Musical note segmentation employing combined time and frequency analyses

In this paper we describe a musical stream note segmentation method that employs time-domain and frequency-domain analysis methods working in conjunction. The method has two demonstrated benefits: first, it leads to reliable results with very low probability of either missing or of falsely detecting notes, and second, it has temporal resolution on the order of 1 msec. We have applied the method to a variety of monophonic musical instrument recordings, including the clarinet, piano and violin, with results that vary from 95% to 100% accuracy.