A case study about the effort to classify music intervals by chroma and spectrum analysis

Abstract

Recognition of harmonic characteristics from polyphonic music, in particular intervals, can be very hard if the different instruments with their specific characteristics (overtones, formants, noisy components) are playing together at the same time. In our study we examined the impact of chroma features and spectrum on classification of single tone pitchs and music intervals played either by the same or different instruments. After the analysis of the audio recordings which produced the most errors we implemented two optimization approaches based on energy envelope and overtone distribution. Themethods were compared during the experiment study. The results show that especially the integration of instrument-specific knowledge can significantly improve the overall performance. © Springer International Publishing Switzerland 2013.