The musical world does not move in a single straight line. While researchers test artificial intelligence with auditory illusions, other systems are learning to turn sound into notes. Musicians are returning to historical instruments, performing enormous symphonies from memory, and composing songs between the morning coffee and the walk to school.
We have gathered five events of the year 2026 — five notes of one great composition about how music, technology, and the very way we listen are changing.
First note. The music that isn't there
September 2, 2026 Hayun Kim, Yunis Hon, and Kyogu Lee presented the preprint Auditory Illusion Benchmark for Large Audio Language Models — the first specialized test of large audio language models using auditory illusions.
The test included ten illusions related to music, speech, and other sounds. The models' answers were compared with the results of controlled listening sessions involving people.
The researchers found a notable difference. In relatively simple acoustic illusions, most models more often remained faithful to the physical characteristics of the signal. A human, meanwhile, could hear a holistic image or a musical structure that was literally not present in the recording.
But when perception depended on knowledge of language or learned musical patterns, some systems began to answer in a more human-like way. Probably, training on an enormous amount of speech and music allowed them to adopt part of our expectations.
At the same time, not a single model reproduced the human perceptual profile completely.
This is for now — a preprint that has not undergone scientific peer review. The work does not prove that the machine possesses consciousness or experiences music like a human does. But it opens a new path: to evaluate artificial hearing not only by correct answers, but also by the character of its mistakes.
The main question:
If the ear receives sound and the brain completes what is heard — where is music actually born?
Second note. He played — and the music recorded itself
September 4, 2026 a preprint was published Harmonica: Accurate and Lightweight Instrument-Agnostic Music Transcription.
Longshen Ou, Hector Martel, Joe Hennessey-Priest, and Taemin Cho have developed a family of models capable of converting a musical recording into a sequence of notes regardless of the instrument being played.
One of the main difficulties of automatic transcription lies in timbre. The same note on a piano, violin, trumpet, or human voice has a different overtone structure. The system needs to separate the pitch from the unique coloring of the instrument—to hear the note within the timbre.
The most compact version of Harmonica contains just 26,3 thousand parameters and, according to the authors' results, processes material more than 1600 times faster than its real duration. The large version showed the best result among the models with which the researchers compared it.
However, this is also an un-peer-reviewed preprint, so the stated performance figures still require independent verification. The system also does not replace the musician: musical notation does not fully convey timbre, breath, micro-dynamics, and the living movement of a performance.
But the prospect is exciting. An improvisation that sounded only once and vanished into thin air will be able to take on a notated form almost instantly—so that the musician can preserve, develop, or pass it on to others.
Third note. The symphony the whole orchestra remembers
On August 1 and 2, 2026 musicians of the Aurora Orchestra, conducted by Nicholas Collon, performed Gustav Mahler's First Symphony entirely from memory at London's Albert Hall.
Before the symphony, the audience was presented with a dramatized journey into the composer's inner world. The first part featured actors Jack Bardo and Sarah Toomey, as well as mezzo-soprano Natalie Lewis. The musicians not only played but also moved around the stage.
Mahler's story was intertwined with his meeting with Sigmund Freud, personal losses, and the imagery of the symphony. The audience was shown how the familiar melody Frère Jacques is transformed by Mahler into a funeral march.
But the main miracle happened after the intermission.
The orchestra members came out without scores or music stands. They could see each other more freely, move, and respond more attentively to the conductor and the overall sound.
Performing a symphony of such length and complexity from memory is more than just a memory exercise. Each participant must not only know their own part but also understand the architecture of the work: hearing where the music is now and where it is headed.
Thus, the orchestra becomes like a single organism. Everyone retains their own voice but exists within a shared knowledge of the work.
Fourth note. Can the past be heard?
September 9, 2026 The Guardian published an insightful piece on historically informed performance—an approach in which musicians study the instruments, techniques, and musical concepts of the era in which a work was created.
Early music sounds different not only because historical instruments have a different timbre. A natural horn without modern valves requires a different technique from the performer. Gut strings respond differently to the touch of the bow. And the harpsichord produces sound by plucking a string, which does not allow for volume control in the same way as a modern grand piano.
The instrument changes the musician's gestures, the expressive means available to them, and the very nature of the performance.
But it is impossible to completely restore the original sound: no recordings exist from the era of Bach and Mozart. At the same time, instruments, scores, treatises, and contemporary accounts have survived, allowing researchers to reconstruct many features of performance practice.
Therefore, it is more accurate to speak not of an absolutely “authentic” performance, but of a historically informed one. Its goal is not to preserve the past, but to make old music a living event of the present once again.
Fifth note. A song between coffee and the school run
September 9, 2026 The War on Drugs have released Who’s That? —their first new studio song in nearly four years.
Writer, vocalist, and producer Adam Granduciel said he wrote it in about ten minutes—between his first morning coffee and the school run with his child.
It is a song about trying to love oneself despite conditions that a person is not always able to control: anxiety, depression, and doubts about one's own worth.
Along with the track, an official music video directed by Max Flick was released.
This story brings our musical review back from laboratories, archives, and concert halls into ordinary human life. Sometimes a work does not require perfect silence, a special ritual, or months of preparation. Music arrives in the middle of the morning—if a person manages to hear its impulse and does not postpone creativity until the “right moment.”
When five notes become music
Each of these notes sounds in its own way.
The first reveals the creative nature of human perception.
The second helps preserve a vanishing improvisation.
The third turns the orchestra into a single memory.
The fourth connects times.
The fifth returns creativity to the living flow of everyday life.
At first glance, these stories belong to different worlds and eras. But together they add up to a composition about the main thing: music is never equal to just a sound signal.
A note can be preserved in a score, in human memory, or in a digital model. An instrument can have been created centuries ago or exist as a computer system. But music arises at the moment of encounter: sound joins with expectation, the musician with the instrument, the performer with the other participants, and the work with the listener, who brings to it their own memory and imagination.
Perhaps the future of music is determined not by who turns out to be more perfect—a human or a machine. Far more interesting is what forms of creativity will be born when the precision of technology meets the human ability to hear more than is contained in the signal.
