Perceptual audio quality assessment is a task that involves the characterization and estimation of perceived quality of an audio signal. Many existing systems, depending on psychoacoustic principles and statistical models, achieve reasonable performance under specific conditions (e.g., type of artifacts, impairment levels, etc.) but do not generalize well when these conditions vary. This lack of generality often limits their utility in real-world scenarios. In this paper, we address this challenge by leveraging the domain knowledge from several state-of-the-art expert systems. Particularly, we explore the idea of training a multitask student model using unlabeled data and the pseudo labels from multiple expert (i.e. teacher) systems. Evaluation is conducted using a variety of test datasets, and the results show that our proposed system compares favorably with the state-of-the-art systems and achieves the highest overall performance.