Calfa workshop at TUMO where AI and the music of Komitas intertwine
In March 2024, Chahan and Baptiste led and animated a workshop dedicated to Artificial Intelligence and the musicologist Komitas a the TUMO Center in Erevan. For two weeks, TUMO students, aged 13 to 18, learned to create an AI model capable to read the music sheets of the famous Armenian musicologist and composer, whom we are indebted to for the preservation of numerous popular songs.
The students were taught how to create an AI model from scratch, starting with data creation, model training and finally the prediction of results. Like for our previous workshops at TUMO, our purpose remains to mix teaching about AI with the preservation and promotion of the Armenian cultural heritage. Previously, our workshops were focused on the creation of models dedicated to the recognition of handwritten documents and to the detection of images in pages from the archives of the FSL library of Erevan.
This year, the first week of the workshop was focused on OMR (Optical Music Recognition): the goal was to train a model capable to read the music sheets of Komitas, on the basis of digitized versions extracted from first editions of book collections from the Samuelian Oriental Bookshop in Paris, where Calfa has started a digitization and preservation project.
This session was to the opportunity for the teenagers to study neural networks and computer vision with YOLO in the applied case of music detection. The students also learned how to process an AI project, through environment implementation, data creation, training configuration (number of iterations, learning rate, etc) and of course the evaluation of the results (metrics curves, confusion matrix etc.). Throughout the whole week, they have been working on first editions of the music sheets of Komitas dating from 1920s and 30s.
This first week ended with the students visiting the Komitas Center in Erevan, thanks to Dr. Tatevik Shakhkulyan, head of research, who conducted and made this visit possible. The aim of the visit was to break away temporarily from the workshop, the coding and to give substance and visibility to the link between digital technologies and the tangible collections of the museum. The students discovered the rich collections of the museum: archives, personal belongings and the original music sheets of Komitas, on which they had been working the whole week.
The second week of the workshop was dedicated to the creation of a generative AI, capable of creating music: it was the opportunity for the students to learn about generative networks and the learning phases of models. The training were carried out on the TUMO computers, as well as possible, given their limited performances compared to the usually required computing power for the training of these technologies. Nevertheless, we were able to manage some notes of duduk and some percussion sounds, after a 12h training set, until the ram of the computers was overloaded! Although far from the big generative networks, that require very large computing power to be developed, the teenagers were still able to set foot in the very rich field of AI generative research and to experience first hand the implementation and processing of packages used by the professionals of the field.
Finally, each student was able to code their own webapp with Streamlit, on order to portage their model on a demo interface. On the very last day of the workshop, all the results achieved were made public. Find the content of the workshop (program and code) on GitHub!
