Explainability and Real-Time in Music Information Retrieval: Motivations and Possible Scenarios

In recent years, various explainable artificial intelligence methods have been proposed for different Music Information Retrieval (MIR) applications. However, real-time aspects of explainability and the musical applications that they can enable have been largely overlooked by researchers. In this position paper, we propose a vision for real-time explanation systems in the context of MIR applications. We describe three application scenarios where such systems would be useful, although the same concepts can be applied to other musical cases. In the first scenario, we propose a system that provides explanations for an emotion recognition model in the context of music composition and production. In the second scenario, we investigate the benefits of explaining a music genre recognition system for live performances. In the third scenario, we explore the utility of an explanation system for an application in musical instrument learning. Subsequently, we discuss the main challenges associated with the envisioned real-time explanation systems and use cases, especially in relation to the need for easy-to-grasp explanations and to the possible embedding of such systems into smart musical instruments.