Music: Algorithms at the controls

Music: Algorithms at the controls

The music industry is increasingly interested in artificial intelligence. In addition to making custom playlists or composing new tracks, machines are now being used to predict future hits.

Painting, literature, journalism… There are few fields today in which the machine hasn’t burst in. And each time, the same question is asked, tinged with fantasies and fears: is artificial intelligence going to replace, if not replace, the human being, or at least do better than him? Music is no exception. At the end of 2016, Sony’s Computer Science Laboratory (CSL) presented the title “Daddy’s car”. Inspired by the world of the Beatles, the song was partially created by a computer. A few months later, a complete album was presented by the Flow-Machines project, with artificial intelligence as its main “feature”.

And if these sounds are still far from delighting all ears, machines are already shaping our musical universes. On Spotify, Apple Music and Youtube, the algorithms have turned into custom DJs, with playlists designed just for us.

The Assisted Human

The next step: predicting the commercial success of a song. In 2018, researchers at the University of California (Irvine) claimed to have found the secret ingredients for hits in the making. Using an algorithm that scanned more than 500,000 songs released in the UK over the past 30 years, they indicated that they could predict a song’s success “85%” based on 18 variables.

In Finland, Hyperlive also plays the Mrs. Irma of music. This start-up boasts of having predicted the success of new songs by Justin Bieber, Billie Eilish or Katy Perry. And this, “simply based on the analysis of the song, without measuring other parameters such as activity on social networks, addition to playlists, media buzz, etc.”. »

To achieve this, we still had to resort to good old-fashioned human intervention; first, by defining a certain number of criteria to qualify the pieces (rhythm, tone, speed, genre, etc.). A musicologist then characterizes the songs that make up the corpus according to the emotions they bring (sadness, cheerfulness…). Finally, it is time for the machine to digest the whole, and to release regularities likely to be found in future hits to be analyzed.

Seeking to optimize the recommendation capacity of their algorithms, music streaming platforms have understood the interest of start-ups active in computer-assisted analysis. For example, Spotify has acquired French start-up Niland. In addition to Hyperlive or Musiio (based in Singapore), one of the most advanced companies in the field is the Belgian company Musimap. From several billion stored data and an analysis of 50 million tracks, Musimap seeks to develop recommendation tools that take into account the emotions of the user. Seduced by the project of the Liège-based company, Quincy Jones, the legendary producer of Thriller (Michael Jackson) became a shareholder last summer.

 

No Comments

Post A Comment