Automatic Chord Identification
Abstract. Context information is vital in musical listening. We present a chord transcription algorithm using dynamic Bayesian networks (DBN) that achieves context-awareness by simultaneously estimating beat position, key, bass note and the chord itself. The given data are audio waves of pop songs. Several pre-processing steps including automatic tuning and automatic beat-segmentation enhance the quality of the features used as observations in the model. Qualitative evaluation of chord transcriptions of selected songs demonstrates the method’s capability of reliably estimating more diverse chord types than our (and other authors’) previous methods have while maintaining low fragmentation.