Abstract:

This survey gives an overview of the subsymbolic systems and paradigms that are currently used in music analysis and in models of music perception. This includes a discussion of neural network and connectist systems for learning, as well as systems that deal with the preprocessing stage of musical signal processing (such as auditory systems). In the recent past, some advanced subsymbolic paradigms for music have been based on data driven long-term processing (learning), while other systems have been based on shema-based short-term processing. Subsymbolic systems for music, nowadays, are able to transform acoustical signals into auditory images and develop stable perception structures which in turn can be used to process music at higher semantic levels. The survey will provide a general framework for subsymbolic research in music and by means of examples, it will give an overview of the work that has been done in tone perception, rhythm analysis, and timbre recognition. Systems within hybrid environments (that combine both subsymbolic and symbolic approaches) are discussed, as well as the significance for computer interactive systems and cognitive modelling in general. Finally, the subsymbolic paradigm is discussed within the context of its epistemological relevance (the question to what extend subsymbolling modelling can/should be related to empirical and physiological data).