Classifying Musical Performance by Statistical Analysis of Audio Cues

Abstract

In this paper an attempt is presented to classify concert music performances played by different performers (and hence characterized by different expressive styles). The classification is obtained from statistical analysis of a set of simple audio cues, extracted in real time by a set of tools implemented in the EyesWeb open platform. The cue extraction process is carried out by looking at a “note” and a “phrase” profile (obtained by squaring and low pass filtering the audio signal). These are exploited to obtain, for each note, parameters regarding tempo, articulation and dynamics. Statistical analysis is then performed on the extracted data in order to establish whether more or less remarkable differences among the original performances can be traced. Once this is confirmed, a Hidden Markov Model is proposed, developed in MatLab and able to recognize the players’ styles on the basis of audio cues.