作者: Emilios Cambouropoulos , Simon Dixon
DOI:
关键词: MIDI 、 Computer science 、 Musical 、 Beat (music) 、 Musical form 、 Piano 、 Speech recognition 、 Beat tracking 、 Time signature 、 Beat detection
摘要: When a person taps foot in time with piece of music, they are performing beat tracking. Beat tracking is fundamental to the understanding musical structure, and therefore an essential ability for any system which purports exhibit intelligence or understanding. We present off-line multiple agent estimates locations beats MIDI performance data. This approach requires no prior information about input data, such as tempo signature; all required derived from For constant performances, previous systems have proved successful; however, these fail when there large variations tempo. examine role knowledge guiding process, show that equipped salience able track music even presence variations. Results presented corpus expressively performed classical piano (13 complete sonatas), containing full range tempos much variability within sections. With disabled, tracked 75% correctly; inclusion raises this figure over 90%.