Entropy of Teager Energy in Wavelet-domain algorithm applied in note onset detection

Note segmentation is a crucial step in content-based musical signal analysis and processing. Considering the characters of multi-resolution of wavelet transform, anti-noise performance of TEO (Teager Energy Operator) and good statistical performance of information entropy, this paper combined this three features and proposed a novel note onset detection algorithm—Entropy of Teager Energy in Wavelet-domain (ETEW). Compared with the Adaptive Sub-band Spectrum Entropy (ASSE) which was a typical and effective note onset detection algorithm, the detection curve obtained from ETEW was smoother and the note boundaries were more obvious, which led to a 10% increase in the note segmentation accuracy. Especially for pieces played by percussion instruments, the results would be better. The experiment data set contained several groups played by 7 different kinds of instruments and had 2000 notes in total. Experiments indicated that the advantages of ETEW became much prominent when pieces were played by a variety of instruments or accompanied by background music. What's more, the anti-noise performance was improved in a great extent especially with lower SNR.