At the Spanish University of Pompeu Fabra, a new type of artificial intelligence was developed, which was taught to play the violin. More precisely, the process is close to completion, and now the authors of the development are wondering which further development direction to choose. Turn an AI into a concert violinist, or start working on a feedback system so that a machine can train people?
Spanish engineers used the Myo gesture recognition armband, which was attached to the hands of a professional violinist. He consistently, as part of the educational process, demonstrated the classical styles of playing the instrument. Detache, Martele, Spiccato, Staccato, Sotiye and others - the AI meticulously recorded all changes in the position of the musician's hands. In parallel, the original sound of the violin was recorded.
Further, using machine learning algorithms, the system learned to compare sounds and movements, learned the sequence of gestures necessary to reproduce specific sounds, and identified the patterns of musical styles. With further iterations, the AI learned to determine the performance technique of different violinists with an accuracy of up to 94%. He can play anything himself - there would be a suitable hardware base, manipulators with sufficient accuracy and freedom of movement.
The potential of such a system for training new musicians is that one AI has the knowledge and skills of teachers of different levels, from beginner to advanced. And he can immediately point out the mistakes of students with high accuracy. But here's the trouble - he doesn't understand anything about training, an individual approach and other important aspects of teaching music. That is why it will not make serious competition for human violinists yet.