Everyday life actuators such as music can be used as neurofeedback to improve quality of life and performance. To track one's performance, we develop a performance decoder that captures the time-varying nature of the process noise. We first design a performance state-space model within an autoregressive conditional heteroskedasticity (ARCH) framework to enable adaptive performance state estimation. Then, we design an expectation-maximization algorithm to decode a hidden performance state and estimate the model parameters. Particularly, by considering the sequence of responses and the corresponding reaction times as the observation vector, we employ particle-filtering to track the hidden performance state. We investigate the decoder's performance on experimental data. The estimated performance state is aligned with different task difficulty levels. During the experiment, music was used as neurofeedback to regulate the arousal. Our results indicate music can be utilized to regulate arousal and modulate performance in smart environments. Adaptive performance estimation in varying environments in presence of neurofeedback is a key step for improving performance in real-world settings. Envisioned cyber-physical systems applications include improving productivity in smart workplaces and enhancing learning in online educational systems.