All current and future music has its roots in earlier music. Even the most avantgarde needs a prior knowledge of the foregoing in order to break it. This relationship between musician and its predecessors determines its musical characteristics. Thus, the relationship between previous, present and future musical elements determines the definition of musical influence. With this definition, we may tend to think that any music is an influence relationship. The main difficulty lies in determining what types of connections are influential and which not.
In this thesis we analyse the musical influence from a computational standpoint developing a basic algorithm that allows the finding of relationships between musical passages that can be considered as influential.
Unfortunately, the problem of influence has rarely been addressed with a computational perspective. For this reason, we present a conceptual framework that defines the needs of the problem and allows us to face analysis with guarantees. This conceptualization presents the different factors needed to discern if a musical relationship is influential or not as well as the various factors that condition relationships, the diverse musical dimensions involved and different levels of interaction.
Using this conceptual framework, we particularize the problem to a specific case: to find influences from different musical passages in Progressive Rock Music. To accomplish this, we used similarity between different music items as a measurable index of influence relationships. Our study compares a particular influenced segment against all the derived ones from a database of songs of four groups considered the most influential: King Crimson, Yes, Genesis and ELP.
For our approach we use several techniques developed in other fields of MIR such as how to obtain descriptors that characterize the different levels and computing music similarity distances. During the development of the thesis, we discuss and analyse many details as well as the problems encountered in the development of the algorithm derived from our particular approach. The results suggest that the automatic determination of influential relationships is a feasible task.