Team

AI and Magritte, or how the KMSKA brought the voice of the iconic painter back to life

An interview with Claudio Carreon, ICT staff member

Visitors can immerse themselves in the world of the modern master in Magritte. La ligne de vie until 22 February. The exhibition is based on a unique lecture that the painter himself gave at the KMSKA in 1938. The fascinating insight into Magritte’s thoughts on both art and society is illustrated in the exhibition not only through an important selection of his works. Quotations from his lecture, delivered by the reconstructed voice of the Belgian surrealist himself, also guide visitors through the experience.

How do you start such an innovative project?

To create the AI voice, we used ElevenLabs, a platform specifically focused on AI audio applications. Using existing audio clips, a voice model is created, which can then be used to convert your own texts—in this case, his lecture—into audio. The better the source material, the better the result.

What did this mean in the case of Magritte?

We were provided with an existing interview from 1965. The segments in which Magritte spoke were extracted and then uploaded to ElevenLabs. In fact, the first voice model was immediately usable because the original interview was of sufficiently clear quality.

Curious about the original interview? You can listen to it again here.

Do unexpected challenges also arise during such a process?

We noticed that, in this case, most of the time was spent on the so-called prosodic elements of speech. These elements include things like rhythm, intonation, and stress. They were checked and adjusted throughout the lecture wherever necessary. For example, sometimes Magritte speaks about a work, and the pronunciation of Il peint "Le Chant d'amour" differs from Il peint le chant d'amour.

And what about the language? Can you trust AI for this?

With the help of colleagues and guides, we listened to the AI-generated French lecture and provided feedback whenever adjustments to the speech were needed. With the help of an intern, we also manually added many short pauses throughout the lecture. This prevents it from running continuously, making it easier for visitors to understand.

This is the KMSKA’s second project using AI and speech recognition. Earlier, a smart visitor assistant was also developed for the Studio Rubens restoration workshop. If you want to learn more about that, read this article.