New approaches for rhythmic description of audio signals

The subject of this work is the study of how rhythmical description of music can be included in Music Information Retrieval works, but from a more musical point of view. The goal is to provide a suitable representation of rhythm and explore how useful it can be for many different applications. This new representation is based on the so called Rhythm Transform that transforms data from the time domain to a so called rhythm domain. This transformation has proved to give good results for automatic classification and similarity applications. The theoretical and computational issues of the Rhythm Transform are explained, and some results for specific applications are also shown.