Automatic Location And Measurement Of Score-based Gestures In Audio Recordings

This paper reports on our first experiments in using the feature extraction tools of the MPEG-7 international standard for multimedia content description on a novel problem, the automatic identification and analysis of scorebased performance features in audio recordings of music. Our test material consists of recordings of two pieces of 17thand 18th-century lute music in which our aim is to recognise and isolate performance features such as trills and chord-spreadings. Using the audio tools from the MPEG-7 standard facilitates interoperability and allows us to share both score and audio metadata. As well as using low-level audio MIR techniques within this MPEG7 context, the work has potential importance as an ’ornamentation filter’ for MIR systems. It may also form a useful component in methods for instrumental performer identification.