Simon Dixon

Queen Mary University of London, UK (bio)

Automatic Transcription: An Enabling Technology for Music Analysis


The most common format for representing a musical performance is an audio recording, which provides a detailed low-level portrayal of the sound, but without human interpretation gives little insight into the concepts and structures discussed by human analysts. To bridge this semantic gap, audio content analysis algorithms are developed to extract features such as note onset time, pitch, duration, tempo, key and instrumentation from recordings. Automatic transcription is a technology whose goal is to combine these features in order to produce a score in some standard musical notation. In this talk I describe our recent advances in transcription research, and how they have been put to use in three music analysis scenarios: the characterisation of composer style from audio recordings, the analysis of temperament in harpsichord recordings, and the modelling of timbre in cello performance.