In this paper we propose a real-time signal processing framework for musical audio that 1) aligns the audio with an existing music score or creates a musical score by automated music transcription algorithms; and 2) obtains the expressive feature descriptors of music performance by comparing the score with the audio. Real-time audio segmentation algorithms are implemented to identify the onset points of music notes in the incoming audio stream. The score related features and musical expressive features are extracted based on these segmentation results. In a realtime setting, these audio segmentation and feature extraction operations have to be accomplished at (or shortly after) the note onset points, when an incomplete length of audio signal is captured. To satisfy real-time processing requirements while maintaining feature accuracy, our proposed framework combines the processing stages of prediction, estimation, and updating in both audio segmentation and feature extraction algorithms in an integrated refinement process. The proposed framework is implemented in a MATLAB real-time signal processing framework.