Effective Visualization of Individual Piano Performance Style Preferences Using Tempo and Dynamics Features with AIME
Ayako Minematsu, Takafumi Nakanishi · 2024
In appreciating piano performances, listeners often develop preferences for specific performance styles even for the same musical piece. This study proposes and validates a method for visualizing individual preferences in professional piano performances. By analyzing performance features, such as tempo and its first and second derivatives, and dynamics and its first and second derivatives, this method generates visual representations of preferred stylistic traits. Furthermore, by leveraging the Approximate Inverse Model Explanation (AIME) method from Explainable AI (XAI), the method extracts comprehensive preference features for each listener across multiple pieces. The key strength of this method is its ability to articulate a listener's preferred performance style in concrete terms. This empowers individuals to incorporate these elements into their play, thereby facilitating self-improvement. This also facilitates connections between individuals with shared stylistic preferences, fostering a deeper appreciation for musical nuances. This study presents a practical tool with the potential to translate abstract musical preferences into concrete, understandable, and actionable insights, thereby enhancing the experience of both performers and listeners.