This thesis contributes to several facets of real-time computer music, using the construct of dynamic music representations as a platform towards the online retrieval and pro- cessing of information from sound and its (re)interpretation into visual and auditory display. We design and implement a real-time performance tracker using the audio-to-audio alignment of acoustic input with a pre-existing reference track, without any symbolic score information. We extend this system with a robust tempo modelling engine, that adaptively switches its source between the alignment path and a beat tracking module. Through several case studies, we contrast this system with the audio-to-score following paradigm, as instanced by the existing state-of-the-art Antescofo software. We describe our contributions to the user interface for the programming and execution of Antescofo scores, focussing on visual models for the coherent notation of dynamic processes. We also investigate avenues for automated generation of interactive music scores by way of formal grammars. Finally, we harness the potential of musical sound for data exploration and performance mediation, by implementing an EEG sonification system as part of an interactive theatre piece. Throughout, our objective is to increase the capacity of software-based systems for dynamic interaction, which enables more immediate and expressive modes of human- computer music performance.