Towards a style-specific basis for computational beat tracking

Outlined in this paper are a number of sources of evidence, from psychological, ethnomusicological and engineering grounds, to suggest that current approaches to computational beat tracking are incomplete. It is contended that the degree to which cultural knowledge, that is, the specifics of style and associated learnt representational schema, underlie the human faculty of beat tracking has been severely underestimated. Difficulties in building general beat tracking solutions, which can provide both period and phase locking across a large corpus of styles, are highlighted. It is probable that no universal beat tracking model exists which does not utilise a switching model to recognise style and context prior to application.