.

Natural Language Processing of text about music found online such as blog posts. It then “figures out what people are saying about specific artists and songs — what adjectives and language is frequently used about those songs, and which other artists and songs are also discussed alongside them.” It’s like performing what I have been doing to discover new acts by reading about musicians but at a mind-boggling scale.

Spotify recommendations

UNIQUE RECOMMENDATIONS. By tapping big data and machine learning, Spotify is able to give accurate and unique recommendations to its users.

The third model is analysis of raw audio to “understand fundamental similarities” between songs based on such things as time signature, key, tempo and loudness. Audio analysis allows Spotify to recommend new songs with few listens, songs that wouldn’t have been picked up by the filtering and natural language processing.

By tapping these recommendation models, Spotify creates unique playlists for users that show a fundamental understanding of their individual tastes.

Loading Facebook Comments ...

Leave a Reply

Your email address will not be published. Required fields are marked *