Deep Q-Network-Based Optimization of Piano Playing Skills and Psychological Feedback

Traditional methods often fail to precisely quantify and analyze playing skills, and it is difficult to make personalized adjustments based on individual differences of performers. At the same time, the influence of the performer’s psychological state on the performance quality is ignored. For this reason, this paper uses Deep Q-Network (DQN) to study the optimization of piano playing skills and the psychological feedback mechanism. The collected data are preprocessed using a short-time Fourier transform to extract multidimensional features. An optimization model based on DQN is constructed, which approximates the state-action value function through a neural network and generates a target Q value. The network weight is updated by minimizing the mean square error between the Q value output by the current network and the target Q value, and the state space, action space and reward mechanism are designed. The psychological feedback mechanism is applied to guide the model to adjust the performer’s emotions and psychological state while optimizing the playing skills through a variety of reward functions. The study shows that the average note playing accuracy of the optimization model studied in this paper is 96.15%, with a training time of 2.4[Formula: see text]s and a response time of 1.3[Formula: see text]s; the anxiety index of the psychological feedback mechanism based on DQN is 4.3 points, and the psychological state recovery time after playing is 8[Formula: see text]min. This paper’s research on piano playing skill optimization and psychological feedback based on DQN not only achieves precise quantitative analysis of playing skills, but also effectively solves the problem that traditional methods ignore the regulation of psychological state.