Research On Entropy Of Teager Energy In Wavelet-domain Algorithm Applied In Note Onset Detection

Music note onset detection is a crucial step in content-based musical signal analysis and processing. Especially for query by humming (QBH) system, which opens a new prospect in musical retrieval field,if a kind of effective and simple note segmentation program is combined, it is that not only improve the retrieval efficiency to a considerable extent, but also be extremely convenient for the users, who could search directly by humming naturally avoiding some limitation. Therefore, it was in truth that music note onset detection was an indispensable and important process in content-based music information retrieval areas. Firstly, some famous and effective algorithms of musical signal process in recent years were presented in the paper. And the advantages and disadvantages of each algorithm were analyzed in very great detail. Then, aiming at the specific signal characters of music notes, several effective character parameters were combined together to form a kind of more high-efficiency and novel music onset detection method——Entropy of Teager Energy in Wavelet Domain Algorithm.It is considered that Teager energy operator not only retains the information of signals’ range, but also extracts the frequency characters. The above property justly could precisely describe the note signals in music streams, whose energy distribution is with variation in different frequency band. At the same time, the information entropy, with good statistical performance, could reflect the distribution of character parameters well. And it won’t be affected by the variation of a few sample signal point. Consequently, the two properties were combined to form a new character of entropy of Teager energy. After extracting the new character, the logarithmic function were used to equalize the parameters’ peaks and the dual thresholds method were used to extract the peaks, which was the whole process of note segmentation. Compared with the adaptive sub-band spectrum entropy algorithm, which is a typical and effective method in audio signal processing area, the entropy of Teager energy algorithm was much more quick and easy and the calculation complexity algorithm was reduced nearly 60%. At the same time, the obtained detection curve became more flat and smooth, and note boundaries became more obvious, which made the accuracy of the note segmentation increased by 10%. Especially for the music of percussion, the result would be better. However, when there is noise in musical signals which is the experiment data to be processed, especially the noise signal in high frequency band, the detection result of entropy of Teager energy algorithm degrades much. For this problem, wavelet transform were used before extracting the character of entropy of Teager energy. In that process, the high frequency noise was removed, and the analysis and processes were only applied to low frequency signal. What’s more, the experimental results indicate that anti-noise performances were improved much. Thus, a novel and high-efficiency method——Entropy of Teager Energy in Wavelet Domain Algorithm, was proposed at last. In this paper, the experimental data contains different kinds of musical files, which includes 4 kinds,7groups, and more than 2000 notes in total. It was found that the advantages of the new algorithm proposed in this paper were more obvious when the tested music was played by a variety of instruments or accompanied by background music. Indeed, the entropy of Teager energy in wavelet domain algorithm was a kind of more excellent note onset detection method.