Content-based description of music has become a significant research topic, since technological advances let the amount of digitally available music explode. One of the most interesting and challenging problems, among the wide range of disciplines in this field, is that of music transcription. The term transcription refers to the task of estimating the temporal locations of sound events and recognising the instruments which have been used to produce them. Research in this discipline primarily focused on the extraction of melodic and tonal information, until more recently the extraction of rhythmic structures received the same degree of attention. As percussive instruments form the rhythmic backbone of a musical piece, their transcription is a key component in representing and understanding music. This thesis explores state of the art signal processing techniques that have found application in percussion transcription and describes a template-matching-based transcription system in more detail, which has been implemented and evaluated in the course of this thesis.

Top