Women Composers’ Missing Music Is Also Missing From AI Training Data

A song written almost 900 years ago by Contessa de Dia survives as the only melody by a woman from her musical tradition in medieval southern France. For classical singer Gabriella Laccio, that absence raised a question she had not thought to ask: who decides which voices survive?

A secondhand book market in London gave her an unexpected answer in the form of an encyclopedia listing 6,000 women composers. Di Laccio says she could name only five. She began looking up the composers and listening to their music, then sharing it constantly. In a presentation, she paired Marianna Martines, a leading musical voice in 18th-century Vienna with more than 70 compositions, with Florence Price, whose work spanned major musical forms. Price was the first African American woman to have a symphony performed by a major orchestra.

Di Laccio’s husband, a railway engineer, eventually asked whether she planned to do anything beyond playing the music at breakfast. She made a list of women composers and posted it online. Women around the world wrote to thank her; some asked to be added as composers themselves. The list grew into a foundation that, Di Laccio says, is uncovering more than 18,000 women composers across music history.

The scale of the gap, she argues, helps explain why even a professional musician could spend a lifetime training and performing without encountering thousands of women’s work. A chart in her presentation, credited to Donne Foundation and USC Annenberg, gives the share of music written by men as 92.5 percent in classical music, 96.6 percent in film and 86.2 percent in pop.

Di Laccio says the gap means audiences have missed an enormous body of music.

The imbalance is not only a matter of music history. Di Laccio connects the underrepresentation of women in recorded music to the data available to systems that generate music. She cites more than 50,000 AI-generated tracks being added every day on one platform and argues that these systems can learn only from the data available to them. If that data is missing much of women’s musical creativity, she says, AI is learning from that same silence.

Di Laccio is explicit about her reservations. She believes musicians should write music rather than machines, and says generating music from artists’ work without consent or proper remuneration is wrong. But she also treats the technology as something that is here to stay. Her proposed response is to record both historical composers and contemporary musicians who remain underrepresented.

A recording, in Di Laccio’s account, makes music searchable and teachable, and can bring it to classrooms, radio stations and streaming platforms. It also makes the music part of the systems shaping what the world will hear next. Her proposal assigns responsibility beyond historians, performers and archives: she asks technology companies making AI music, and anyone profiting from it, to invest in making the musical catalog more complete.

Di Laccio’s closing example makes the stakes of preservation personal. Ilse Weber was a poet and singer imprisoned with her family in the Theresienstadt ghetto during the Second World War. There, she wrote songs to entertain the elderly and children. Weber and her son died in Auschwitz. Her husband buried her manuscripts under a tool shed, and a few songs survived.

One of them is “Wiegala,” a lullaby Di Laccio sings. The story turns the question of survival into a concrete chain of events: a song was written for people in captivity, manuscripts were hidden, and some of the music made it through.

“If you can’t hear this part, perhaps history never recorded it,” reads the slide shown during the performance. Di Laccio’s point is that what becomes part of history depends partly on what is recorded, and that recordings can shape what future listeners, classrooms and music systems encounter. “Let’s teach it to remember the sounds of all of us,” she says.