2021 - 2026: the great epic of musical AI - L'ADN

Since the appearance of the K7 in 1963, listening media have literally mutated under the impetus of digital technologies. Today, to discover new sounds and listen to them wherever we want, most of us follow the recommendations of streaming platforms that run on artificial intelligence. But, as Frédéric Amadu and Nicolas Pingnelain of Ircam Amplify told us, this is only the beginning.

Ambient intelligence, geographical continuity, spatial restitution: how will AI change the lives of music lovers?

How the stream invaded the world

Between the metro, work, sleep, a routine has set in: the music stream. In France, it was in 2016 that dedicated platforms imposed themselves on the market by becoming the main source of income for the music industry. By capillarity, streaming consumption has since continued to climb to explode in 2019 and reach records with the Covid crisis. According to a Kantar study, 25% of French people have consumed more streaming music since the first confinement, and 65% of them are determined not to lose this good habit.

The principle of music recommendation that we all love now on streaming services works thanks to artificial intelligence and more precisely to Machine Learning, a technology that learns our tastes to suggest a host of playlists. These cross styles, genres, atmospheres and novelties that we consumed during the week. Sometimes, the AI ​​that determines these compilations also offers thematic medleys supposed to accompany the mood of daily activities (sports session, work session) or echo an emotion felt.

But for Frédéric Amadu, CTO of Ircam Amplify, these themed suggestions are too automatic, linear and do not really keep their promises. “On the major music platforms, the playlists offered are heterogeneous and static. The problem is that they are common to all. They cannot correspond to what everyone expects as an atmosphere for such and such an activity, or as a response to an emotion”, he explains to us. In addition, adds Nicolas Pingnelain, sales manager at Ircam Amplify: “Today we have a plethoric production flow and therefore a multitude of choices. Yet we have never listened to the same things so much.

Are we insidiously locking ourselves up in a musical filter bubble, as Eli Pariser theorized? Accomplices, the algorithms that we encounter daily participate, in intelligence with our cognitive biases, in this intellectual compartmentalization that reinforces our tastes and us away from the unexpected. On Spotify, personalized playlists are responsible for half of monthly listens for more than 8,000 artists. According to researchers Jean-Samuel Beuscart, Samuel Coavoux and Sisley Maillard in their publication Music Recommendation Algorithms and Listener Autonomy, “work evaluating the effects of algorithmic recommendations focuses primarily on their effects on content diversity: because it is based on the competition of consumption, collaborative filtering can promote confinement in a portfolio of very similar products. While it is thought of as a way to promote the discovery of little-known artists, it can also paradoxically strengthen the place of stars.

To help us break with these (bad) habits and promote new music consumption, Ircam Amplify, a subsidiary of Ircam (Institute for Acoustic/Music Research and Coordination) launched Metasound in March. This solution aims to create new interconnections between titles, genres and different universes, but also to offer intelligent and scalable playlists (customizable and contextualized). Titles and catalogs passing under the radar of the general public can thus be valued. This is encouraging news for young artists, their distributors and users eager for new discoveries!

2023: the real-time descriptor for smart playlists

Fast forward to the soundtrack of the future. Within two years, real-time descriptors will be legion. But what is it? This technology analyzes live audio broadcast and the sound environment of the listening place. For example, it will take into account intonation, intensity, rhythm, frequencies and background noises in order to adapt its volume and refine the musical proposals.

A descriptor to pimp family reunions, dinners with friends or rainy Sundays? That's the idea! If music is the cornerstone of these cherished moments, it will not necessarily be the same depending on the mood. Collective or introspective, listening is supposed to embrace a moment and accompany it from start to finish. To prevent it from being read according to AIs and linear playlists, the descriptor can adapt it to the energy of the room, the ambient sound, the intensity, and thus self-regulate and become one with the atmosphere felt, in search of the perfect atmosphere: the right content, at the right time, for a given use. Because "what does not work today is to pretend to have a playlist to work on, to be calm and focused, without taking into account the tastes and the atmosphere in which the user is immersed", points out Frédéric Amadu, "and then, adds Nicolas Pingnelain, during dinners, we sometimes need the music to adapt to the intonations, the energy of the debates or the fatigue of the guests, by offering an adequate sound environment, capturing or not the be careful with known titles or new songs”.

Today, discovering new songs can be made difficult, especially because of this bubble of filters that we have trouble getting rid of. But why ? AIs are based on the structure or style of a piece, defining rigid standards to qualify our tastes in music. Too pre-formatted, the listening suggestions supposed to marry our tastes and push us to eclecticism, remain “heterogeneous and static”. "Today and in two years, it will be necessary to create homogeneous "identity cards" between the pieces, therefore to identify and automatically extract the basic data of any music and why not, to specific uses, to increase their content through manual editorial actions. Not only to link them by genre, but also by emotion, intensity, atmosphere... Today, some models design their playlists in an automated way. Others will have to match the editorial approach and the usage data to be as close as possible to the user's desires at the moment T, and thus be fairer", explains Nicolas Pingnelain, and to add: " otherwise there will be a risk of constantly falling back on the songs that comfort us. You have to push the boundaries and allow the pieces to fit into new personalized listening contexts (genres, emotions, moods, etc.). By qualifying musical catalogs of millions of titles using machine learning, we will thus allow more discoveries, for more adapted and richer musical journeys”.

Adaptability is therefore the key word for this AI conductor and its increasingly sophisticated algorithm that will generate intelligent playlists according to people and their environment.

What if there were other settings to make the experience even more immersive?

2026: from geographical continuity to spatial restitution

In five years, technologies will accompany the user from point A to point B while adapting to their environment, their equipment and its geographical position (or even its emotional state!).

To infinity and beyond? For Frédéric Amadu, the future is turned towards an accompanying AI where continuity will be an essential parameter: “in five years, music will follow us everywhere. It will start via the broadcaster at home, the song will automatically resume in the car, then in the headphones on the way to work to continue alone on the computer”. And to change according to the places? A big "yes!" for Nicolas Pingnelain: “it is possible to imagine that the music changes as soon as you get close to work, once you are wide awake to give motivation! ". However, a problem is raised by the duo: this continuity would work for an introspective listening to the music. It will therefore have to adapt, thanks to real-time sound descriptors, if the user is accompanied by friends or children in his car. Two rooms, two atmospheres, but a thousand possibilities.

On a technical level, restitution comes into play. The equipment differs, from the house to the helmet to the car. The AI ​​will be, in these different spaces, able to reproduce the appropriate sound in intelligence and by playing on the spatialization or the binaural sound (in particular via the audio headsets) - all this "thanks to the microphone and the capture via the equipment" , explains Frédéric Amadu.

For Nicolas Pingnelain, thanks to this spatialization and real-time descriptors, the future will be hybrid experiences, propelled by the health crisis and technological advances. "In a few years, it will be possible to experience a concert at home thanks to this spatialization and these descriptors, for example to integrate your encouragement with the sound of the fans live in the stadium during a football match". More personal and intimate, our relationship to music, to listening to it, will quickly be transformed by artificial intelligence. A traveling companion for music lovers, it will be able to adapt to both their private environments and their technological equipment.