AI is struggling to hear the Gulf’s music

When I arrived in Abu Dhabi 13 years ago, I had an idea. As a composer, I had previously worked in Portugal on developing computer programmes capable of generating music, including jazz. I wanted to explore whether similar technology could be used to create music inspired by the traditions of the Gulf.

There was just one problem – I couldn’t find the recordings. Everyone I spoke to described the richness and diversity of the region’s music, but much of it was surprisingly difficult to find, so I began recording local musicians myself. The following year, I travelled to Kuwait to record traditional pearl-diving music. What began as an experiment in computer-generated music gradually became a broader effort to document, understand and make these traditions more accessible.

That experience revealed a problem that has become increasingly relevant as artificial intelligence transforms how we experience music. We have never had access to so many recordings, yet the algorithms that help us navigate them are much better at recognising some musical traditions than others. Paradoxically, the digitisation of music may be widening the divide between what gets heard and what remains largely invisible.

01:03

Why 2026 could be the UAE's busiest year for live music

Music recommendation systems are remarkably effective because they can draw on enormous quantities of data. They can distinguish between Taylor Swift and Ariana Grande, identify similarities between recordings and suggest songs that match our tastes. But ask the same systems to recognise Al Ayyala, a traditional Emirati performing art involving music, poetry and movement, and the task becomes considerably harder. There simply aren’t comparable quantities of recordings or detailed musical information available to train them, making these traditions harder to discover even when recordings are available online.

Addressing this imbalance is one of the challenges facing Music Information Retrieval, a field combining musical knowledge, engineering, computer science and AI to analyse and organise music digitally. Much of the technology has been developed around musical traditions with extensive recordings and established theoretical frameworks. Arabic music, however, uses maqam, a melodic system that can incorporate intervals outside western equal-tempered tuning. Indian classical music has its own sophisticated melodic and rhythmic structures. Algorithms trained primarily on western music may struggle to recognise these characteristics.

At NYU Abu Dhabi, we’re addressing this by combining computational methods with the knowledge of musicians and researchers, rather than expecting computers to learn everything from recordings alone. In our research into Carnatic music from southern India, for example, we’re developing methods to distinguish the 13 different types of stroke played on a mridangam, a double-headed drum. Existing systems can readily identify instruments in a western drum kit, but recognising the playing techniques of a mridangam remains a research challenge.

The potential extends beyond identifying instruments or musical styles. Centuries of trade and migration across the Indian Ocean have connected the Gulf’s musical traditions with those of East Africa and southern India. In my teaching, I sometimes play students’ recordings of music from Ghana alongside Gulf pearl-diving songs and Brazilian samba. Despite their different origins, their rhythmic characteristics, including handclapping and percussion patterns, can be strikingly similar. Computational analysis could help us investigate such relationships across large collections of recordings, revealing patterns that would be difficult to examine through listening alone.

Music recommendation systems are remarkably effective because they can draw on enormous quantities of data

We have already applied this approach in research into Candomble, an Afro-Brazilian religious tradition, examining relationships between its rhythmic patterns and those found in West African music. Similar methods could deepen our understanding of the Gulf’s musical connections with other regions. Interpreting these findings, however, requires the knowledge of musicians, historians and ethnomusicologists alongside computer scientists.

In 2016, I co-founded the Music and Sound Cultures, or MaSC, research group at NYU Abu Dhabi with ethnomusicologist Andy Eisenberg. Our work includes recording traditional Emirati music, researching Carnatic percussion and developing creative projects around traditional crafts. One ambition is to make musical connections more visible. Imagine listening to a contemporary musician incorporating Gulf pearl-diving rhythms and having a streaming platform introduce you to traditional recordings from Kuwait or Bahrain. The technology must first learn to recognise those relationships.

This month, NYU Abu Dhabi will host the International Society for Music Information Retrieval Conference for the first time in the Arab world, following a collaborative effort involving researchers at our university, Mohamed bin Zayed University of Artificial Intelligence and Zayed University. We hope to encourage greater international interest in the region’s musical traditions and expand the research community studying them. Among the work being presented is the first Arabic music dataset created by NYU doctoral student Peter Traver, MaSC members Eisenberg and Robert Rowe, in collaboration with virtuoso violinist Layth Siddiq, which combines expert musical knowledge with computational methods.

Show caption: A musical performance at New York University Abu Dhabi. NYUA…

As a composer, my interest extends beyond analysing and archiving music. We’re also exploring tools that allow contemporary artists to experiment with regional rhythms, melodic systems and performance practices. Musicians such as Ghazi Al-Mulaifi, through his work with Boom.Diwan, demonstrate how traditional Gulf music can continue to inspire contemporary artistic expression.

This reflects something fundamental about cultural heritage. Traditions survive because people continue to practise, interpret and adapt them. Recording music for future generations is essential, particularly where documentation is limited, but we should also create opportunities for people to engage with that heritage today.

I often explain our approach by saying that we don’t have a freezer, we have an oven. We want to preserve these traditions, certainly, but we also want people to create something with them. As AI becomes increasingly capable of generating new music, its potential to help us understand what already exists deserves equal attention. The Gulf’s musical traditions carry centuries of history and cultural exchange. Making them more accessible will help future generations discover that history, draw inspiration from it and keep it alive.